Immunogroup analysis method and system for cancer diagnosis and treatment prognosis

Abstract

The present invention provides a method of providing a cancer assessment for a subject, comprising: (i) measuring values of one or more cancer traits of at least one tumor-associated immune cell (TAIC) or TAIC-derived biomarker in a liquid sample obtained from the subject to obtain a test profile; (ii) comparing the test map to one or more reference maps; (iii) wherein the similarity or difference of the test profile to one or more reference profiles is indicative of one or more cancer characteristics; and (iv) providing a cancer assessment of the subject based on the one or more cancer characteristics.

Description

[0001] Related Applications

[0002] This application claims priority to U.S. Provisional Application No. 63 / 330,123, filed on April 12, 2022, the entire contents of which are hereby incorporated by reference. Technical Field

[0003] The present disclosure relates generally to cancer detection and prognosis, and more particularly, to methods and devices for early cancer detection and prognosis using tumor-associated immune cells as biomarkers. Background Art

[0004] The global cancer burden continues to increase, with an estimated 18 million cancer cases and 9.6 million cancer deaths in 2018. Most cancers are curable when diagnosed while they are still confined to the organ of origin. Therefore, rapid screening for cancer incidence is key to curing cancer.

[0005] Unfortunately, many types of cancer are diagnosed only at advanced stages because diagnostic interventions are often associated with symptoms. Therefore, cancer screening is needed as an early cancer intervention strategy to detect disease markers before obvious symptoms appear.

[0006] Early detection and monitoring of recurrence have been shown to have the potential to reduce cancer mortality. So far, non-invasive cancer diagnostic studies have relied on endogenous biomarkers such as circulating cell-free DNA, circulating tumor cells, and exosomes. The sensitivity of cancer diagnosis using endogenous biomarkers is limited due to rapid biomarker clearance, low concentrations in the circulation, and high background signals, as these biomarkers are also present in healthy individuals. Therefore, attention needs to be paid to the immune system, which is the first line of defense against abnormal changes in homeostasis such as cancer.

[0007] Studies have shown that immune responses are associated with tumor growth by changing the tumor microenvironment and are associated with cancer progression and survival (Kresovich, JK et al., 2020). The immune profile obtained after cancer diagnosis may be associated with prognosis, but whether the immune profile can be used as a diagnostic marker remains underexplored. SUMMARY OF THE INVENTION

[0009] According to a broad aspect, at least one exemplary embodiment described in accordance with the teachings herein provides a method of determining one or more cancer characteristics of a subject, the method comprising:

[0010] measuring the value of one or more cancer traits of at least one tumor-associated immune cell (TAIC) or TAIC-derived biomarker in a fluid sample obtained from the subject to obtain a test profile;

[0011] comparing the test profile to one or more reference profiles;

[0012] Wherein the similarity or difference between the test profile and the one or more reference profiles indicates one or more cancer characteristics.

[0013] In one embodiment, the method further comprises providing a cancer assessment for the subject based on the one or more cancer characteristics.

[0014] In one embodiment, the cancer characteristic is: the presence or absence of cancer, cancer type, cancer stage, cancer grade, primary cancer, metastatic cancer, cancer metastatic potential, molecular phenotype of cancer, benign cancer, malignant cancer, tissue of origin of cancer, location of cancer metastasis, or a combination thereof.

[0015] In another embodiment, the at least one tumor-associated immune cell comprises: a T cell, a NK cell, a macrophage, a dendritic cell, a B cell, or a neutrophil, or a combination thereof.

[0016] In another embodiment, the one or more TAIC-derived biomarkers comprise a metabolic change, activation state, or surface phenotype of at least one tumor-associated immune cell.

[0017] In another embodiment, the one or more TAIC-derived biomarkers comprise one or more extracellular vesicles from at least one tumor-associated immune cell.

[0018] In another embodiment, the one or more TAIC-derived biomarkers comprise cell-free nucleic acid of at least one tumor-associated immune cell.

[0019] In another embodiment, the cell-free nucleic acid is: cell-free DNA, miRNA, IncRNA, mRNA, or histone-associated DNA.

[0020] In another embodiment, the cell-free DNA is molecularly modified by one or more of methylation, oxidation, phosphorylation, and acetylation.

[0021] In another embodiment, one or more TAIC-derived biomarkers comprise the structural and molecular composition of cell-free DNA.

[0022] In another embodiment, the one or more TAIC-derived biomarkers comprise the DNA methylation status of at least one tumor-associated immune cell.

[0023] In another embodiment, one or more cancer traits are measured by Raman spectroscopy.

[0024] In another embodiment, one or more cancer traits are measured by DNA next generation sequencing, PCR, repertoire sequencing, immunosequencing, single cell sequencing, single cell RNA sequencing, unique molecular identifiers (UMIs), light microscopy, optionally confocal microscopy, immunohistochemistry, super-resolution microscopy, flow cytometry, or mass cytometry.

[0025] In another embodiment, the reference profile is derived from one or more reference TAICs.

[0026] In another embodiment, the reference TAIC is: a T cell, a NK cell, a macrophage, a dendritic cell, a B cell, or a neutrophil.

[0027] In another embodiment, the reference TAIC is: a cancer cell-associated T cell, a cancer stem cell-associated T cell, or a tumor-associated stem-like T cell.

[0028] In another embodiment, the reference TAIC is: a cancer cell-associated NK cell, a cancer stem cell-associated NK cell, or a tumor-associated stem cell-like NK cell.

[0029] In another embodiment, the reference TAIC is: a cancer cell-associated macrophage, a cancer stem cell-associated macrophage, or a tumor-associated stem cell-like macrophage.

[0030] In another embodiment, the reference TAIC is: a cancer cell-associated dendritic cell, a cancer stem cell-associated dendritic cell, or a tumor-associated stem cell-like dendritic cell.

[0031] In another embodiment, the reference TAIC is: a cancer cell-associated B cell, a cancer stem cell-associated B cell, or a tumor-associated stem-like B cell.

[0032] In another embodiment, the reference TAIC is: cancer cell-associated neutrophils, cancer stem cell-associated neutrophils, or tumor-associated stem-like neutrophils.

[0033] In another embodiment, the fluid is: plasma, serum, urine, feces, mucus, or cerebrospinal fluid.

[0034] In another embodiment, providing a cancer assessment includes providing a cancer type, a cancer location, a cancer stage, a cancer metastasis potential, or a cancer treatment efficacy.

[0035] In another embodiment, providing a cancer assessment includes providing a prognosis for a patient, diagnosing early cancer, determining whether a tumor is benign or malignant, determining whether a tumor is primary or metastatic, determining whether a primary tumor has metastatic potential, determining the progression of cancer, determining lymph node metastasis of cancer, determining clinical metastasis of cancer, predicting patient survival, providing a prognosis for a patient, providing early diagnosis of cancer (including providing early diagnosis of difficult-to-detect cancers), determining the presence of aggressive brain cancer, providing the location of a tumor, monitoring cancer recurrence during or after treatment, and / or determining the presence of minimal residual disease.

[0036] In another embodiment, the method further comprises determining whether the tumor has malignant potential when the tumor is benign.

[0037] In another embodiment, the cancer is brain cancer, optionally a glioblastoma, astrocytoma, or oligodendroglioma.

[0038] In another embodiment, the cancer is:

[0039] Ovarian cancer, brain cancer, bladder cancer, breast cancer, colon cancer, esophageal cancer, stomach cancer, liver cancer, intestinal cancer, lung cancer, head and neck cancer, rectal cancer, prostate cancer, pancreatic cancer, thyroid cancer, cervical cancer, melanoma, nasopharyngeal cancer, testicular cancer, or uterine cancer.

[0040] According to another broad aspect, at least one exemplary embodiment described according to the teachings of this document provides a method for determining one or more cancer characteristics of a subject, the method comprising: separating a volume of liquid from a liquid sample of the subject, the volume of liquid comprising at least one tumor-associated immune cell (TAIC) or a TAIC-derived biomarker; adding at least a portion of the volume of liquid to a nanosensor, the nanosensor comprising nanoparticles configured to capture at least one TAIC or TAIC-derived biomarker and amplify a signal emitted by at least one TAIC or TAIC-derived biomarker during Raman spectroscopy; performing Raman spectroscopy on the volume of liquid on the nanosensor to produce a sample Raman spectrum, the sample Raman spectrum having an amplified signal, the amplified signal indicating the presence of at least one TAIC or TAIC-derived biomarker on the nanosensor; and processing the sample Raman spectrum using data from a template Raman spectrum having cancer characteristics to detect whether the sample contains one or more cancer characteristics.

[0041] In one embodiment, the method further comprises providing a cancer assessment for the subject based on the detected one or more cancer characteristics.

[0042] In one embodiment, one or more cancer signatures of a sample are detected based on determining which correlation values ​​are greater than a correlation threshold, wherein the correlation values ​​are obtained by correlating an amplified signal of a Raman spectrum of the sample with a template Raman spectrum from a known cancer sample having the cancer signature.

[0043] In another embodiment, one or more cancer signatures of a sample are detected by: performing feature extraction on Raman sample spectral data to extract feature values; performing classification by applying the feature values ​​to at least one set of classification models determined for at least one TAIC or TAIC-derived biomarker to detect one or more cancer signatures; and providing a cancer assessment by combining each detected cancer signature, wherein the classification model is determined using a template Raman spectrum from a known cancer sample.

[0044] In another embodiment, feature extraction is performed using principal component analysis, multivariate curve resolution analysis, or a combination thereof.

[0045] In another embodiment, the classification model comprises: partial least squares discriminant analysis (PLSDA), support vector machine discriminant analysis (SVMDA), artificial neural network analysis (ANN) TSNE and random forest classification.

[0046] In another embodiment, the sample Raman spectrum includes a second amplified signal, which indicates the presence of a second biomarker on the nanosensor, and the method further includes: further data processing the sample Raman spectrum to compare the second amplified signal with the second template Raman spectrum to determine a correlation between the sample Raman spectrum and the second template Raman spectrum, the template Raman spectrum having a known cancer feature; and providing a diagnosis of cancer in the patient based on the correlation between the sample Raman spectrum and the template Raman spectrum and the second correlation between the sample Raman spectrum and the second template Raman spectrum.

[0047] According to another broad aspect, at least one exemplary embodiment described in accordance with the teachings herein provides a method for generating reference tumor-associated immune cells (TAICs) for cancer assessment, the method comprising co-culturing one or more naive immune cells with a tumor and / or cancer cell line until the one or more reference tumor-associated immune cells (TAICs) express one or more cancer traits.

[0048] In one embodiment, the one or more primary immune cells comprise: T cells, NK cells, macrophages, dendritic cells, B cells, neutrophils, or a combination thereof.

[0049] In another embodiment, the cancer assessment includes providing the type of cancer, the location of the cancer, the stage of the cancer, the grade of the cancer, the metastatic potential of the cancer, or the efficacy of a treatment of the cancer.

[0050] According to another broad aspect, at least one example embodiment described in accordance with the teachings herein provides a computing device for providing a cancer assessment for a subject, or for determining one or more cancer characteristics of a subject.

[0051] In at least one embodiment, the processing unit is further configured to perform any steps of the method described according to the teachings herein.

[0052] These and other features and advantages of the present application will become apparent from the detailed description below in conjunction with the accompanying drawings. However, it should be understood that although the detailed description and specific examples show the preferred embodiments of the present application, they are only given by way of example, because those skilled in the art can clearly understand various changes and modifications within the spirit and scope of the present application from the detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] To better understand the various embodiments described herein, and to more clearly show how these various embodiments may be implemented, reference will now be made to the accompanying drawings showing at least one example embodiment, by way of example. The drawings are not intended to limit the scope of the teachings described herein.

[0054] Figure 1 Shown are a flow chart (a) illustrating a workflow for discovering cancer-specific signatures in immune cells of patients diagnosed with cancer and (b) a potential workflow for cancer diagnosis in a clinical setting using the published methods.

[0055] Figure 2 Evaluation of tumor-associated immune cell viability is shown for comprehensive cancer assessment. a) Similarity analysis of individual immune cell signatures in buffy coat profiles of validated cancer patients, b) Principal component analysis showing overlap of tumor-associated immune signatures between genders (top). Bar graphs of individual macromolecule peak positions do not show statistical significance (bottom), c) Principal component analysis showing overlap of tumor-associated immune signatures between age groups under 45 years and over 75 years (top). Bar graphs of individual macromolecule peak positions do not show statistical significance (bottom).

[0056] Figure 3Figure 2 shows the discovery of unique T cell signatures specific to cancer, a) t-distributed stochastic neighbor embedding (tSNE) plot showing T cell phenotypic differences between tumor-associated T cells and non-tumor T cells, b) violin plot showing that T cells in early-stage cancers exhibit significant phenotypic differences compared to non-cancers. c) tSNE plot showing that T cells in early-stage cancers exhibit significant phenotypic differences compared to late-stage cancers, d) heat map (bottom) and dendrogram (top) showing that significant differences are caused by peaks represented in PC1.

[0057] Figure 4 The application of unique tumor-associated T cell signatures in cancer diagnosis and early cancer diagnosis is shown. a) Scatter plot showing the probability of diagnosing a sample as cancer based on the tumor-associated T cell signature. b) Calibration curve showing the performance of a machine learning model for diagnosing cancer using the tumor-associated T cell signature, with sensitivity and specificity both 100%. c) Principal components contributing to accurate cancer diagnosis. d) ROC curve for cancer prediction, e) Scatter plot showing the probability of diagnosing a sample as early cancer based on the tumor-associated T cell signature, f) Calibration curve showing the performance of a machine learning model for diagnosing early cancer using the tumor-associated T cell signature, with sensitivity and specificity both 100%. g) Principal components contributing to accurate early cancer diagnosis, h) ROC curve for early cancer prediction, with an AUC of 1.00.

[0058] Figure 5 Figure 1 shows the application of unique tumor-associated T cell signatures in cancer staging and metastasis diagnosis. a) Scatter plot showing the probability of cancer staging a sample based on the tumor-associated T cell signature. b) Calibration curve showing the performance of the machine learning model for cancer staging using the tumor-associated T cell signature, with a sensitivity of 98.4% and a specificity of 100%. c) Principal components contributing to accurate cancer staging. d) ROC curve for cancer staging showing an AUC of 1.00. e) Scatter plot showing the probability of a sample being metastatic based on the tumor-associated T cell signature. f) Calibration curve showing the performance of the machine learning model for detecting metastasis using the tumor-associated T cell signature, with a sensitivity of 96.2% and a specificity of 100%. g) Principal components contributing to accurate detection of metastasis, and h) ROC curve for predicting cancer metastasis, showing an AUC of 1.00.

[0059] Figure 6A schematic diagram of the method for cancer diagnosis using the circulating natural killer cell profiling (CNKP) and tumor immune probe platform is shown. The SERS training data came from in vitro generated cancer cell-associated NK cells and cancer stem cell (CSC)-associated NK cells (left, top panel) as well as healthy NK cells (left, bottom panel). Middle: Three NK cell phenotypes form the basis for cancer diagnosis using a simple machine learning paradigm. Once the principal components that distinguish cancer-associated NK cells from healthy NK cells are identified from the training data, the model can be used to make cancer predictions using SERS data obtained from patient blood samples (right panel).

[0060] Figure 7 Figure 1 shows the discovery of unique NK cell signatures specific to cancer, a) tSNE plot showing NK cell phenotypic differences between tumor-associated NK cells and non-tumor NK cells, b) NK cells in early cancers show significant phenotypic differences compared to non-cancers, c) NK cells in early cancers show significant phenotypic differences compared to late-stage cancers, d) dendrogram (left) and heat map (right) showing that the significant differences are caused by peaks represented in PC1.

[0061] Figure 8 The application of unique tumor-associated NK cell signatures in cancer diagnosis and early cancer diagnosis is shown. a) Scatter plot showing the probability of diagnosing a sample as cancer based on the tumor-associated NK cell signature. b) Calibration curve showing the performance of the machine learning model for diagnosing cancer using the tumor-associated NK cell signature, with a sensitivity and specificity of 100%. c) Principal components contributing to accurate diagnosis of cancer, d) ROC curve for cancer prediction, with an AUC of 1.0. e) Scatter plot showing the probability of diagnosing a sample as early cancer based on the tumor-associated NK cell signature, f) Calibration curve showing the performance of the machine learning model for diagnosing early cancer using the tumor-associated NK cell signature, with a sensitivity and specificity of 100%. g) Principal components contributing to accurate early cancer diagnosis, h) ROC curve for early cancer prediction, with an AUC of 1.00.

[0062] Fig. 9Figure 1 shows the application of unique tumor-associated NK cell signatures in cancer staging and cancer metastasis diagnosis. a) Scatter plot showing the probability of cancer staging a sample based on the tumor-associated NK cell signature, b) Calibration curve showing the performance of the machine learning model for cancer staging using the tumor-associated NK cell signature, with a sensitivity and specificity of 100%. c) Principal components contributing to accurate cancer staging, d) ROC curve for cancer staging, with an AUC of 1.00. e) Scatter plot showing the probability of a sample being metastatic based on the tumor-associated NK cell signature, f) Calibration curve showing the performance of the machine learning model for detecting metastasis using the tumor-associated NK cell signature, with a sensitivity of 99.6% and a specificity of 100%. g) Principal components contributing to accurate detection of metastasis, h) ROC curve for predicting cancer metastasis, with an AUC of 1.00.

[0063] Fig.10 The unique macrophage signature found to be specific to cancer is shown, a) tSNE plot showing the macrophage phenotypic differences between tumor-associated macrophages and non-tumor macrophages, b) macrophages in early cancers show significant phenotypic differences compared to non-cancers. c) macrophages in early cancers show significant phenotypic differences compared to late-stage cancers, d) dendrogram (left) and heat map (right) showing that the significant differences are caused by the peaks represented in PC1.

[0064] Fig.11 The application of unique tumor-associated macrophage features in cancer diagnosis and early cancer diagnosis is shown. a) Scatter plot showing the probability of diagnosing a sample as cancer based on the tumor-associated macrophage feature. b) Calibration curve showing the performance of the machine learning model for diagnosing cancer using the tumor-associated macrophage feature, with a sensitivity and specificity of 100%. c) Principal components contributing to the accurate diagnosis of cancer. d) ROC curve for cancer prediction, with an AUC of 1.00. e) Scatter plot showing the probability of diagnosing a sample as early cancer based on the tumor-associated macrophage feature. f) Calibration curve showing the performance of the machine learning model for diagnosing early cancer using the tumor-associated macrophage feature, with a sensitivity of 98.5% and a specificity of 100%. g) Principal components contributing to the accurate diagnosis of early cancer, h) ROC curve for early cancer prediction, with an AUC of 1.00.

[0065] Fig.12Figure 1 shows the application of unique tumor-associated macrophage signatures in cancer staging and cancer metastasis diagnosis. a) Scatter plot showing the probability of staging a sample into cancer based on the tumor-associated macrophage signature, b) Calibration curve showing the performance of the machine learning model for cancer staging using the tumor-associated macrophage signature, with specificity and sensitivity both 100%. c) Principal components contributing to accurate cancer staging, d) ROC curve for cancer staging, with an AUC of 1.00. e) Scatter plot showing the probability of metastasis based on the tumor-associated macrophage signature, f) Calibration curve showing the performance of the machine learning model for detecting metastasis using the tumor-associated macrophage signature, with specificity and sensitivity both 100%. g) Principal components contributing to accurate detection of metastasis, h) ROC curve for predicting cancer metastasis, with an AUC of 1.00.

[0066] Fig.13 Figure 1 shows the discovery of unique DC signatures specific to cancer. a) tSNE plot showing differences in DC phenotypes between tumor-associated DCs and non-tumor DCs. b) DCs in early-stage cancers show significant phenotypic differences compared to non-cancers. c) DCs in early-stage cancers show significant phenotypic differences compared to late-stage cancers. d) Dendrogram (left) and heatmap (right) showing that significant differences are caused by peaks represented in PC1.

[0067] Fig.14 The application of unique tumor-associated dendritic cell features in cancer diagnosis and early cancer diagnosis is shown. a) Scatter plot showing the probability of diagnosing a sample as cancer based on the tumor-associated dendritic cell feature. b) Calibration curve showing the performance of the machine learning model for diagnosing cancer using the tumor-associated dendritic cell feature, with a specificity and sensitivity of 100%. c) Principal components contributing to accurate cancer diagnosis. d) ROC curve for cancer prediction. e) Scatter plot showing the probability of diagnosing a sample as early cancer based on the tumor-associated dendritic cell feature. f) Calibration curve showing the performance of the machine learning model for diagnosing early cancer using the tumor-associated dendritic cell feature, with a sensitivity of 97.1% and a specificity of 96.7%. g) Principal components contributing to accurate early cancer diagnosis, h) ROC curve for early cancer prediction, with an AUC of 0.998.

[0068] Fig.15 Figure 2 shows the discovery of unique B cell signatures specific to cancer. a) tSNE plot showing differences in B cell phenotypes between tumor-associated B cells and non-tumor B cells, b) B cells in early-stage cancers show clear phenotypic differences compared to non-cancers, c) B cells in early-stage cancers show clear phenotypic differences compared to late-stage cancers, d) dendrogram (left) and heatmap (right) showing that the clear differences are caused by peaks represented in PC1.

[0069] Fig.16 The application of unique tumor-associated B cell signatures in cancer diagnosis and early cancer diagnosis is shown. a) Scatter plot showing the probability of diagnosing a sample as cancer based on the tumor-associated B cell signature, b) Calibration curve showing the performance of the machine learning model for diagnosing cancer using the tumor-associated B cell signature, with sensitivity and specificity both 100%. c) Principal components contributing to accurate cancer diagnosis, d) ROC curve for cancer prediction, e) Scatter plot showing the probability of diagnosing a sample as early cancer based on the tumor-associated B cell signature, f) Calibration curve showing the performance of the machine learning model for diagnosing early cancer using the tumor-associated B cell signature, with sensitivity and specificity both 100%. g) Principal components contributing to accurate early cancer diagnosis, h) ROC curve for early cancer prediction, with AUC of 1.00.

[0070] Fig.17 The unique characteristics of tumor-associated immune cells are shown to be used in the early diagnosis of multiple cancers and the determination of tumor tissue sources in breast cancer, lung cancer, colorectal cancer, kidney cancer, uterine cancer, testicular cancer, brain cancer (including astrocytoma, oligodendroglioma, glioblastoma), ovarian cancer, and head and neck tissue cancer. a) Tumor-associated T cells are used for the early diagnosis of multiple cancers, b) Tumor-associated macrophages are used for the early diagnosis of multiple cancers, c) Tumor-associated dendritic cells are used for the early diagnosis of multiple cancers, d) Tumor-associated NK cells are used for the early diagnosis of multiple cancers, e) The accuracy of using tumor-associated immune cells for the early diagnosis of multiple cancers is close to 100%. f) The specificity of using tumor-associated immune cells for the early diagnosis of multiple cancers is close to 100%. g) The sensitivity of using tumor-associated immune cells for the early diagnosis of multiple cancers is close to 100%.

[0071] Fig.18 Figure 1 shows the application of the unique signature of tumor-associated immune cells in early tumor diagnosis based on size. a) Scatter plot showing the probability of diagnosing a sample as cancer based on the tumor-associated immune cell signature. b) Calibration curve showing the performance of the machine learning model for diagnosing cancer using the tumor-associated immune cell signature, with a sensitivity of 100% and a specificity of 98.3%. c) Principal components contributing to the accurate diagnosis of cancer. d) ROC curve for cancer prediction. e) Multiple cancer diagnosis correctly diagnoses 5 types of cancer at an early stage based on tumor size of 0.5-2.4 cm.

[0072] Fig.19 The diagnosis of multiple cancers using the combined signature of tumor-associated immune cells (including T cells, NK cells, B cells, dendritic cells and macrophages) was shown. Specifically, the specificity, sensitivity and accuracy of the model were all 100%.

[0073] Fig. 20 A schematic diagram of neutrophil interactions in the tumor microenvironment (TME) of ovarian cancer (OC) is shown. The top figure depicts a schematic diagram of healthy and cancerous ovaries. Neutrophils were observed to interact with ovarian cancer cells and ovarian cancer stem cells in the OC TME. After the interaction, the neutrophils re-entered the circulation with an altered phenotype. The middle figure demonstrates the use of Raman spectroscopy and machine learning to diagnose OC. Non-interacting neutrophils (control) and tumor-associated neutrophils (TAN) were used as training data to train the ANN algorithm. Blood from OC patients was used as test data. The bottom figure shows the Raman spectra of early and late neutrophils. Using these phenotypes as training data and blood from OC patients as test data, OC can be clearly distinguished as early and late stages.

[0074] Fig.21 Tumor-associated immune cells in the TME of ovarian cancer are shown. A shows a heat map demonstrating the immune landscape in the tumor microenvironment of ovarian cancer (n=430). In the ovarian tumor microenvironment, signal transduction between immune cells and ovarian cancer cells can affect the immune response and regulate cancer progression. B shows the proportion of neutrophils in the tumor microenvironment of ovarian cancer. Neutrophils are the main inflammatory cells among immune cells. C shows a schematic diagram of neutrophils on the nanosensor. Immune cells were isolated from peripheral blood, 5 μl of each immune cell was dropped on the sensor, and Raman spectra were taken. D shows the Raman spectra of neutrophils, dendritic cells (D cells), monocytes, NK cells, and T cells. E shows multiple clusters corresponding to different immune cells. PCA can visualize the changes in Raman signals between cells. F shows hierarchical cluster analysis (HCT), confirming the different characteristics of immune cells.

[0075] Fig. 22 Shown are unique neutrophil signatures found to be specific to the presence of ovarian cancer. A shows a schematic diagram of non-interacting (i) and interacting (ii) neutrophils on the nanosensor. B shows the Raman spectra of non-interacting (i) and tumor-interacting (ii) neutrophils, showing the contributions of lipids, proteins, and nucleic acids identified by the grey boxes. C shows a schematic diagram of neutrophils acting through oxidative and non-oxidative mechanisms. D shows a biplot of non-interacting and ovarian cancer-interacting neutrophils. E shows the loading plot obtained from PCA, highlighting the significant peaks that contributed to the classification. F shows a hierarchical cluster analysis confirming the distinct clusters formed by non-interacting and tumor-interacting neutrophils.

[0076] Fig.23Figure 2 shows the changes in neutrophils interacting with ovarian cancer cells and with ovarian cancer stem cells over time. A shows a schematic diagram of the heterogeneous ovarian cancer tumor microenvironment with ovarian cancer, ovarian cancer stem cells, and neutrophils. B and C show the Raman spectra of ovarian cancer-associated (B) and ovarian cancer stem cell-associated (C) neutrophils at 0, 6, 12, and 24 hours, respectively. D and E show the changes in the Raman signals of neutrophils at different time points when interacting with ovarian cancer cells (D) and ovarian cancer stem cells (E). From left to right are the Raman signals of carbohydrates, lipids, and proteins, respectively. F shows that PCA highlights the different clustering of ovarian cancer and cancer stem cells at 0h (i), 6h (ii), 12h (iii), and 24h (iv).

[0077] Fig.24 Raman signatures of programmed death ligand 1 (PD-L1) and evaluation of PD-L1 peaks in early and late neutrophils are shown. A shows the Raman signature of the immune checkpoint protein PD-L1, highlighting the significant peaks. B shows simulations of early and late neutrophils using TGFβ (left, top) and all-trans retinoic acid (left, bottom), and comparison of the Raman signatures of early and late neutrophils with the significant peaks of the PDL-1 immune checkpoint protein (right). C shows the Raman signatures of early (left) and late (right) neutrophils. D shows a Student's t-test indicating the peaks at 546 cm -1 、672cm -1 、853cm -1 、1004cm -1 、1452cm -1 、1589cm -1 、1672cm -1 (from left to right), there is a significant difference between early and late neutrophils (p<0.0001=****).

[0078] Fig.25 The ability of the model to predict healthy vs. ovarian cancer status using patient blood is shown. A) Schematic diagram of OC detection predicted using machine learning-based diagnostic model and scatter plot. B) Sensitivity and specificity curves for OC detection. C) Calibration curve vs. predicted probability. D) Classification accuracy curve for OC detection. E) Neural network scatter plot indicating clear clustering between healthy and cancer. F) Sensitivity and specificity curves using a blood-based training dataset. G) Neural network scatter plot for OC detection, where using OC patient data as test and training (80:20) allows the neural network to achieve 100% sensitivity and 100% specificity.

[0079] Fig.26The ability of the unique neutrophil signature to classify early-stage cancer and late-stage cancer is shown. A) Schematic diagram of early OC detection using a machine learning-based diagnostic model. B) Sensitivity and specificity curves for early ovarian cancer detection. C) Calibration curve versus predicted probability. D) Classification accuracy curve of the early diagnosis model. The model has an area under the curve of 1.000, a classification accuracy of 0.900, a sensitivity of 0.900, a specificity of 1.000, a precision of 0.917, and a recall of 0.900. E) Scatter plot of early and late-stage ovarian cancer, indicating clear clustering.

[0080] Fig. 27 In vitro validation of T cell DNA as a biomarker for early detection of breast cancer is shown. (a) Schematic diagram of the experimental design (left) and SERS spectra of DNA isolated from healthy T cells and T cells interacting with breast cancer stem cells (right), showing clear distinction of peaks. (b) PCA scatter plot showing clear differences between the SERS signatures of DNA from healthy T cells and T cells associated with breast cancer stem cells (BCSC). (c) PC loading plot of principal components PC1, PC2, PC3, and PC4 from the PCA analysis, showing the contributing peaks to the differences observed between healthy and BCSC-associated T cells.

[0081] Fig.28 Figure 2 shows the similarity analysis between breast cancer stem cell-associated T cell DNA, breast cancer tumor-associated T cell DNA, and cell-free DNA present in liquid biopsy samples. (a) Schematic diagram of the experimental design, (b) tSNE analysis showing the similarity between liquid biopsy and breast cancer CSC-associated T cell DNA. (c) tSNE analysis showing the similarity between serum and tumor DNA. (d) tSNE analysis showing the similarity between breast cancer stem cell-associated T cell DNA and tumor DNA. (e) tSNE analysis, (f) dendrogram (left) and heat map (right), and (g) hierarchical clustering, all showing the similarity between breast cancer stem cell-associated T cell DNA, cell-free DNA in serum, and tumor DNA, confirming the ability to apply immune cell-associated cell-free DNA to cancer diagnosis.

[0082] Fig.29Validation of nanosensors for early diagnosis of breast cancer using organoid cultures of different sizes is shown. (a) Schematic showing organoid cultures of different sizes mimicking different grades of tumors, cultured by seeding (i) 10 million cells (ii) 1 million cells (iii) 10,000 cells (iv) 1000 cells. (b) SERS signature of DNA isolated from T cells interacting with breast cancer tumor organoids of different sizes. (c) Optical microscopy images showing the formation of breast cancer tumor organoids in AggreWell plates, (d) tSNE analysis confirms discrete clustering, (e)-(i) Comparison of the intensity of SERS spectra of DNA from T cells interacting with breast cancer tumor organoids at the following peak positions: (e) 1003cm -1 , (f)1142cm -1 , (g) 1208cm -1 , (h)1342cm -1 , and (i) 1492 cm -1 .

[0083] Fig.30 The applicability of cell-free DNA from T cells in cancer diagnosis is shown. (a) and (b) are schematic diagrams of the experimental design for training data (a) and test data (b). (c) A scatter plot showing the probability of diagnosing a sample as early-stage cancer (compared to healthy) based on T cell-related cell-free DNA features. (d) A calibration curve showing the performance of a machine learning model for diagnosing early-stage cancer using T cell-related cell-free DNA features, with a sensitivity of 93.6% and a specificity of 100%. (e) Principal components contributing to the accurate diagnosis of early-stage cancer, (f) ROC curve for early-stage cancer prediction, (g) Schematic diagram of a cancer staging experiment using T cell-related cell-free DNA features, (h) A scatter plot showing the probability of diagnosing a sample as late-stage cancer (compared to early-stage) based on T cell-related cell-free DNA features, (i) A calibration curve showing the performance of a machine learning model for diagnosing late-stage cancer using T cell-related cell-free DNA features, with a specificity of 86.8% and a sensitivity of 99%. (j) ROC curve for advanced cancer prediction, (k) Schematic diagram of cancer cell differentiation diagnosis using T cell-related cell-free DNA features. (i) Scatter plot showing the probability of a sample being poorly differentiated, moderately differentiated, or well differentiated based on T cell-related cell-free DNA features, (m) Calibration curve showing the performance of a machine learning model for diagnosing cancer differentiation status using T cell-related cell-free DNA features, with a specificity of 91.3.8% and a sensitivity of 98.2%. (n) ROC curve for cancer differentiation status prediction.

[0084] Fig.31A unique immunogenomic signature of breast cancer is shown. a) Schematic diagram of the experimental configuration for determining the immunogenomic signature of breast cancer. b) SERS spectral signatures showing characteristic peaks associated with DNA. c) Scatter plot showing the discrimination of immunogenomic signatures associated with breast cancer compared to non-cancer using cluster analysis. d) Loading plot showing the principal components of maximum variance (i.e., PC1 and PC2). e) Violin plot showing the distribution of F1 scores calculated using discriminant analysis. f) Heat map (bottom) and dendrogram (top) showing the correlation and heterogeneity of immunogenomic signatures of breast cancer and non-cancer.

[0085] Fig.32 Figure 2. Immunogenomic signatures showing differences between breast cancer subtypes associated with HER2 expression. a) Schematic diagram of the experimental setup to determine the immunogenomic signatures of breast cancer subtypes based on HER2 expression. b) Representative SERS spectra showing the immunogenomic signatures of HER2-positive and HER2-negative breast cancers. c) Boxplot comparing the normalized intensities of DNA-related signature peaks in HER2-positive and HER2-negative breast cancers. d) Correlation matrix showing the linear correlation between the normalized peak intensities associated with HER2-positive and HER2-negative breast cancers. e) Multidimensional scaling analysis to assess the similarities and differences between the immunogenomic signatures of HER2-positive and HER2-negative breast cancers.

[0086] Fig.33 Figure 1 shows that HER2-positive metastatic breast cancer exhibits a unique immunogenomic signature compared to HER2-positive primary breast cancer. a) Schematic diagram of the experimental model used to establish the immunogenomic signature of HER2-positive primary and metastatic breast cancer. b) Representative SERS spectra of HER2-positive primary breast cancer and HER2-positive metastatic breast cancer. c) Graphical network analysis showing the differences between HER2-positive primary breast cancer and HER2-positive metastatic breast cancer based on the immunogenomic signature. d) Boxplot comparing the normalized SERS intensity of DNA peaks that contribute to the differences in the immunogenomic signature of HER2-positive primary breast cancer and HER2-positive metastatic breast cancer. For each comparison in d), the groups were statistically different using Student's t-test with p values ​​<0.001 to <0.0001.

[0087] Fig.34Validation of the exclusivity of the immunogenomic signature of metastatic breast cancer in clinical samples is shown. a) Hierarchical clustering (left) and violin plot (right) showing the similarities of DNA from metastatic breast cancer tissue biopsies and HER2-positive metastatic breast cancer profiles. b) Graphical network analysis of serum from clinical samples of patients diagnosed with metastatic breast cancer compared to HER2-positive metastatic breast cancer profiles. c) Hierarchical clustering (left) and violin plot (right) showing the differences of DNA from metastatic breast cancer tissue biopsies and HER2-positive primary breast cancer profiles. d) Graphical network analysis of serum from clinical samples of patients diagnosed with metastatic breast cancer compared to HER2-positive primary breast cancer profiles.

[0088] Fig.35 Deconvolution of SERS spectral data using synthetic mixtures is shown, a) SERS spectra of tumor DNA mixed with different percentages of immune cell DNA. b) SERS spectra of epithelial DNA mixed with different percentages of immune cell DNA. c) Linear dependence of characteristic peak intensity on the percentage of immune cell DNA in the mixture, d) Linear regression analysis using immune genome features to determine cell type, e) Estimated and actual cell type ratios for each percentage of immune cell DNA. f) Synthetic mixture data was used to train the algorithm, and the algorithm was validated using independent clinical tissue biopsy samples (left) and liquid biopsy samples (right) cohorts.

[0089] Fig.36 Figure 1 shows that the metastatic immune genomic signature can accurately predict breast cancer metastasis and whether the metastasis is intracranial. a) Schematic diagram of the experimental design, b) Model prediction of HER2-positive phenotype in an independent clinical sample cohort, c) Performance of the machine learning model evaluated using the predicted specificity (99.3%) and sensitivity (97.3%), d) Decision tree showing the main features of the data that contribute to the accurate prediction of the HER2-positive phenotype. The HER2-positive immune genomic signature can accurately classify the phenotype of breast cancer, e) Model prediction of metastatic phenotype in an independent clinical sample cohort, f) Performance of the machine learning model evaluated using the specificity (97.2%) and sensitivity (100%) of the classification algorithm, g) Decision tree showing the main features of the data that contribute to the accurate classification of the metastatic phenotype, h) Model prediction of intracranial metastatic sites in an independent clinical sample cohort, i) Performance of the machine learning model. J) Principal components that contribute to the accurate prediction of intracranial metastatic sites.

[0090] Fig.37Figure 3 shows that NK cell DNA signatures can distinguish between tumor-interacting and non-tumor-interacting phenotypes. A shows a schematic of the experimental design using the sensor to generate Raman spectra. B shows Raman spectra of non-interacting (bottom) and tumor-interacting (top) NK cell DNA. C shows a univariate analysis using a t-test to compare the major peaks in the two phenotypes. * indicates p<0.05. ** indicates p<0.01. *** indicates p<0.001. D shows a principal component analysis (PCA) showing that the two samples have a clear clustering relationship. E shows a hierarchical cluster analysis, which further supports the clear separation of the clusters seen in the PCA, indicating that the phenotypes are different.

[0091] Fig.38 Figure 3 shows that NK cell DNA methylation signatures can distinguish phenotypes that interact with colorectal tumors from those that do not. A shows the best standard curve generated with 5-mC standard control based on OD at 450nm. B shows the linear correlation between peak intensity and methylation percentage. C shows SERS spectra of 5-mC with different methylation percentages. Peak intensity also rises as the methylation percentage increases. D shows colorimetric and Raman calculation results for comparison of methylation-specific peaks (765, 924, 1020, 1310, and 1555) in colo205 interaction and healthy controls (HCT). E shows comparison of methylation peaks in NK cell DNA interacting with colorectal cancer and healthy NK cell DNA. F shows principal component analysis with methylation-specific peaks. G shows Raman spectra comparison of NK cell methylation-specific peaks in colorectal cancer patients and healthy patients. Increased methylation of immune cells is seen in colorectal cancer patients.

[0092] Fig.39 NK cell methylation signatures are shown to differentiate between colorectal cancer patients and healthy volunteers. A shows Raman spectra of blood (buffy coat) from colorectal cancer patients (n=7). B shows Raman spectra of blood (buffy coat) from healthy volunteers (n=10). C shows a scree plot for determining the quantities of selected major components. Maximum variance was obtained using PC1 and PC2. D shows principal component analysis showing clear separation of clusters using Raman spectra of blood (buffy coat) from colorectal cancer patients and blood (buffy coat) from healthy volunteers. E shows a loading plot showing characteristic peaks that differentiate Raman spectra from healthy people and cancer patients. F shows a heat map showing the differences in spectral intensity of Raman spectra of samples from colorectal cancer patients and healthy volunteers.

[0093] Fig.40Exploratory analysis of enhanced data from cell cultures that generated NK cells that interacted with colorectal tumors in vitro and NK cells that did not interact with tumors is shown with DNA structure and methylation peaks. (A) PCA shows clear separation of DNA clusters that did not interact and DNA clusters that interacted with tumors. (B) Biplot for diagnosis. The healthy control feature used for classification is 1276 cm -1 、1358cm -1 、1515cm -1 、1662cm -1 and 1673cm -1 The colorectal cancer signature used for classification is 765cm -1 、924cm -1 、1011cm -1 、1123cm -1 、1310cm -1 、1555cm -1 、1560cm -1 and 1623cm -1 . PLSDA was then trained using the DNA structure and methylation peaks in the patient data. 10 spectra (5 from healthy subjects and 5 from buffy coats of patients) were separated as a blind dataset for testing. C shows the ROC of the prediction with an AUC of 1.0. The PLSDA CV is Venetian blinds and the ANN CV is 80:20 data. D shows the performance of the ANN model, including model accuracy and model loss.

[0094] Fig.41A schematic diagram of glioblastoma (GBM) detection by GBM-associated NK cell-derived extracellular vesicles and supervised machine learning classifiers is shown. The left figure shows the collection of training datasets, with GBM-associated immune vesicle profiles and tumor-independent immune vesicle profiles as GBM and healthy categories, respectively. The GBM-associated immune vesicle profile was obtained by co-culturing NK92 with A172, U118, and T98 cells to represent the cancer dataset. The tumor-independent immune vesicle profile was obtained from non-interacting NK92 to represent the healthy dataset. The middle figure shows model learning, where the training dataset was reduced in dimensionality using PCA (principal component analysis) and then PLSDA (partial least squares discriminant analysis) and ANN (artificial neural network) algorithms were applied. The right figure shows circulating immune vesicle (CIV) biomarkers that cross the blood-brain barrier (BBB) ​​for cancer detection. The trained algorithm was validated using spectra from GBM cancer patient sera and healthy donor sera. The sensitivity of predicting GBM using this model was 96.82% and the specificity was 100%, the area under the curve (AUC) of PLSDA was 0.9568, the sensitivity and specificity were 100%, and the AUC of ANN was 1.0.

[0095] Fig.42Characterization of immune cells and their EVs is shown, and NK cell EV signatures are identified as candidates for the diagnosis of glioblastoma. (A) Natural killer cells infiltrate glioblastoma tumors, (i) Heat map showing the immune landscape in glioblastoma tumors (n = 174). (ii) Relative proportions of immune cell populations found in glioblastoma. (B) Relative number of immune cells infiltrating glioblastoma tumors with the highlighted NK cells used in this study, (iii) Relative gene expression of immune cells in glioblastoma patients, (iv) Heat map showing immune cell gene expression in 174 glioblastoma samples. (B) Unique spectral signatures of natural killer cell EVs in mixed immune cell populations, (a) Schematic diagram of the interaction of immune cells and their extracellular vesicles with the immunoassay platform, (b) SERS spectra of single immune cell extracellular vesicles, (i) B cell EVs, (ii) dendritic cell EVs, (iii) macrophage EVs, (iv) NK cell EVs, (v) T cell EVs. (c) PCA can distinguish natural killer cell EVs from immune cell EV mixtures. (C) Similarities between NK cells and their EVs. (i) Schematic diagram of extracellular vesicles (exosomes and microvesicles) released by natural killer cells, (ii) Transmission electron microscopy (TEM) of NK cells and their associated extracellular vesicles, (iii) TEM of NKEVs, which have a nearly spherical morphology, electron-dense outer membrane, and electroluminescent inner layer, (iv) and (v) SERS spectra of NK cell EVs (iv) and NK cells (v). (vi) Chord diagram, where the red chords represent parental NK cells and the blue chords represent NKEVs. The lower half of the circle represents immune cells and their components. The interconnected chords show the similarities between NK cells and NKEVs. (D) Characterization of NK cell EVs. (i) Flow cytometry dot plot of CD56-positive NKEVs in healthy patient serum, and histograms of CD56-positive and -negative populations, (ii) Snapshot of NKEVs during the measurement, with an average concentration of 8.36 x 10 7 The NKEV size ranged from 50 to 450 nm with an average of 205 nm (n=101).

[0096] Fig.43Different phenotypes of glioblastoma (GMB)-associated and non-associated CIVs are shown. (A) Schematic diagram depicting the interaction between immune cells (NK cells) and tumor cells in the tumor microenvironment. Glucose and amino acid consumption deplete the fuel of immune cells, so immune cells undergo metabolic reprogramming. In addition, fatty acids and lactate produced by tumor cells acidify the tumor environment, exacerbating the reprogramming process. (B) SERS spectra of GBM-associated and non-interacting immune cell vesicles. The GBM-associated immune cell vesicle map was obtained by co-culturing A172, U118, and T98 GBM cells with NK-92 cells. (C) Bar graphs showing the differences in pyruvate (left), lactate (middle), and glucose (right) between non-interacting and GBM-associated vesicles, and their corresponding regression plots (R2 = 0.9392 for glucose-pyruvate and R2 = 0.9396 for glucose-lactate, P < 0.0001). (D), (E), and (F) show the PA of lipid, protein, and nucleic acid composition, respectively. A clear separation of clusters between vesicles interacting with GBM and non-interacting vesicles can be seen. (G), (H), and (I) show the two-tailed t-test of lipid, protein, and nucleic acid between non-interacting immune cell vesicles and GBM-associated immune cell vesicles, respectively. -1 In GBM-associated immune vesicles, this led to increases in CH vibrations of lipids, bone-stretching lipids, fatty acids, and CH2 vibrations in lipids, respectively. In non-interacting immune vesicles, amide I (1672 cm -1 ) and phenylalanine (1000cm -1 ) increased; while in GBM-associated immune vesicles, amide III (1235cm -1 ) and the Ch2 deformation of proteins (1447 cm -1 ) increased (****=P<0.0001, ***=P<0.001, **=P<0.01).

[0097] Fig.44 Comparison of PDL-1 and CTLA-4 markers in GBM-associated and unassociated NK cell EVs is shown. (A) Characteristic Raman spectra of PDL1 and CTLA-4. (B) Raman spectra of GBM-associated immune vesicles (top) and non-interacting immune vesicles (bottom). (C) PDL1 and CTLA-4 peaks can be traced back to 848 cm -1 、855cm -1 、931cm -1 、936cm -1 、1121cm -1 、1250cm -1 、1335cm-1 、1344cm -1 、1463cm -1 、1460cm -1 、1658cm -1 (D) and (E) show PCA and biplots for PDL1 and CTLA-4, respectively. (F) and (G) t-tests were used to show the relationship between the expression of PDL1 (F) and CTLA-4 (G) in GBM-associated immune vesicles and non-interacting immune vesicles, showing significant differences in PDL1 expression and CTLA4 between EV phenotypes. (H) Flow cytometry showed the expression of PDL1 (i) and CTLA-4 (ii) in GBM-associated immune vesicles and non-interacting immune vesicles. **** = P < 0.0001, *** = P < 0.001, * = P < 0.005.

[0098] Fig.45 Unique features of NK cell EVs are shown to be useful for diagnosing glioblastoma (GBM). (A) Immunoassays can identify GBM features in NK cell-derived extracellular vesicles, as well as non-cancer features in non-interacting NK cell-derived extracellular vesicles. SERS spectra of EVs derived from NK-92 interacting with (top to second from bottom) A172, U118, and T98 were used as input training data for the cancer dataset, and SERS spectra of non-interacting NK-92 cell-derived EVs (bottom) were used as healthy controls. Validation was performed using SERS spectra from serum samples of GBM patients (v) and healthy controls (vi). (B) Principal component analysis of non-interacting vesicles and vesicles interacting with GBM, showing separate distinct clusters. (C) Partial least squares discriminant analysis showing score scatter plots for training and validation models (n=20). (iii) Model prediction results using test data (n=20) from GBM and healthy patient sera, and model performance evaluation using ROC analysis. Sensitivity = 96.82%, specificity = 100%, AUC = 0.9568. (viii) VIP scores of variables with importance scores are plotted on the x-axis and the corresponding variables are plotted on the y-axis. (D) Artificial neural network (ANN) analysis, the results complement PLSDA. ANN showed 100% sensitivity, specificity, accuracy and precision.

[0099] Fig.46Circulating immune vesicles (CIVs) interacting with three brain cancers (astrocytoma, oligodendroglioma, and glioblastoma) are shown to display three distinct phenotypes. (A) SERS spectra of astrocytoma-associated CIVs. (B) SERS spectra of oligodendroglioma-associated CIVs. (C) Principal component analysis showing PC1 and PC2 score plots with separate clusters of oligodendroglioma-, astrocytoma-, and GBM-associated CIVs. Red, glioblastoma-associated CIVs. The top cluster is oligodendroglioma-associated CIVs. The bottom right cluster is astrocytoma-associated CIVs. The bottom left cluster is glioblastoma-associated CIVs. (D) and (E) show the loading plots of PC1 and PC2, respectively, showing classification markers for oligodendroglioma versus GBM and astrocytoma versus GBM, respectively. (F) Comparison of spectral markers for three different brain cancer types. (G) Hierarchical clustering analysis showed clear clustering of three different types of brain tumors (astrocytoma, oligodendroglioma, and glioblastoma).

[0100] Fig.47 Diagnosis of ovarian cancer using T cell extracellular vesicles (EVs) associated with ovarian cancer is shown. The upper left shows Raman spectra of EVs from healthy T cells (bottom) and ovarian cancer-associated T cells (top). The upper right shows a clear change in Raman intensity, where the heat map (top) and PCA map (bottom) show clear clustering, thus showing the discrimination ability of the method. The clusters on the left are EVs from ovarian cancer-associated T cells, and the clusters on the right are EVs from healthy T cells. The middle figure shows the PLSDA validation of the model using patient plasma, showing the ability to accurately classify ovarian cancer. The horizontal dashed line represents a prediction above 50%. Healthy plasma is represented by light circles, while ovarian cancer plasma is represented by dark circles. The lower left shows an evaluation of the performance of the model for classifying samples as healthy or ovarian cancer using a receiver operating characteristic curve (ROC) analysis. ROC shows the accuracy of the classification, with an area under the curve (AUC) = 0.9412. The calibration curve is shown at the lower right, indicating a predicted response sensitivity of 60% and a specificity of 100%.

[0101] Fig.48 Showing that T cell signatures associated with brain tumors can predict early and late stage cancer diagnosis. The left image shows Raman maps of T cells associated with early brain cancer (bottom) and late stage brain cancer (top). The top right shows principal component analysis showing changes in T cell signatures between early stage cancer (right cluster) and late stage cancer (left cluster), allowing for early brain cancer diagnosis. The bottom right shows hierarchical cluster analysis showing different clusters in the dendrogram for T cells associated with early stage brain cancer (bottom cluster) and T cells associated with late stage brain cancer (top cluster).

[0102] Fig.49Showing that immune cell signatures can distinguish between benign and malignant cancers. The upper left corner is the SERS spectrum of T cells associated with benign brain tumors. The upper right is the SERS spectrum of T cells associated with malignant tumors. The middle shows that the different signatures of T cells associated with benign and malignant tumors form different clusters in the PCA analysis. The figure below shows the ANN prediction model that can classify benign and malignant tumors with 100% accuracy.

[0103] Fig.50A is a block diagram showing steps of a method of cancer detection according to at least one example embodiment described herein.

[0104] Fig.50B According to at least one example embodiment described herein, Fig.50A Block diagram of the hardware components used to obtain Raman spectra of biomarkers on the nanosensor during the methodological process.

[0105] Fig.50C It is available with Fig.50B A block diagram of an example implementation of a computing device and software used together with a hardware configuration.

[0106] Fig.50D is a flow chart of a method for evaluating a patient sample for cancer using a Raman spectrum obtained from the patient sample according to at least one example embodiment described herein, the method may be performed by Fig.50C The software is executed by the computing device.

[0107] Fig.50E is a flowchart of the method for training to obtain a model, which can be Fig.50C The computing device's software is used to detect various conditions from patient samples.

[0108] Other aspects and features of the example embodiments described herein will become apparent from the following description taken in conjunction with the accompanying drawings. DETAILED DESCRIPTION OF THE INVENTION

[0110] Various apparatuses, methods, and compositions are described below to provide examples of at least one embodiment of the claimed subject matter. Any embodiment described below does not limit any claimed subject matter, and any claimed subject matter may cover different devices and methods than described below. The claimed subject matter is not limited to apparatuses, methods, and compositions having all the features of any one of the apparatuses, methods, or compositions described below, or to features common to multiple or all of the apparatuses, methods, or compositions described below. The apparatuses, methods, or compositions described below may not be embodiments of any claimed subject matter. Any subject matter disclosed in the apparatuses, methods, or compositions described herein but not claimed in this document may be the subject of another means of protection, such as a continuing patent application, and the applicant, inventor, and / or owner do not intend to abandon, waive, or disclose to the public any such invention by disclosing such invention in this document.

[0111] In addition, it should be understood that for simplicity and clarity of explanation, reference numerals may be repeated in the figures to indicate corresponding or similar elements where deemed appropriate. In addition, in order to provide a thorough understanding of the example embodiments described herein, many specific details are also set forth. However, it should be understood by those of ordinary skill in the art that the example embodiments described herein may be practiced without these specific details. In other cases, well-known methods, procedures, and components are not described in detail to avoid blurring the example embodiments described herein. In addition, this description should not be considered to limit the scope of the example embodiments described herein.

[0112] It should be noted that the degree terms such as "substantially", "about" and "approximately" used in this article mean a reasonable amount of deviation of the modified term so that the end result will not be significantly changed. If such deviation does not negate the meaning of the modified term, these degree terms should be understood to include the deviation of the modified term, such as 1%, 2%, 5% or 10%.

[0113] In addition, any numerical ranges expressed herein by endpoints include all numbers and fractions within that range (e.g., 1-5 includes 1, 1.5, 2, 2.75, 3, 3.90, 4, and 5). It should also be understood that all numbers and fractions thereof are modified by the term "about," meaning that the referenced number may vary by a certain amount, such as 1%, 2%, 5%, or 10%, if the end result is not significantly changed.

[0114] It should also be noted that the term "and / or" as used herein is intended to mean an inclusive "or". That is, "X and / or Y" is intended to mean, for example, X, Y, or X and Y. As a further example, "X, Y, and / or Z" is intended to mean X or Y or Z, or any combination thereof. In addition, the expression A, B, and C means various combinations, including A; B; C; A and B; A and C; B and C; or A, B, and C.

[0115] As used herein, the term "subject" includes all members of the animal kingdom, including mammals, and appropriately refers to humans. Therefore, the methods and uses of the present application are applicable to both human treatment and veterinary applications. In one embodiment, the subject is a patient.

[0116] As used herein, the term "cell" refers to a single cell or a plurality of cells, including cells in cell culture or in a subject.

[0117] The example embodiments of the apparatus, system or method described in accordance with the teachings herein may be implemented as a combination of hardware and software. For example, the embodiments described herein may be implemented at least in part by using one or more computer programs that are executed on one or more programmable devices comprising at least one processing element and at least one storage element (i.e., at least one volatile memory element and at least one non-volatile memory element). The hardware may include an input device, and (depending on the implementation of the hardware) one or more of a display, a speaker, a printer, etc., wherein the input device includes at least one of the following: a touch screen, a keyboard, a mouse, a button, a key, a slider, etc.

[0118] It should also be noted that there may be some elements for implementing at least a portion of the embodiments described herein that can be implemented by software written in a high-level programming language (e.g., object-oriented programming). The program code can be written in MATLAB, Julia, Python, C, C++, or other languages. ++ Or any other suitable programming language, and may contain modules or classes, as known to those skilled in object-oriented programming. Alternatively or in addition, some of these elements implemented by software may be written in assembly language, machine language, or firmware as desired. In either case, the language may be a compiled language or an interpreted language.

[0119] At least some of these software programs can be stored on a computer-readable medium, such as but not limited to: ROM, disk, CD, USB key, etc., which can be read by a device having a processor, an operating system, and related hardware and software, which are necessary to implement the functions of at least one embodiment described herein. When the device reads the software program code, it configures the device to operate in a new, specific and predefined way to perform at least one method described herein.

[0120] In addition, at least some of the programs associated with the devices, systems, and methods of the embodiments described herein may be distributed in a computer program product comprising a computer-readable medium carrying computer-usable instructions, such as program code, for one or more processing units. The medium may be provided in a variety of forms, including non-transitory forms, such as, but not limited to, one or more floppy disks, optical disks, tapes, chips, and magnetic and electronic memory. In alternative embodiments, the medium may be transient in nature, such as, but not limited to, wired transmissions, satellite transmissions, Internet transmissions (e.g., downloads), media, digital and analog signals, and the like. Computer-usable instructions may also be in a variety of formats, including compiled and non-compiled code.

[0121] The following description is not intended to limit or define any subject matter that has been claimed or not yet claimed. Subject matter that may be claimed may exist in any combination or subcombination of elements or method steps disclosed in any part of this document (including its claims and drawings). Therefore, it should be understood by those skilled in the art that the device, system or method disclosed according to the teachings of this document may embody any one or more features contained herein, and these features may be used in any specific combination or subcombination that is physically feasible and achievable to achieve its intended purpose.

[0122] Recently, there has been an increasing interest in developing new technologies for early cancer detection and prognosis. In particular, there has been an increasing interest in identifying new biomarkers in human physiological (e.g., blood) samples for early cancer detection and prognosis. In particular, the present disclosure relates to the use of tumor-associated immune cells as biomarkers for cancer.

[0123] Thus, in one embodiment, provided herein is a method for providing a cancer assessment to a subject, the method comprising:

[0124] measuring the value of one or more cancer traits of at least one tumor-associated immune cell (TAIC) or TAIC-derived biomarker in a fluid sample obtained from the subject to obtain a test profile;

[0125] comparing the test profile to one or more reference profiles;

[0126] wherein similarity or difference between the test profile and one or more reference profiles is indicative of one or more cancer characteristics; and

[0127] Based on the one or more cancer characteristics, a cancer assessment is provided for the subject.

[0128] Also provided herein is a method of determining one or more cancer characteristics in a subject, the method comprising:

[0129] measuring the value of one or more cancer traits of at least one tumor-associated immune cell (TAIC) or TAIC-derived biomarker in a fluid sample obtained from the subject to obtain a test profile;

[0130] comparing the test profile to one or more reference profiles;

[0131] Wherein a similarity or difference between the test profile and one or more reference profiles is indicative of one or more cancer characteristics.

[0132] As used herein, the term "tumor-associated immune cell" or "TAIC" refers to an immune cell that is or has been present in the tumor microenvironment. "Tumor microenvironment" or "tumor microenvironment" (TME) refers to the environment surrounding the tumor, including surrounding blood vessels, immune cells, fibroblasts, signal transduction molecules, and extracellular matrix (ECM).

[0133] Immune cells include, but are not limited to, T cells, natural killer (NK) cells, macrophages, neutrophils, myeloid-derived suppressor cells (MDSC), dendritic cells (DC), B cells, and mast cells.

[0134] TAICs include, but are not limited to: tumor-associated T cells, including cancer cell-associated T cells, cancer stem cell-associated T cells, and tumor-associated stem cell-like T cells; tumor-associated NK cells, including cancer cell-associated NK cells, cancer stem cell-associated NK cells, and tumor-associated stem cell-like NK cells; tumor-associated macrophages, including cancer cell-associated macrophages, cancer stem cell-associated macrophages, and tumor-associated stem cell-like macrophages; tumor-associated neutrophils, including cancer cell-associated neutrophils, cancer stem cell-associated neutrophils, and tumor-associated stem cell-like neutrophils; tumor-associated dendritic cells, including cancer cell-associated dendritic cells, cancer stem cell-associated dendritic cells, and tumor-associated stem cell-like dendritic cells; and tumor-associated B cells, including cancer cell-associated B cells, cancer stem cell-associated B cells, and tumor-associated stem cell-like dendritic cells.

[0135] As used herein, the term "tumor-associated immune cell (TAIC)-derived biomarker" refers to a biomarker derived from or associated with a tumor-associated immune cell. Examples of "tumor-associated immune cell (TAIC)-derived biomarkers" include, but are not limited to, extracellular vesicles of tumor-associated immune cells, cell-free DNA from tumor-associated immune cells, and methylated DNA from tumor-associated immune cells. Other examples of "tumor-associated immune cell (TAIC)-derived biomarkers" include metabolic changes, activation states, and surface phenotypes of TAIC.

[0136] In one embodiment, TAIC or a TAIC-derived biomarker circulates in the patient. Thus, in one embodiment, the method comprises obtaining a liquid sample from the subject. The liquid sample is optionally blood, plasma, serum, urine, feces, mucus, saliva, cerebrospinal fluid, and / or amniotic fluid. In one embodiment, the liquid sample is blood.

[0137] In another embodiment, TAIC or a TAIC-derived biomarker is present in a patient's tissue. Thus, in one embodiment, the method comprises obtaining a tissue sample from a subject, optionally by tissue biopsy.

[0138] As used herein, the term "cancer trait" refers to a property of or associated with TAIC or a TAIC-derived biomarker that indicates the presence or absence of cancer or is associated with the presence or absence of cancer. In one embodiment, a "cancer trait" is a measurable value.

[0139] For example, in one embodiment, the cancer trait is a Raman spectrum of at least one TAIC or TAIC-derived biomarker. In one embodiment, the Raman spectrum is obtained using the methods described in detail herein and in PCT Application No. PCT / CA2022 / 050232, the entire contents of which are incorporated herein.

[0140] According to the methods described herein, the values ​​of one or more cancer traits are measured to obtain a test profile (also referred to herein as a sample profile). The test profile can be a single measurement or multiple measurements. In one embodiment, the test profile is a Raman spectrum, optionally a surface enhanced Raman spectrum (SERS).

[0141] In one embodiment, the value of one or more cancer traits is measured by DNA next generation sequencing, PCR, repertoire sequencing, immunosequencing, single cell RNA sequencing, DNA methylation sequencing, unique molecular identifiers (UMIs), light microscopy, optionally confocal microscopy, immunohistochemistry or super-resolution microscopy, flow cytometry or mass cytometry.

[0142] The method also includes determining a level of similarity between the test profile and one or more reference profiles (also referred to herein as templates or template profiles).

[0143] The reference profile can be a reference value and / or can be from one or more samples, optionally from historical data of a subject or subject pool known to have or not have cancer or to have a particular type, grade or stage of cancer. The historical data can be values ​​that are continuously updated as further samples are collected and it is determined whether the subject has cancer, or to have a particular type, grade or stage of cancer. For example, a control database can be stored in an online database that is continuously updated with data from confirmed patients.

[0144] The reference profiles are optionally derived from one or more reference TAICs. In one embodiment, the reference TAICs are immune cells, such as immune cells from healthy subjects (also referred to herein as "naive immune cells"), which are cultured in the presence of tumors and / or cancer cell lines until one or more reference tumor-associated immune cells (TAICs) express one or more cancer traits.

[0145] In another embodiment, the reference TAIC is an immune cell obtained from a cancer subject.

[0146] A reference profile generated by one or more reference TAICs is an example of a "cancer-specific reference profile." A reference profile generated by one or more naive immune cells is an example of a control reference profile, a healthy reference profile, or a "non-cancer-specific reference profile."

[0147] When the reference profile is a cancer-specific reference profile (e.g., a reference profile indicative of cancer or a cancer having one or more cancer characteristics), (i) the similarity of the test profile to the cancer-specific reference profile is high; and / or (ii) the similarity of the test profile to the cancer-specific reference profile is higher than the similarity to the non-cancer-specific reference profile, indicating the presence of cancer or a cancer having one or more cancer characteristics, or an increased likelihood of cancer or a cancer having one or more cancer characteristics.

[0148] When the reference profile is a non-cancer specific reference profile (e.g., a reference profile that is not indicative of cancer), (i) the similarity of the test profile to the non-cancer specific reference profile is low; and / or (ii) the similarity to the non-cancer specific reference is lower than the similarity to the cancer specific reference profile, indicating the presence of cancer or a cancer with one or more cancer characteristics, or an increased likelihood of cancer or a cancer with one or more cancer characteristics.

[0149] As used herein, "cancer characteristic" may include, but is not limited to, the presence or absence of cancer, cancer type, cancer stage, cancer grade, primary cancer, metastatic cancer, cancer metastatic potential, molecular phenotype of cancer, benign cancer, malignant cancer, cancer tissue of origin, cancer metastasis location, or a combination thereof.

[0150] In one embodiment, a higher similarity to a cancer-specific reference profile compared to a non-cancer-specific reference profile is indicated by tighter clustering between the test profile and the cancer-specific reference profile than between the test profile and the non-cancer-specific reference profile, as measured by, for example, unsupervised multivariate analysis (principal component analysis or PCA).

[0151] Likewise, lower similarity to a cancer-specific reference profile than to a non-cancer-specific reference profile is indicated by tighter clustering between the test profile and the non-cancer-specific reference profile than between the test profile and the cancer-specific reference profile, as measured by, for example, unsupervised multivariate analysis (principal component analysis or PCA). "Tight clustering" can be indicated, for example, by two profiles falling within one cluster or forming their own separate clusters.

[0152] In another embodiment, a higher similarity to a cancer-specific reference profile compared to a non-cancer specific reference profile is indicated by a calculated correlation value between the test profile and the cancer-specific reference profile being higher than an equivalent calculated correlation value between the test profile and the non-cancer specific reference profile, optionally wherein the correlation value is a correlation coefficient.

[0153] In another embodiment, a lower similarity to a non-cancer specific reference profile compared to a cancer specific reference profile is indicated by a calculated correlation value between the test profile and the non-cancer specific reference profile being lower than an equivalent calculated correlation value between the test profile and the cancer specific reference profile, optionally wherein the correlation value is a correlation coefficient.

[0154] Without being bound by theory, the reference profile of the cancer characteristic "metastatic cancer" according to the methods described herein can be used to identify early stage samples or preliminary diagnostic samples having the cancer characteristic "cancer metastatic potential".

[0155] In another embodiment, the correlation coefficient is a linear correlation coefficient, optionally a Pearson correlation coefficient or a Spearman correlation coefficient.

[0156] As used herein, the term "cancer" refers to a cell proliferative disease state.

[0157] In one embodiment, the cancer is brain cancer, optionally a glioblastoma, astrocytoma, or oligodendroglioma. In another embodiment, the cancer is ovarian cancer. In another embodiment, the cancer is bladder cancer, breast cancer, colon cancer, esophageal cancer, stomach cancer, liver cancer, intestinal cancer, lung cancer, ovarian cancer, head and neck cancer, rectal cancer, prostate cancer, pancreatic cancer, thyroid cancer, cervical cancer, melanoma, nasopharyngeal cancer, testicular cancer, or uterine cancer.

[0158] In one embodiment, the breast cancer is HER2+ breast cancer, triple positive breast cancer, or triple negative breast cancer.

[0159] In one embodiment, the cancer is an adult cancer or a childhood cancer.

[0160] As used herein, the term "cancer assessment" includes, but is not limited to, any assessment that can be provided to a patient or a care provider, including, for example, providing the presence of cancer, the type of cancer, the location of cancer, the stage of cancer, the grade of cancer, the potential for cancer to metastasize, or the effectiveness of cancer treatment. "Cancer assessment" also includes providing a patient with a prognosis, early diagnosis of cancer, determining whether a tumor is benign or malignant, determining the grade of cancer, determining whether a tumor is primary or metastatic, determining whether a primary tumor has metastatic potential, determining the progression of cancer, determining lymph node metastasis of cancer, determining clinical metastasis of cancer, predicting patient survival, providing a patient with a prognosis, providing an early diagnosis of cancer, including providing an early diagnosis of difficult-to-detect cancers, determining the presence of aggressive brain cancers, providing the location of a tumor, monitoring cancer recurrence during or after treatment, and / or determining the presence of minimal residual disease.

[0161] The present disclosure also provides methods of treating a subject diagnosed with cancer, comprising: (a) assessing the subject for cancer according to the methods described herein; and (b) treating the subject.

[0162] As used herein, the term "treating / treatment" is well known in the art and refers to a method for obtaining beneficial or desired results (including clinical results). Beneficial or desired clinical results include, but are not limited to, relieving or improving one or more symptoms or conditions, reducing the extent of the disease, stabilizing (i.e., not worsening) the disease state, preventing the spread of the disease, delaying or slowing the progression of the disease, improving or alleviating the disease state, reducing disease recurrence, and alleviating (whether partially or completely) the disease, whether detectable or undetectable. "Treatment" can also mean extending the survival period compared to the expected survival period without treatment. For example, a subject with early-stage cancer can be treated to prevent the progression of the disease, or a subject in remission can be treated to prevent recurrence. Therapeutic methods can include administering to the subject a therapy such as a chemotherapeutic drug, radiation, and / or performing surgery on the subject.

[0163] As used herein, "treating cancer" includes, but is not limited to, reversing, alleviating or inhibiting the progression of cancer or symptoms or conditions associated with cancer. "Treating cancer" also includes extending the patient's survival. The survival period is optionally extended by at least 1, 2, 3, 6 or 12 months, or at least 2, 3, 4, 5 or 10 years compared to the expected survival period without treatment with the cytotoxic agent or composition described herein. "Treating cancer" also includes reducing tumor mass and / or shrinking tumors. Optionally, after treatment with the cytotoxic agent or composition described herein, the tumor mass and / or tumor burden is reduced by at least 5%, 10%, 25%, 50%, 75% or 100%. "Treating cancer" also includes reducing the aggressiveness, grade and / or invasiveness of the tumor.

[0164] The present application also provides a method for generating reference tumor-associated immune cells (TAICs) for cancer assessment, the method comprising co-culturing one or more initial immune cells with a tumor and / or cancer cell line until one or more reference tumor-associated immune cells (TAICs) express one or more cancer traits.

[0165] In one embodiment, the one or more primary immune cells comprise T cells, NK cells, macrophages, neutrophils, dendritic cells, or B cells, or a combination thereof.

[0166] Now referring to the accompanying drawings, Fig.50A An example embodiment of a method 100 of assessing cancer in a subject according to the teachings herein is shown. Fig.50B An example embodiment of a system 200 for acquiring a Raman spectrum of a biomarker captured by a nanosensor, as described in the methods described herein, is shown.

[0167] exist Fig.50A In step 102, a liquid sample (e.g., raw plasma; plasma collected from a subject) is collected and processed to produce a certain volume of cell-free plasma, a certain volume of buffy coat, and a certain volume of remaining product (e.g., red blood cells, etc.). The processing in step 102 can be performed using a suitable technique, such as but not limited to density gradient centrifugation, to obtain cell-free plasma, buffy coat, and remaining product (e.g., red blood cells, etc.).

[0168] In at least one embodiment, in step 103, a volume of plasma (e.g., cell-free plasma) produced in step 102 is dripped onto a first nanosensor to detect cellular component biomarkers. In at least one embodiment, the volume of plasma contains at least one tumor-associated immune cell (TAIC) or a tumor-associated immune cell (TAIC)-derived biomarker.

[0169] In at least one embodiment, the amount of plasma added to the nanosensor is in the range of about 10 microliters (μL).

[0170] The surface enhanced Raman scattering nanosensor for biomolecule detection herein includes one or more nanoparticles (e.g., nanoprobes). The nanoparticles are made of materials such as gold, silver, platinum, titanium, silicon, aluminum, nickel and / or graphite, but are not limited thereto. The nanoparticles described herein may also be referred to as quantum dots (i.e., nanoparticles having a particle size of less than about 5 nm).

[0171] Unlike all other types of quantum dots, the quantum dots of the nanosensors described in this article are generally non-toxic, making them particularly suitable for biomedical applications. In addition, these quantum dots are free of contamination and generally do not react / interfere with target molecules.

[0172] Compared to conventional quantum dots, the quantum dots of the nanosensors described herein are smaller and have a unique structure (e.g., having a high crystalline nanoparticle vacancy density), which translates into higher detection sensitivity and pushes the detection limit to lower concentrations. Therefore, the quantum dots of the nanosensors described herein can be used to detect previously undetectable trace amounts (undetectable due to low concentrations).

[0173] In at least one embodiment, the nanosensor can amplify very weak signals of one or more TAIC or TAIC-derived biomarkers. The nanosensor includes one or more self-assembled three-dimensional nanoprobes that can detect very low levels of biomolecular structures. The nanosensor is a three-dimensional porous nanoprobe network. The interconnected and criss-crossed nanosensors can be used as capture and screening devices. In addition to the capture of biological components by the nanosensor, liquids are easily discharged from the nanosensor, thereby improving the interaction between the biological components and the nanosensor.

[0174] Typically, the nanosensors used in the methods and systems described herein have a three-dimensional (3D) structure that contains self-assembled closed rings and bridges, which cause the nanoparticles to aggregate together. The self-assembled closed rings and bridges are interconnected pores. The size of the ring is approximately equal to the wavelength of the laser beam used to make the sensor.

[0175] In at least one embodiment, nanoparticles contain abundant crystal defects. In at least one embodiment, the size of nanoparticles can be regulated between about 100nm to about 1nm. For example, in at least one embodiment, the nanoparticle size can be less than about 1nm.

[0176] In at least one embodiment, the nanosensors used in the methods and systems described herein are manufactured by the method described in U.S. Provisional Patent Application US63 / 059,079 filed on July 30, 2020, entitled "Ultrashort Laser Synthesis of Isotope Nanoparticles," the contents of which are incorporated herein by reference.

[0177] In one embodiment, in step 104, optionally, a volume of the buffy coat produced in step 102 can be dropped onto a second nanosensor to detect immune cell biomarkers. In at least one embodiment, the volume of the buffy coat comprises one or more TAIC or TAIC-derived biomarkers.

[0178] In at least one embodiment, the volume of buffy coat added to the nanosensor is in the range of about 10 μL.

[0179] According to the teachings herein, the nanosensors used can detect low concentrations of TAIC or TAIC-derived biomarkers.

[0180] Optionally, in step 105, the remaining product produced during step 102 may be discarded.

[0181] In step 106, in at least one embodiment, after the first incubation time, the first nanosensor including the volume of plasma is scanned under a Raman microscope to obtain a Raman spectrum.

[0182] In at least one embodiment, the first incubation time ranges from about 1 to about 2 minutes. The length of the first incubation time depends on the time required for the biomarker to adsorb to the nanosensor surface. The first incubation time can be determined empirically.

[0183] Also in step 106, in at least one embodiment, after the second incubation time, a second nanosensor including a volume of buffy coat is scanned under a Raman microscope to obtain a Raman spectrum.

[0184] In at least one embodiment, the second incubation time is in the range of about 1 to about 2 minutes. Similarly, the duration of the second incubation time depends on the time required for the biomarker to adsorb to the nanosensor surface. The second incubation time can be determined empirically.

[0185] In step 107, one or more Raman spectra of a volume of cell-free plasma on the nanosensor are generated by a Raman spectroscopy system.

[0186] In step 108, one or more Raman spectra of a volume of the buffy coat on the nanosensor are generated by a Raman spectroscopy system.

[0187] Fig.50B An example of a Raman spectroscopy system for obtaining a Raman spectrum of a sample on a nanosensor in steps 107 and / or 108 of method 100 is provided. Fig.50B An example embodiment of a system 200 is shown having one or more hardware components for, for example, in accordance with at least one example embodiment described herein. Fig.50A During step 106 of the method, a Raman spectrum of the biomarker on the nanosensor is obtained.

[0188] System 200 includes: computing device 202, excitation path (including laser 204, wave plate 206, beam deflector 208, beam expander 210, Rayleigh filter 212, filter and wave plate 214) and movable stage 216. The nanosensor is located on movable stage 216. System 200 also includes a return path, which includes microscope 218, filter and wave plate 214, focusing lens 220, adjustable slit 222, collimator 224, diffraction grating 226, focusing lens 228 and CCD camera 230. In at least one embodiment, system 200 is a confocal Raman microscope system, for example, with an excitation wavelength of 785nm. The microscope system provides Fig.50B All components shown except computing device 202 .

[0189] During use, the computing device 202 includes one or more software programs (see, e.g. Fig.50C ) for Rayleigh imaging and analysis of a sample 217 (such as a plasma or buffy coat sample) that has been placed on the nanosensor. Specifically, the computing device 202 is connected to the laser 204 and the CCD camera 230 via I / O hardware (which may include an analog-to-digital converter and / or a USB cable) so that the computing device 202 can control the laser 204 to generate laser pulses and the CCD camera 230 to record images. The laser pulse from the laser 204 is a transient light beam that is polarized by the wave plate 206 and then steered by the beam deflector 208. The deflected light beam is then expanded in diameter by the beam expander 210 and then filtered by the Rayleigh filter 212 and the filter and wave plate 214 to purify the scattered light, which is then provided to the microscope 218, which can be a standard microscope that includes a focusing lens for focusing the excitation light pulse onto the sample 217.

[0190] The excitation light pulse excites the molecules in the sample 217, which emit scattered photons of different energies and frequencies. The electric dipole-electric dipole polarization also changes, and the resulting Raman scattering is proportional to the change in polarization. The scattered photons (called the Raman signal) are then focused and cleaned up by optical elements in the return path before reaching the CCD camera 230. The CCD camera 206 then records the image and transmits the image to the computing device 202 so that it can perform processing, as described below.

[0191] For example, now refer to Fig.50C , which shows a computing device 202 and Fig.50B 202 is an example embodiment of a block diagram of an example embodiment of software for use with a hardware configuration of FIG. 202. It should be noted that this is an example embodiment and that in other embodiments, the computing device 202 may have more or fewer components or alternative layouts. The computing device may be a desktop computer or other suitable computing device that can wirelessly communicate with the Fig.50B The computing device communicates with some hardware component in the computer, in this case, a tablet.

[0192] The computing device 202 typically includes a processing unit 250, an analog-to-digital converter (ADC) 256, a data storage 252, a display 254, and an input / output interface 258, which can be connected to various peripheral components, such as the laser 204 and the CCD camera 230, or a pre-packaged Raman microscope including these hardware components. The computing device 202 may also include a power supply unit (not shown) or be connected to a power supply to receive the power required to operate its components.

[0193] Processing unit 250 is operably connected to other components of computing device 202 to control various operations and perform certain functions, such as setting or modifying stimulation parameters (i.e., wavelength and intensity) of laser 204, data acquisition processes (i.e., controlling image acquisition, etc.), and evaluating patient samples for various aspects of cancer as described herein.

[0194] Processing unit 250 may be any suitable processor, controller, or digital signal processor that can provide sufficient processing power according to the operational requirements of computing device 202, as known to those skilled in the art. For example, processing unit 250 may be a high-performance processor. In alternative embodiments, processing unit 250 may include multiple processors, each configured to perform different dedicated tasks.

[0195] The data storage 252 includes volatile and non-volatile memory elements, such as, but not limited to, one or more RAMs, ROMs, one or more hard disk drives, one or more flash drives, or some other suitable data storage element. The data storage 252 can be used to store operating systems and programs, as known to those skilled in the art. For example, the operating system provides various basic operating processes for the processing unit 252, and the programs include various operations and user programs so that the user can interact with the computing device 202 to perform Raman imaging on the sample, and then perform cancer assessment on the sample to determine and / or update the model used in the cancer assessment (e.g., through training), or a combination of these two operations.

[0196] The data storage 252 may also include software code for various components for implementing various aspects of Raman imaging, model training, and cancer assessment according to the teachings herein, as well as for storing values ​​of various operating parameters of Raman imaging. For example, the data storage 252 may include programs for implementing the input / output module 252a, the cancer assessment module 252b, the Raman imaging module 252c, and the training module 252d. It should be noted that there may be other embodiments in which the software modules may be organized in a different manner; however, performing the same functions as described herein.

[0197] The input / output module 252a may include program instructions for receiving acquired Raman image data and user control data. The input / output module 252a may also include program instructions for outputting and / or storing raw Raman image data, pre-processed Raman image data, and cancer assessment data.

[0198] The cancer assessment module 252b includes program instructions for acquiring and preprocessing Raman spectral data of patient samples, performing feature extraction on the preprocessed Raman spectral data, and providing feature extraction values ​​to various models (including classifiers for detecting various cancer features), wherein the classifiers can be used to perform cancer assessment on patient samples based on one or more biomarkers described herein. Cancer features may include cancer type, cancer stage, cancer grade, cancer metastasis, potential cancer metastasis, or any combination thereof. If the analysis is based on two or more biomarkers, the results of each biomarker can be combined to provide a more accurate cancer assessment than an analysis based on only one biomarker. Reference Fig.50D The operation of the cancer assessment module 252b is further described.

[0199] The Raman imaging module 252b is used to control the laser 204 to generate excitation laser pulses and record the generated images when the patient sample 217 is located on the movable stage 216. The Raman image can then be directly processed by the cancer assessment module 252b and optionally stored on the data storage 252.

[0200] The training module 252d includes program instructions for using the training data to determine the models (e.g., classifiers) used by the cancer assessment module 252b, and for updating these models over time as more training data is obtained. The training data may be used Fig.50B The training data may be obtained by hardware settings of the embodiment of the present invention, or may be obtained by other means, and then accessed by the input / output module 252a when performing training. In some embodiments, the training data may be stored in the data storage 252. Once the model is trained, the model parameters 252e may be stored in the data storage 252. The operation of the training module 252d will refer to Fig.50E Describe in more detail.

[0201] Display 254 can be any device suitable for displaying images and various types of information, such as an LCD monitor or a touch screen display. Input / output interface 258 can be various ports, such as one or more of USB, FireWire, serial and parallel ports, which can be connected to various peripheral devices for data input or output. Examples of these devices include laser 204 and CCD camera 230, and may also include, but are not limited to: keyboard, mouse, touch pad, touch interface and printer or any combination thereof. An analog-to-digital converter 256 may be required to convert any analog data received by input / output interface 258 into digital data.

[0202] return Fig.50A In step 110, the computing device 202 (i.e., its processor) performs processing to analyze the one or more Raman spectra obtained from steps 107, 108. For example, in one embodiment, in step 110, when the processor is executing one of the computer programs, one or more obtained Raman spectra are compared with one or more template Raman spectra using correlation analysis.

[0203] In at least one embodiment, the correlation analysis of one or more acquired Raman spectra with a template Raman spectrum comprises comparing characteristic peaks of Raman bands of the acquired Raman spectrum with characteristic peaks of Raman bands of the template Raman spectrum. The template Raman spectrum comprises an average spectrum of multiple individual Raman spectra from samples with known cancer characteristics.

[0204] For example, the Raman spectra of plasma / cfDNA / exosomes contain rich information. This information is usually presented in the form of characteristic peaks. For example, a peak at a specific wave number usually represents a biological molecule. Some representative Raman assignments are provided here as examples. For example, 782cm -1 The peak at 1445 cm is attributed to the ring breathing mode of DNA / RNA bases. -1 The peak at CH2 is attributed to the bending mode of the protein, 1670 cm -1The peak at 2850 cm-1 is attributed to the stretching mode of C=C in lipids. -1 The peak at is attributed to CH2 symmetric stretching in lipids. Raman assignments have been well documented by known literature. It should be understood that specific characteristic peaks of the spectrum can be selected based on the application of the data.

[0205] In at least one embodiment, the correlation analysis may include the use of one or more correlation analysis tools. Some tools for performing correlation analysis may include: Pearson correlation analysis, Spearman correlation test, heat map and / or a correlation matrix using an artificial neural network to plot eigenvalues. In at least one embodiment, during the correlation analysis, ten measurements are obtained from a sample.

[0206] In at least one embodiment, a large number of peaks in any given spectrum can be selected. The number of peaks can be determined by the molecular vibrations in each cell or biomarker. In general, for each cell or biomarker used, at least three peaks are considered. In at least one embodiment, the number of peaks used can be in the range of 3-10. In at least one embodiment, in the Raman spectrum of relatively large macromolecules (such as proteins, lipids and / or nucleic acids), there may be many peak overlaps. Therefore, when the spectrum is qualitatively analyzed and certain peaks are assigned to specific biochemical components, many inaccuracies may be introduced. These errors are caused by visual inspection and guesswork to determine the biochemical components based on intensity changes, because it is always possible that several components make a combined contribution to a peak. In addition, the omitted area of ​​the spectrum may lose important information. In order to overcome these errors, the chemometric method of multivariate data analysis can be adopted, as described in reference to Figures 61C and 61D.

[0207] In at least one embodiment, if there is a high degree of similarity between the sample Raman spectrum and the template Raman spectrum, a diagnosis can be made. For example, in at least one embodiment, if a correlation analysis of the obtained Raman spectrum (from the subject sample) and the template Raman spectrum shows a correlation of 65% or more, a diagnosis can be made based on the potential cancer characteristics of the template Raman spectrum.

[0208] In at least one embodiment, the obtained Raman spectra can be compared with multiple templates simultaneously to perform multiple diagnostics. In at least one embodiment, the comparison of the obtained Raman spectra with the templates can be performed by a computer program. For example, in at least one embodiment, Python TM and / or Modules can be used for data comparison and data interpretation.

[0209] Reference now Fig.50D, which shows a flow chart of a method 300 for evaluating a subject sample for cancer using a Raman spectrum obtained from the subject sample, which can be performed by software of a computing device 202 according to at least one example embodiment described herein. In alternative embodiments, certain aspects of the method 300 can be modified. For example, although Fig.50D The application of four classifiers is shown, but some classifiers may not have been applied based on the underlying biomarkers used to perform the classification.

[0210] In operations 302 and 304, a Raman spectrum of a patient sample is acquired and preprocessed. This can be done using Fig.50B The Raman spectrum may have been acquired, in which case it may be accessed from storage and may have been preprocessed. Preprocessing involves removing any amplitude shift caused by the baseline in the Raman spectrum data when it was first acquired. The baseline may cause uneven amplitude changes between different wave numbers and negatively affect the analysis performed by the cancer assessment module 252b. The baseline correction performed during the preprocessing process of operation 304 is known to those skilled in the art.

[0211] In operation 306, feature extraction is performed on the pre-processed Raman spectrum to obtain eigenvalues ​​used by the model in the subsequent steps of method 300. The feature extraction results in a reduction in the data dimension. Several different techniques can be used to perform the feature extraction. For example, in step 306, principal component analysis (PCA) can be used as a non-parametric method that does not require any explicit background model. In principal component analysis, the input Raman spectral data is decomposed, and the generated output is the principal component. The first principal component can be defined as the direction of maximizing the variance. The i-th component is orthogonal to the first principal component that maximizes the variance. In short, the principal component is the eigenvector of the covariance matrix of the input Raman spectral data. We can determine how many principal components to use for analysis and ignore the remaining components. PCA can be performed using known techniques to generate a certain number of principal components, such as 10 principal components. In at least one embodiment described herein, the component that shows the maximum amount of data is used (for example, 3-5 components are usually selected, which represent at least 70% of the spectral data). The principal component is a feature used as a classifier input. The calculated principal components can be further used to solve the classification problem of one or more components (e.g., extracellular vesicles) detected after the sample is placed on the nanosensor and the Raman image is captured.

[0212] Alternatively, in operation 306, multivariate curve resolution analysis (MCR) can be used to estimate the contribution of pure components to mixed measurements (e.g., a mixture of various types of cfDNA in a plasma sample), thereby providing information about the compositional profile to generate scientifically meaningful information, which in this case are features that are subsequently provided to the classifier. In this way, we can easily understand the interpretation of results from complex and large amounts of data collected by Raman spectroscopy.

[0213] Compared with PC analysis, MCR analysis generally provides more accurate information about the pure components in the mixture. Therefore, by applying it to extracellular vesicles, changes in the pure components of extracellular vesicles can be deduced. In other words, PC analysis is faster than MCR analysis. However, it should be understood that the choice of performing PC analysis or MCR analysis depends on which type of analysis is used when generating / training a classifier (e.g., a classification model). For example, if PC analysis is used to generate / train a cancer type classifier used in operation 308, PC analysis is performed in operation 306 and the extracted feature values ​​are input to operation 308. At the same time, if MCR analysis is used to generate / train a cancer stage classifier used in operation 310, MCR analysis is performed in operation 306 and the extracted feature values ​​are input to operation 310. Therefore, if PC analysis and MCR analysis are used respectively when determining / training at least one of the classifiers used in operations 308 to 314, PC analysis and MCR analysis can be performed simultaneously in operation 306.

[0214] In operation 308, the feature values ​​extracted from operation 306 are provided as input to a cancer type classifier, which provides probabilities for different cancer type categories, which are used to detect the presence of at least one type of cancer in the patient sample. The detection results are then provided to operation 316, where they are incorporated into the cancer assessment performed. It is worth noting that if no cancer type is detected in step 310, steps 310-314 do not need to be performed.

[0215] In operation 310, the feature values ​​extracted from operation 306 are provided as input to a cancer stage classifier, which provides probabilities for different cancer stage categories, which cancer type categories are used to detect what the cancer stage is if cancer is present in the patient sample. This detection result is then provided to operation 316, where it is incorporated into the cancer assessment performed.

[0216] In operation 312, the feature values ​​extracted from operation 306 are provided as input to a metastasis classifier, which provides a probability of whether the cancer has metastasized when cancer is present in the patient sample. The detection result is then provided to operation 316, where it is incorporated into the cancer assessment performed.

[0217] In operation 314, the feature values ​​extracted from operation 306 are provided as input to a potential metastasis classifier, which provides a probability of potential cancer metastasis when cancer is present in the patient sample. The detection result is then provided to operation 316, where it is incorporated into the cancer assessment performed.

[0218] It should be noted that in some embodiments, one or more of operations 308 through 314 may be optional.

[0219] In operation 316, the test results from the performed operations 308-314 (some of which may be optional) are used to provide a cancer assessment, which may be considered as a cancer diagnosis for the subject based on the analysis of the subject's sample. For example, the cancer assessment may be that the subject has a certain cancer type, a certain cancer stage, a certain cancer grade, whether the cancer has metastasized, and the likelihood of cancer metastasis. For example, the subject may have stage 4 breast cancer that has metastasized, and based on further potential metastases, a 5-year survival estimate for the subject may be provided.

[0220] The classification model used in operations 308-314 can be based on a machine learning model. For example, the machine learning model can be based on partial least squares discriminant analysis (PLSDA), support vector machine discriminant analysis (SVMDA), or PLSDA, or an artificial neural network topology. Alternatively, other types of machine learning techniques can be used, such as (but not limited to) convolutional neural networks and random forest methods.

[0221] In an alternative embodiment, the classification model may be based on a deep learning model, in which case the feature extraction step of operation 306 need not be performed because the deep learning model may perform feature extraction while determining the classification result.

[0222] However, each classifier used in operations 308-314 may be based on different types of models (some examples are listed previously) that can be used Fig.50E 308-314. In particular, the training method 400 can be used to determine the best classification model for each classifier used in steps 308-314. Each classification model parameter can be stored in the data memory 252. In at least one embodiment, the classifier used in operations 308-314 can be implemented using Matlab-based software and / or Python open source libraries.

[0223] The PLSDA-based classifier determines the probability that a sample belongs to each class and the classification threshold. This is achieved by fitting a Gaussian distribution to all calculated class probabilities.

[0224] SVMDA-based classifier uses supervised methods for classification, regression, and outlier detection. SVMDA is an effective tool in high-dimensional space. It separates features into different structural domains and calculates the probability of each sample belonging to a certain category. These probabilities are used to estimate the most likely category for each sample (e.g., predicted probability).

[0225] A classifier using an artificial neural network (ANN) uses a sequential model to build a solution to the problem layer by layer. A "hard sigmoid" activation function can be used for each computational node with binary cross entropy, such as using the Adam optimization algorithm, to predict the probability of the source of the cancer. The source of the cancer can be used to determine the type of cancer, such as lung cancer versus breast cancer. The Adam algorithm uses an adaptive learning rate method to obtain the learning rate of each parameter to optimize the ANN. Alternatively, the ANN can also be used, for example, as a classifier to obtain cancer prognosis trajectory information. For example, the ANN can be implemented using python's keras, numpy, or sklearn functions.

[0226] It is worth noting that method 300 can be performed for each biomarker used to perform cancer assessment, which can improve the accuracy of cancer assessment. Each biomarker has its own set of classification models. For example, if two biomarkers are used, such as a first biomarker and a second biomarker, there will be a first group of classifiers, including a first group having a first cancer type classification model, a first cancer stage classification model, a first metastasis classification model, and a first metastasis potential classification model, and a second group having a second cancer type classification model, a second cancer stage classification model, a second metastasis classification model, and a second metastasis potential classification model, wherein the models in the first group all correspond to the first biomarker used to provide the first group of test results, and the models in the second group all correspond to the second biomarker used to provide the second group of test results. Then, in operation 316, the first and second test results are combined to provide a more accurate cancer assessment. In at least one embodiment, any combination of features of cell-free DNA, immune biomarkers, exosomes, and methylation markers can be used as a training data set. Compared with the use of a single biomarker, the combination of biomarkers improves the accuracy of the detection process. The first and second biomarkers of the classification model can be tumor-associated immune cells, cell-free DNA, immune biomarkers, exosomes, or methylation markers.

[0227] Thus, in at least one embodiment, the cancer assessment in operation 316 can be comprehensive, depending on the type of biomarker used to perform method 300. For example, in at least one embodiment, the cancer assessment can include: cancer location, cancer stage, cancer grade, cancer characteristics (such as but not limited to metastatic potential), treatment efficacy, and the development of adoptive immunity to cancer. Any of the biomarkers described herein can be used to detect these cancer characteristics, although the specificity and accuracy may vary depending on the specific biomarker used to detect a given cancer characteristic. In at least one embodiment, the comprehensive cancer analysis can include the prognosis of a cancer patient.

[0228] Reference now Fig.50E , which shows a flow chart of an example embodiment of a method 400 for performing training to obtain a model that can be used by software of computing device 202 when detecting one or more cancer features from a patient sample. For example, method 400 can be used to obtain an optimal model for each classifier used in operations 308-314 of cancer assessment method 300. In alternative embodiments, some aspects of method 400 can be modified.

[0229] Operations 402 and 404 are performed in a similar manner to operations 302 and 304 of method 300. Operation 406 involves using either PC analysis or MCR analysis, which can be performed in the manner described previously. However, based on the type of model being generated / trained, only one of the PC or MCR analysis is performed. This can be determined based on the nature of the potential biomarker for which the model is targeted.

[0230] It is worth noting that operations 402-406 are performed on several samples to obtain training data, which is then used to generate / train a classification model in operation 408. For example, ten or more known cancer samples may be used for training.

[0231] Then, operation 408 involves using the training data from operations 402-406. For each sample, there are several features for training, depending on the model structure. For example, when the model is based on PLSDA, the features can be provided by MCR analysis, in which case the first 5 MCR components can be used.

[0232] In operation 410, the generated / trained model can be evaluated for accuracy, which can be done by preprocessing and feature extracting Raman spectra of known cancer samples to run multiple classifiers in series. The classifier that gives the highest classification probability is used to determine the cancer type of the Raman spectrum. Raman spectra associated with this cancer type can be removed and the analysis repeated to classify the remaining cancer types.

[0233] In operation 412, if multiple classification models have been generated / trained, the most accurate model is selected as the best model. For example, feature extraction can be performed by performing 2,000 iterations and using the feature values ​​as a training data set to further optimize the model. Alternatively, at least 10 spectra can be used in the training data for statistical purposes to ensure repeatability. Each spectrum is collected three times and averaged. Example

[0234] Example 1 - T cells for cancer diagnosis

[0235] The immune system is mainly composed of immune cells in the peripheral circulation (such as lymphocytes and monocytes), among which T cells and B cells play a key role in the immune response to the disease (Ichimura, T et al., 2016). Among lymphocytes, T cells are the most numerous. Zhao, W et al. (2021) performed single-cell RNA sequencing on 8397 cells from biopsies of glioblastoma patients and generated a sequencing library (Zhao, W et al., 2021). Oncogenes were collected from the Cancer Gene Census (CGC), OncoKB, Cancer Gene Network (NCG), TSGene, IntOGene, and Volkan Okur and Wendy K. Chung, 2017. T cells showed elevated expression of oncogenes, which confirmed the presence of cancer-related features, making T cells a potential diagnostic element.

[0236] T cells play a crucial role in regulating the cancer microenvironment 1 There are differences in T cell phenotypes between cancer and non-cancer states. For example, the differentiation state is critical for T cell exhaustion, anti-tumor immunity, and drug intervention. 2-3 Strategies have been developed to produce less differentiated T cells. The morphological characteristics of individual T cells can also accurately distinguish cancer cells from non-cancerous cells. Therefore, understanding T cell phenotypes is crucial to distinguish cancer and non-cancerous microenvironments.

[0237] The immune system mainly determines the growth, progression and metastasis of tumors. Therefore, the presence of immune cell infiltration within the tumor microenvironment largely determines the patient's prognosis (Varn, FS et al., 2018). In addition, research results show that immune cells can have pro-tumor and anti-tumor effects based on the tumor stage (Wellenstein, MD et al., 2018). Different cancer types are associated with characteristic immune activation states in the tumor microenvironment. For example, studies have shown that a high abundance of intratumoral CD8 T cells in breast cancer (Savas P et al., 2016) is associated with a high survival rate of patients.

[0238] Similarly, the presence of CD4 T cells is associated with poor outcomes in bladder cancer and good outcomes in lung cancer (Gentles, AJ et al., 2015). These clinical insights suggest that changes in the immune profile are phenotypically and functionally distinct and can be used to characterize cancers based on molecular subtypes. Figure 1 Schematic diagram of the experimental design for obtaining immune cell signatures (a) and the clinical workflow (b) are shown. Figure 2 The feasibility of tumor-associated immune cell-based signatures in comprehensive cancer assessment is shown (a), and further shows that tumor-associated immune profiles show significant overlap between sexes (b) and do not change with age (c). Therefore, different immune profile changes can be used for cancer diagnosis.

[0239] Figure 3 The t-distributed stochastic neighbor embedding (tSNE) analysis observed in a) shows the changes in T cell phenotypes between cancer and non-cancer, indicating that by utilizing tSNE, the Raman signatures of tumor-associated T cells are significantly different between non-cancer and cancer. Tumor-associated T cells are used for comprehensive cancer assessment, including cancer diagnosis, early cancer detection, cancer tissue origin identification, and cancer metastasis detection. Figure 3 As can be observed in c), the characteristics of tumor-associated T cells also differ between early-stage cancer and late-stage cancer. Phenotypes specific to the cancer-derived tissue were also observed. Similarly, Figure 3 , The heat map in d) shows that most of the differences between cancer-associated T cells and non-cancer-associated T cells are contributed by PC1 corresponding to the peak position in Table 1. This unique phenotype can be used to determine the tissue of origin of cancer and the ability to diagnose a variety of cancers at an early stage.

[0240] Table 1. Peak positions corresponding to PC1 contributing to the differences between cancer-associated T cells and non-cancer-associated T cells.

[0241] PC1-peak distribute 676.22 DNA(G) 854.81 Glycogen 953.12 Proline / Valine 1129.6 Lipid acyl backbone 1339.5 Tryptophan 1419.8 <![CDATA[CH2 shear vibration (lipid band)]]>

[0242] like Figure 4 As shown, these unique differences between tumor-associated T cells and non-cancer T cells have been developed for cancer diagnostic testing. Figure 4 , a) shows a scatter plot showing the probability of classifying a sample as cancer using this unique tumor-associated T cell signature. Figure 4 ,It can be observed in a), that the classification is accurate, only one sample is misclassified as cancer, and no cancer samples are misclassified as non-cancer. Figure 4, b) Calibration curve showing the performance of the machine learning model for cancer diagnosis using the unique features of tumor-associated T cells identified above. Based on the machine learning model, 100% sensitivity and specificity were achieved. Next, the principal components that contributed to accurate cancer diagnosis were analyzed. Figure 4 , c) It can be observed that the main principal component PC1 shows the greatest contribution in assessing the probability of a sample being diagnosed as cancer. In addition, in order to evaluate the performance of the machine learning model, the trade-off between sensitivity and specificity, the so-called receiver operating characteristic (ROC) curve, is used. Figure 4 , d), it can be observed that the area under the curve is 1.00, confirming that accurate cancer classification can be performed using tumor-associated T cell signatures.

[0243] Next, the unique tumor-associated T cell signature was used to evaluate the suitability of this signature for early cancer diagnosis, such as Figure 3 Early diagnosis of cancer is crucial as it increases the chances of successful treatment and survival 4 . In addition, delayed or inaccessible access to cancer care can reduce the chance of survival, make treatment-related problems more severe, and increase the cost of care. Further research has shown that T cells play a vital role in fighting cancer both directly and indirectly. Killer T cells can directly kill cancer cells, while helper T cells organize and coordinate the fight against cancer. 5 Therefore, defining the precise phenotype and function of tumor-specific T cell populations is a central goal in tumor immunology, with important implications for the diagnosis, patient prognosis, and treatment of both early-stage and metastatic cancers. 6 In addition, recent studies have shown that genetic variants in cancer immune response pathways can influence outcomes. They could serve as predictors of treatment outcomes in patients with early-stage lung cancer. 7 .

[0244] Figure 4 , e) shows a scatter plot showing the probability of classifying a sample as early-stage cancer using the tumor-associated T cell signature. It can be observed that the tumor-associated T cell signature can accurately classify early-stage cancer and can distinguish between early-stage and non-cancer samples with 100% sensitivity and specificity ( Figure 4 , f). When analyzing the principal components that contribute to the accurate diagnosis of early cancer, PC1 and PC2 are the main factors for early cancer diagnosis ( Figure 4 , g). Next, the accuracy of the model was evaluated using the ROC curve, where the observed area under the curve was 1.00 ( Figure 4 , h).

[0245] Figure 5, a) shows a scatter plot of the probability of classifying samples based on cancer stage using the tumor-associated T cell signature. It can be observed that the tumor-associated T cell signature can accurately classify cancer stage and can distinguish between early and late stage samples with 100% sensitivity and specificity ( Figure 5 , b). When analyzing the principal components that contribute to the accurate diagnosis of early-stage cancer and late-stage cancer, PC4 and PC5 were identified as the main factors for the diagnosis of early-stage cancer and late-stage cancer ( Figure 5 , c). Next, the accuracy of the model was analyzed using the ROC curve, where the area under the curve was 1.00 ( Figure 5 , d). The prominent peak in PC4 and PC5 is 1002 cm -1 、1090cm -1 、1324cm -1 、1449cm -1 and 1659cm -1 , corresponding to amino acids from proteins, nucleic acids, lipids, and RNA, respectively.

[0246] In addition, unique tumor-associated T cell signatures can be used to distinguish primary cancers from metastatic cancers. Figure 5 , e) shows a scatter plot indicating the probability of a sample being classified as metastatic / potentially metastatic. It can be observed that the tumor-associated T cell signature can accurately classify metastases and distinguish metastatic samples from primary samples with a sensitivity of 96.2% and a specificity of 100% ( Figure 5 , f). When analyzing the principal components that contribute to the accurate diagnosis of metastatic cancer versus primary cancer, PC1 and PC2 were identified as the main factors ( Figure 5 , g). Next, the accuracy of the model was analyzed using the ROC curve, where the area under the curve was 1.00 ( Figure 5 , h).

[0247] In summary, the features from the correlated T cells formed clusters that were separate from the healthy T cells. The peak differences were mainly due to the displacement of protein, DNA, lipid, and RNA peaks. The spectral data were then analyzed by performing t-distributed stochastic neighbor embedding (tSNE) classification. The T cells formed two distinct clusters ( Figure 3 , a), which means that they exhibit different phenotypes and cellular compositions when derived from cancer and non-cancerous samples. Figure 3 The violin plots in b) show the significant differences between the spectral data of early cancer and non-cancer associated T cells.

[0248] Similarly, Figure 2, a) shows a scatter plot of tSNE analysis of SERS data of cancer-related immune cells and buffy coat from patient blood. The clustering is uniform and broad, indicating that immune cells express Figure 3 , Diagnostic features of T cells in a). Therefore, circulating immune cells in patients show similarities, indicating that circulating immune cells reflect the entire spectrum of cancer features and can be used for effective cancer diagnosis through liquid biopsy.

[0249] method

[0250] T cells were isolated from patient tumors by mincing the tumor and digesting the tissue to release tumor-infiltrating T cells according to the following protocol. Surgically resected tumors were minced with a sterile No. 10 blade and enzymatically digested with collagenase and hyaluronidase at 37°C. The resulting suspension was filtered repeatedly through a 100 μm filter. The suspension was then treated with RBC lysis buffer, incubated at 25°C for 5 minutes, then centrifuged and washed with 1X PBS. T cells were isolated from the single-cell suspension by magnetic separation.

[0251] Example 2 - Tumor-associated NK cells for cancer diagnosis

[0252] method

[0253] Clinical sample collection

[0254] This study was conducted in accordance with the Ryerson University Ethics Committee (REB 2020-275). Informed consent and blood samples from cancer patients were obtained from the Ontario Cancer Board (OTB). Buffy coats were extracted by density gradient centrifugation.

[0255] Cell culture

[0256] Primary NK cells (stem cell technology) and NK-92 cell lines (derived from non-Hodgkin's lymphoma) were obtained from the American Type Culture Collection (ATCC, USA) and maintained in Alpha's minimal essential medium containing 0.2mM inositol, 0.1mM 2-mercaptoethanol, 0.02mM folic acid, 12.5% ​​horse serum and 12.5% ​​fetal bovine serum without nucleosides. In order to maintain adequate proliferation, IL-2 was added at a concentration of 150IU / ml. Similarly, cancer cell lines of breast cancer (MDA-MB231), non-small cell lung cancer (H69 AR) and colon adenocarcinoma (colo 205) were obtained from ATCC. Breast cancer cell lines were maintained in DMEM containing 10% FBS. Lung cancer and colon cancer cell lines were maintained in RPMI 1640 containing 10% FBS. All cells were cultured at 37°C, 5% CO2 environment.

[0257] Co-culture assay

[0258] NK-92 cells were co-cultured with three cancer cell lines using a Transwell apparatus (0.4 μm pore size, Corning, Lowell, MA). NK-92 cells were seeded in the upper insert of a 24-well Transwell plate. Cancer cell lines were seeded in three separate lower chambers. Prior to the experiment, cancer stem cells (CSCs) were cultured separately in serum-free medium using ultra-low low attachment plates. Tumor spheroids were verified under a microscope and seeded into a set of other low chambers. After 24 hours of co-culture, NK-92 cells were centrifuged and resuspended in water for SERS analysis. NK cells grown alone without co-culture with cancer cell lines were used as a control.

[0259] RT-qPCR

[0260] RNA was extracted from cell pellets by Trizol / chloroform extraction (Invitrogen) and resuspended in diethylpyrocarbonate (DEPC)-treated water. RNA concentration and ratio were determined photometrically (Nanodrop). BCL-2 expression levels were assessed by RT-qPCR. RT-qPCR reactions were performed in 96-well plates (Micro FastOptical 96-well reaction plates; ABI, Foster City, CA, USA). The relative expression levels of target genes are expressed as fold changes relative to the housekeeping gene GAPDH. Statistical analysis (one-way ANOVA) was performed using Prism Graphpad (v.9) to compare gene expression levels.

[0261] Raman data acquisition

[0262] 5 μl of buffy coat (blood) and 10 μl of cultured NK-92 cells were dropped onto the tumor immune probe platform. After 1 minute (allowing the NK-92 cells to be captured in the sensor grid), Raman spectroscopy was performed at a wavelength of 785 nm. At least 10 spectra were collected using 5W laser power and 3 acquisitions were performed. The spectra were saved and processed collectively using Spectragryph software.

[0263] Data analysis

[0264] Partial least squares regression analysis was performed after multivariate analysis. NK-92 cell culture data were used as training data, and spectra from buffy coats were used as validation data. PLSDA analysis was performed using PLS Toolbox software. For cancer localization, buffy coat data from known patient samples were used to train the algorithm. Statistical analysis of all results is shown in the legends. All data are expressed as mean ± SD. For SERS analysis, a two-tailed Student's t-test was performed. P < 0.05 was considered statistically significant. Experimental results were performed at least three times unless otherwise stated in the legends.

[0265] result

[0266] This study shows that molecular probing of natural killer (NK) cells has the potential to provide diagnostic information for cancer patients. The presence of cancer stem cells (CSCs) can be detected by changes in NK cell expression, as CSCs are resistant to antiproliferative therapy and have the ability to repopulate bulk tumors (Ames E, et al., 2015). Detecting the presence of CSCs is not only crucial for cancer diagnosis, but also for patient care management. To detect the presence of CSCs, NK cells were chosen for a number of reasons. NK cells are an important component of the innate immune system and are the first line of defense against cancer, responsible for cancer immune surveillance (Myers JA and Miller JS, 2021). In addition, NK cells do not require any prior sensitization to identify tumors (Tallerico R et al., 2013). In addition, among all immune cells, only NK cells show preferential cytotoxicity against CSCs (Butler KT, et al., 2018). Although CSCs are able to evade attack by other immune cells, CSCs cannot escape NK cell surveillance and show vulnerability to NK cells (Luna JI, et al., 2017). Therefore, the presence of CSCs may naturally activate NK cells with characteristic molecular changes, thereby being able to identify the presence of CSCs and, in turn, the presence of cancer. Figure 6This diagnostic approach is illustrated. NK cells were co-cultured with cancer cells as well as CSCs. This resulted in NK cells exhibiting three phenotypes based on cell-specific associations. Consistent with this idea, initial NK cell spectra, cancer-associated NK cell spectra, and CSC-associated NK cell spectra were obtained from cell cultures. These three phenotypes formed the basis for distinguishing cancer diagnosis in this study. Using a simple machine learning algorithm, the SERS spectra of human blood samples were analyzed based on the similarity with the active SERS spectra of NK cells. It is expected that Raman signals from the interaction of NK cells with cancer cells and CSCs can be detected from the patient's blood. Therefore, NK cells were first co-cultured with cancer cells, CSCs, and non-cancer cells, and the SERS signals were collected using a SERS-functionalized tumor immune probe platform.

[0267] In this study, machine learning (ML), a subfield of artificial intelligence that has developed rapidly in recent years, was used for diagnostic prediction. Unlike traditional techniques, ML techniques are able to solve complex problems involving a large number of combinatorial spaces or nonlinear processes without incurring a lot of computational costs (Guo S., et al., 2021). By adopting ML methods for cancer diagnosis, the use of ML was explored to solve the complex molecular fingerprints of tumor-associated NK cells for predicting cancer. ML was chosen because ML tools have been continuously used to generate, test, and improve scientific models (Meza Ramirez CA, et al., 2021; Sheetz KE and Squier J, 2009; Wong DM, et al., 2010). This series of statistically based methods can predict the properties of molecules and materials without the need for computationally intensive electronic structure calculations, and has the potential to accelerate various applications in chemistry and molecular sciences, including Raman spectroscopy. The spectral dataset from co-culture was used to train a machine learning model with binary classification (cancer and non-cancer). The supervised model successfully classified the co-culture data into two clusters. Blood samples from cancer patients and healthy controls were also classified by the model.

[0268] Figure 7 , a) shows the distinction between normal NK cells and tumor-associated NK cells based on tSNE analysis. PCA analysis shows that the Raman signature of tumor-associated NK cells shows characteristic bands of carbohydrates, proteins, and lipids. The obvious peak is 1003 cm -1 Phenylalanine band at 1450cm -1 CH deformation at 1661cm -1 However, the most obvious band typical of lymphocytes appears at 1522 cm -1 and 1158cm -1The peaks in the spectral region are carotenoids (El-Said WA, et al., 2011). Carotenoids are strong Raman scatterers with characteristic Raman spectra. The presence of carotenoids in NK cells indicates cytotoxicity and cell surface activation. Other spectral peaks originate from enzymes in dense granules in the cytoplasm (Shen Y, et al., 2021).

[0269] Ultrasensitive detection of CNKP using tumor immunosensor-assisted

[0270] Next, the ability of the tumor immunosensor to generate a profile of circulating NK cells was analyzed. Surface-enhanced Raman scattering (SERS) was used for ultrasensitive detection. SERS-based methods are well suited for monitoring intracellular proteins and other macromolecules such as lipids and nucleic acids, enabling in-depth analysis of cellular biochemical composition (Movasaghi Z et al., 2007). SERS-based molecular fingerprints at the subcellular level will provide real-time information in a non-destructive manner (López-Soto A, et al., 2017). This label-free approach is very useful for analyzing tiny intracellular changes (Xu Y, et al., 2020).

[0271] NK cells that are completely free of any tumor cell association show a characteristic Raman spectrum with characteristic bands from carbohydrates, proteins, and lipids. This spectrum can be associated with circulating NK cells. The CH deformation band is located at 1450 cm -1 , the amide I band is located at 1662 cm -1 , the amide II band is located at 1555 cm -1 , the amide III band is located at 1339 cm -1 Other spectral features include 500-550cm -1 disulfide bonds (SS) between the -1 (from phenylalanine), 830cm -1 and 854cm -1 Aromatic amino acids at 1340 cm (from the tyrosine doublet) -1 Tryptophan in the intestine (Jiang S, et al., 1996-1997).

[0272] Evaluating the role of cancer stem cell-associated NK cells in cancer diagnosis

[0273] The main changes in NK-CSC metabolites included a significant decrease in disulfide-extended proteins, a relative increase in the number of nucleic acids, a decrease in tyrosine and phenylalanine, an increase in glycogen and fatty acids and lipids, an increase in carotene, phospholipids, and cytosine, and a decrease in tryptophan, corresponding to 521 cm -1 、787cm-1 、854cm -1 , 1000cm -1 、1048cm -1 、1137cm -1 、1168cm -1 、1268cm -1 、1509cm -1 、1339cm -1 NK cells associated with cancer stem cells at 520 cm -1 The peak at 1522cm decreased, indicating that NK-CSCs have downregulated their killer immunoglobulin receptor (KIR) expression, suggesting that CSCs may have decreased major histocompatibility complex (MHC) expression. This paves the way for increased cytotoxicity. Loss or reduction of expression in CSCs is associated with better clinical outcomes and promising strategies to reverse immune escape (Kornberg MD, 2020). The most prominent band of typical lymphocytes can be seen at 1522cm -1 1158cm -1 , belonging to carotenoids (El-Said WA, et al., 2011). As mentioned above, carotenoids are strong Raman scatterers with characteristic Raman spectra. The significant changes in the CSC-associated NK cell profile indicate that CSC-associated NK cells are suitable for further analysis. A notable feature of this association is that, unlike cancer cells, CSCs cannot escape immune detection by NK cells, and their association easily reflects the tumor profile. Many studies have shown that NK cells have a unique affinity for CSCs, where NK cells lyse and kill CSC populations in mice. The observed spectrum shows the state of immune cell stimulation / inhibition of the cells. Several markers indicative of lymphocyte activation have been identified. 521cm -1 The peak at is a disulfide bond band, indicating immunoglobulin formation (Schafer JR, et al., 2019). Disulfide bonds are a feature in Raman spectra that appear distinct from other peaks and contribute to the conformation of proteins. NK cells associated with CSCs exhibit killer cell immunoglobulin-like receptors (KIRs) on their surface, which are involved in the education of NK cells.

[0274] These unique differences between tumor-associated NK cells and non-cancer NK cells are used in cancer diagnosis, such as Figure 8 shown. Figure 8 , a) shows a scatter plot demonstrating the probability of classifying a sample as cancer using this unique tumor-associated NK cell signature. Figure 8, b) shows the calibration curve of the performance of the machine learning model for cancer diagnosis using the unique features of tumor-associated NK cells. Based on the machine learning model, 100% sensitivity and specificity were achieved. Next, the principal components contributing to accurate cancer diagnosis were analyzed. Figure 8 , c) It can be observed that the main principal component PC6 contributes the most in evaluating the probability of a sample being diagnosed as cancer. In addition, in order to evaluate the performance of the machine learning model, the ROC curve is also evaluated. Figure 8 , d) It can be observed that the area under the curve is 1.00, confirming that accurate cancer classification can be performed using tumor-associated NK cell signatures.

[0275] Next, the unique tumor-associated NK cell signature was used to evaluate the suitability of this signature for early cancer diagnosis. Studies have shown that NK cells have the potential to detect early cancer. In preclinical studies, NK cells have been shown to indicate survival and treatment response in different types of cancer using different surface and functional markers detected by immunohistochemistry, immunofluorescence or flow cytometry. 8 Molecular detection of NK cells has the potential to provide diagnostic information for cancer patients, and the diagnostic value of NK cell activity is significant. Therefore, it can be concluded that the specific contribution of NK cells to early cancer is observed in the Raman spectral features of early and late cancers, such as Figure 7 , as shown in c).

[0276] Figure 8 , e) shows a scatter plot showing the probability of classifying a sample as early-stage cancer using the tumor-associated NK cell signature. It can be observed that the tumor-associated NK cell signature provides accurate classification of early-stage cancer and can distinguish between early-stage and non-cancer samples with a sensitivity and specificity of 100% ( Figure 8 , f). When analyzing the principal components that contribute to accurate diagnosis of early cancer, it was observed that PC2 and PC5 are the main factors for early cancer diagnosis ( Figure 8 , g). Next, the accuracy of the model was analyzed using the ROC curve, where the area under the curve was 1.00 ( Figure 8 , h).

[0277] Fig. 9 , a) shows a scatter plot showing the probability of classifying samples according to cancer stage using the tumor-associated NK cell signature. It can be observed that the tumor-associated NK cell signature provides accurate classification of cancer stage and can distinguish between early and late stage samples with 100% sensitivity and specificity ( Fig. 9 , b). When analyzing the principal components that contribute to accurate diagnosis of early cancer, it was observed that PC4 and PC2 were the main factors for early cancer diagnosis ( Fig. 9, c). Next, the accuracy of the model was analyzed using the ROC curve, where the area under the curve was 1.00 ( Fig. 9 , d).

[0278] NK cells may be used to detect cancer metastasis. NK cells play a key role in controlling metastasis, and they have been shown to preferentially control monoclonal metastases originating from single circulating tumor cells rather than polyclonal metastases originating from cell clusters. 14 Therefore, unique tumor-associated NK cell signatures are used to detect cancer metastasis. Fig. 9 , e) shows a scatter plot showing the probability of determining a sample as a metastasis using the tumor-associated NK cell signature. It can be observed that the tumor-associated NK cell signature provides accurate classification of cancer metastasis and can distinguish primary samples from metastatic samples with 100% sensitivity and specificity ( Fig. 9 , f). When analyzing the principal components that contribute to accurate diagnosis of early-stage cancer, PC4 and PC2 were observed to be the main factors for detecting metastasis ( Fig. 9 , g). Next, the accuracy of the model was analyzed using the ROC curve, where the area under the curve was 1.00 ( Fig. 9 , h), thus indicating that NK cells can be used to detect cancer metastasis.

[0279] Example 3 - Tumor-associated macrophages for cancer diagnosis

[0280] Macrophages play a crucial role in the anti-tumor activity of chemotherapy, radiotherapy and monoclonal antibodies 15 Tumor-associated macrophages (TAMs) play crucial roles in tumor development, regulation of neoangiogenesis, immunosuppression, and metastasis 16 The high proportion of macrophages in cancer is thought to be a mechanism involved in anticancer surveillance 17 Macrophages can promote tumor death by promoting cytotoxicity, and targeting macrophages in cancer is a potential therapeutic approach 18 Since macrophages are involved in various aspects of cancer development and progression, it is crucial to develop cancer diagnostic and therapeutic strategies targeting macrophages. 19 .

[0281] Fig.10 , tSNE analysis observed in a) shows the changes in macrophage phenotypes between cancer and non-cancer, indicating that the Raman signatures of tumor-associated macrophages are significantly different between non-cancer and cancer. By taking advantage of this significant difference, tumor-associated macrophage signatures can be used for comprehensive cancer assessment, including cancer diagnosis, identification of early cancer and cancer-origin tissues, and cancer metastasis detection. Fig.10, c), it can be observed that the characteristics of tumor-associated macrophages also differ between early-stage cancer and late-stage cancer. Phenotypes specific to the cancer-derived tissue were also observed. Similarly, Fig.10 , The heat map in d) shows that most of the differences between cancer-associated macrophages and non-cancer-associated macrophages are contributed by PC1 corresponding to the peak positions shown in Table 2.

[0282] Table 2. Peak Raman shift positions that differ between cancer-associated macrophages and non-cancer-associated macrophages.

[0283]

[0284]

[0285] These unique differences between tumor-associated macrophages and non-cancer macrophages can be used for cancer diagnosis, e.g. Fig.11 shown. Fig.11 , a) shows a scatter plot demonstrating the probability of classifying a sample as cancer using this unique tumor-associated macrophage signature. Fig.11 , b) shows the calibration curve of the machine learning model performance for cancer diagnosis using the unique features of tumor-associated macrophages. Based on the machine learning model, 100% sensitivity and specificity were achieved ( Fig.11 , b). Next, the principal components contributing to accurate cancer diagnosis were analyzed. Fig.11 , c) It can be observed that the main principal component PC1 contributes the most in evaluating the probability of a sample being diagnosed as cancer. In addition, in order to evaluate the performance of the machine learning model, the ROC curve was analyzed. Fig.11 , d) It can be observed that the area under the curve is 1.00, confirming that accurate cancer classification can be performed using the tumor-associated macrophage cell signature.

[0286] Next, the unique tumor-associated macrophage cell signature was used to evaluate the applicability of this signature for early cancer diagnosis, e.g. Fig.10 , as originally shown in c). Macrophages play multiple roles in cancer development, ranging from anti-tumor activity during early stages of progression to tumor promotion in established cancers. 16 Tumor-associated macrophages (TAMs) are one of the important cellular components of the tumor microenvironment and are crucial for tumor growth. Depending on their activation state, macrophages can have a dual impact on tumorigenesis by either promoting or inhibiting cancer development. 20 In the early stages of cancer, macrophages are thought to participate in anticancer surveillance and have antitumor activity. 17 However, some studies have shown that macrophages may play a pro-tumor role in tumorigenesis and tumor progression by promoting tumor growth and metastasis. 20Higher density of M2 macrophages is associated with high tumor cell proliferation, vascularization, immunosuppression, drug resistance, induced histological malignancy, and poor clinical prognosis 21 Thus, macrophages have a complex role in cancer development and their potential use in cancer diagnostics remains an active area of ​​research.

[0287] Fig.11 , e) shows a scatter plot showing the probability of classifying a sample as early-stage cancer using the tumor-associated macrophage signature. It can be observed that the tumor-associated macrophage signature can accurately classify early-stage cancer and can distinguish between early-stage and non-cancer samples with a sensitivity of 98.5% and a specificity of 100% ( Fig.11 , f). When analyzing the principal components that contribute to accurate diagnosis of early cancer, PC1 and PC5 were identified as the main factors for early cancer diagnosis ( Fig.11 , g). Next, the accuracy of the model was analyzed using the ROC curve, where the area under the curve was 1.00 ( Fig.11 , h).

[0288] Fig.12 , a) shows a scatter plot showing the probability of classifying samples based on cancer stage using the tumor-associated macrophage signature. It can be observed that the tumor-associated macrophage signature provides accurate classification of cancer stage and can distinguish between early and late stage samples with a sensitivity and specificity of 100% ( Fig.12 , b). When analyzing the principal components affecting the accuracy of early cancer diagnosis, PC1 and PC9 were identified as the main factors for early cancer diagnosis ( Fig.12 , c). Next, the accuracy of the model was analyzed using the ROC curve, where the area under the curve was 1.00 ( Fig.12 , d).

[0289] Macrophages may be used to detect cancer metastasis. Tumor-associated macrophages (TAMs) are one of the cellular components of the tumor microenvironment and play a key role in tumor development and metastasis. 18 Macrophages can also induce invasion and metastasis, making them a potential target for cancer immunotherapy 24 . Fig.12 , e) shows a scatter plot showing the probability of a sample being metastatic using the tumor-associated macrophage cell signature. It can be observed that the tumor-associated macrophage signature provides accurate classification of cancer metastasis and can distinguish between primary and metastatic samples with 100% sensitivity and specificity ( Fig.12 , f). When analyzing the principal components that contribute to accurate diagnosis of early-stage cancer, PC5 and PC1 were identified as the main factors for detecting metastasis ( Fig.12 , g). Next, the accuracy of the model was analyzed using the ROC curve, where the area under the curve was 1.00 ( Fig.12 , g).

[0290] Example 4 - Tumor-associated dendritic cells for cancer diagnosis

[0291] Dendritic cells (DCs) are a diverse group of specialized antigen-presenting cells that play a key role in initiating and regulating innate and adaptive immune responses. 25 DCs are specialized in antigen recognition and presentation and play a central role in initiating antigen-specific immunity and tolerance. 26 Despite their great potential in promoting anti-tumor responses, tumor-associated dendritic cells are largely defective in their functional activity and may contribute to immunosuppression in cancer. 27 The powerful ability of DCs to initiate and regulate adaptive immune responses is the basis for successful anti-tumor immune responses. 26,28 DCs are the most efficient, powerful, and professional antigen-presenting cells in the immune system, inducing and dispersing primary immune responses by activating T cells. 29 Therefore, the unique characteristics of tumor-associated dendritic cells were discovered using SERS technology and applied to cancer diagnosis.

[0292] Fig.13 , the tSNE analysis observed in a) shows the changes in the dendritic cell phenotype between cancer and non-cancer, indicating that the Raman signatures of tumor-associated dendritic cells are significantly different between non-cancer and cancer. By taking advantage of this significant difference, tumor-associated dendritic cells are used for comprehensive cancer assessment, including cancer diagnosis, detection of early cancer and cancer-derived tissue, and cancer metastasis detection. Fig.13 , c), it can be observed that the characteristics of tumor-associated dendritic cells also differ between early-stage cancer and late-stage cancer. Phenotypes specific to the cancer-derived tissue were also observed. Similarly, Fig.13 , The heat map in d) shows that most of the differences between cancer-associated DCs and non-cancer-associated DCs are contributed by PC1 corresponding to the peak positions shown in Table 3.

[0293] Table 3. Peak Raman shift positions that differ between cancer-associated dendritic cells and non-cancer-associated dendritic cells.

[0294]

[0295] The unique differences between these tumor-associated DCs and non-cancerous DCs were tested for cancer diagnostics, e.g. Fig.14 shown. Fig.14, a) shows a scatter plot showing the probability of classifying a sample as cancer using this unique tumor-associated dendritic cell signature. Fig.14 ,a) It can be observed that the classification is accurate. Fig.14 , b) shows the calibration curve of the performance of the machine learning model for cancer diagnosis using the unique features of tumor-associated dendritic cells. Based on the machine learning model, 100% sensitivity and specificity were achieved. Next, the principal components contributing to accurate cancer diagnosis were analyzed. Fig.14 , c) It can be observed that the main principal component PC3 shows the greatest contribution in evaluating the probability of a sample being diagnosed as cancer. In addition, in order to evaluate the performance of the machine learning model, the ROC curve was analyzed. Fig.14 , d) It can be observed that the area under the curve is 1.00, confirming that accurate cancer classification can be performed using tumor-associated dendritic cell signatures.

[0296] Next, the unique tumor-associated DC signature was used to evaluate the applicability of this signature for early cancer diagnosis, e.g. Fig.13 , originally shown in c). The phenotype of dendritic cells (DCs) has been studied in early cancers. DCs are a diverse group of specialized antigen-presenting cells that play a key role in initiating and regulating innate and adaptive immune responses. 25,27 The phenotype of DCs in cancer is heterogeneous, and different DC subsets have been identified that drive specific types of immune responses. 28,30 .

[0297] Fig.14 , e) shows a scatter plot showing the probability of classifying a sample as early-stage cancer using the tumor-associated dendritic cell signature. It can be observed that the tumor-associated dendritic cell signature can accurately classify early-stage cancer and distinguish between early-stage and non-cancer samples with a sensitivity of 97.1% and a specificity of 96.7% ( Fig.14 , f). When analyzing the principal components that contribute to accurate diagnosis of early cancer, PC1 and PC6 were identified as the main factors for early cancer diagnosis ( Fig.14 , g). Next, the accuracy of the model was analyzed using the ROC curve, where the area under the curve was 0.998 ( Fig.14 , h).

[0298] Example 5 - Tumor-associated B cells for cancer diagnosis

[0299] B cells are a type of lymphocyte that produce antibodies to help the body fight infection and can be involved in the immune response to cancer. 31. The immune system can be divided into the innate and adaptive immune systems, both of which contribute to the recognition and elimination of foreign pathogens as well as tumors. Recent studies have shown that intratumoral B cells are present in human cancers. Studies have shown that B cells can present cognate tumor-derived antigens to T cells and that the functional consequences of this interaction are determined by the B cell phenotype. 32 Therefore, the unique characteristics of tumor-associated B cells were discovered using SERS technology and applied to cancer diagnosis.

[0300] Fig.15 , tSNE analysis observed in a) shows the changes in B cell phenotypes between cancer and non-cancer, indicating that the Raman signatures of tumor-associated B cells are significantly different between non-cancer and cancer ( Fig.15 , b). By utilizing this distinct difference, tumor-associated B cells can be used for comprehensive cancer assessment, including cancer diagnosis, detection of early-stage cancer and cancer-derived tissue, and cancer metastasis detection. Fig.15 , c), it can be observed that the characteristics of tumor-associated B cells also differ between early-stage cancers and late-stage cancers. Phenotypes specific to the cancer-derived tissue were also observed. Similarly, Fig.15 , The heat map in d) shows that most of the differences between cancer-associated B cells and non-cancer-associated B cells are contributed by PC1 corresponding to the peak positions shown in Table 4.

[0301] Table 4. Peak Raman shift positions that differ between cancer-associated B cells and non-cancer-associated B cells.

[0302] Raman shift (1 / cm) distribute 677.94 Ring breathing patterns in DNA bases 850.41 Single bond stretching vibrations of amino acids and valine 1083.4 CN stretching patterns of proteins 1158 CC / CN stretching (protein) 1420 CH2 shear vibration (lipid band) 1722.4 C=O stretching (lipid)

[0303] The unique differences between these tumor-associated B cells and non-cancer B cells can be used for cancer diagnosis, such as Fig.16 shown. Fig.16 , a) shows a scatter plot showing the probability of classifying a sample as cancer using this unique tumor-associated B cell signature. Fig.16 ,a) It can be observed that the classification is accurate. Fig.16 , b) shows the calibration curve of the performance of the machine learning model for cancer diagnosis using tumor-associated B cells. Based on the machine learning model, 100% sensitivity and specificity were achieved ( Fig.16 , b). Next, the principal components contributing to accurate cancer diagnosis were analyzed. Fig.16 , c) It can be observed that the main principal component PC4 shows the greatest contribution in evaluating the probability of a sample being diagnosed as cancer. In addition, in order to evaluate the performance of the machine learning model, the ROC curve was analyzed. Fig.16 , d) It can be observed that the area under the curve is 1.00, confirming that accurate cancer classification can be performed using tumor-associated B cell signatures.

[0304] Next, the unique tumor-associated B cell signature was used to evaluate the suitability of this signature for early cancer diagnosis, such as Fig.15 , originally shown in c). The phenotype of B cells in early lung cancer, the diverse landscape of B cells and plasma cells in the tumor, appears to influence patient prognosis and treatment response 32,35 In addition, the presence of B cells in tertiary lymphoid structures (TLS) is associated with protective immunity in lung cancer patients 36 In contrast, in some tumors, B cells in TLS form germinal centers and actively secrete antibodies that recognize tumor-associated antigens, and patients whose tumor-associated TLS contain germinal centers sometimes have a better prognosis. 31 The phenotype of B cells in different types of B-cell lymphomas can also differ depending on where the cancer cells are located. Overall, the phenotype of B cells in cancer can be affected by the stage and type of cancer.

[0305] Fig.16 , e) shows a scatter plot showing the probability of classifying a sample as early-stage cancer using the tumor-associated B cell signature. It can be observed that the tumor-associated B cell signature can accurately classify early-stage cancer and can distinguish between early-stage and non-cancer samples with 100% sensitivity and specificity ( Fig.16 , f). When analyzing the principal components contributing to accurate diagnosis of early cancer, PC1 was identified as the main factor for early cancer diagnosis ( Fig.16 , g). Next, the accuracy of the model was analyzed using the ROC curve, where the area under the curve was 1.00 ( Fig.16 , h).

[0306] Example 6 - Tumor-associated immune cells for diagnosis of various cancers and identification of cancer-derived tissues

[0307] Identifying the tissue of origin of a cancer is important because it allows for more specific treatment and improves treatment outcomes. 37 Cancers of unknown primary (CUP) origin account for approximately 3% of all cancer diagnoses, and identifying the tissue of origin of these tumors is particularly challenging but critical for effective treatment. 38,39 Understanding the normal cell hierarchy within a given tissue is an important prerequisite for identifying the cells of origin of cancer 40 Although the combination of clinical data, histomics, and immunohistochemistry is often sufficient to classify cancers according to tissue of origin, a small but not insignificant number of cases remain that are difficult to classify. 41 Overall, identifying the tissue of origin of a cancer is critical to determining the most appropriate treatment and improving patient outcomes.

[0308] In order to identify cancer tissue origin using tumor-associated immune cells, a machine learning algorithm is trained using the characteristics of tumor-associated immune cells to provide the probability that a sample is from a specific cancer tissue origin. The tumor-associated immune cells include T cells ( Fig.17 , a), macrophages ( Fig.17 , b), dendritic cells ( Fig.17 , c), NK cells ( Fig.14 , d) and B cells (data not shown). These tumor-associated immune cell signatures can distinguish cancer tissue origin with an accuracy of more than 98% ( Fig.17 , e), with a specificity of more than 99% ( Fig.17 , f), and the sensitivity exceeds 99% ( Fig.17 , g).

[0309] Example 7—Using tumor-associated immune cell signatures to diagnose early-stage tumors based on size

[0310] A key shortcoming of existing screening and early diagnosis technologies is the lack of ability to detect tumors of a few millimeters or centimeters in size at an early stage. When the tumor size is small, the immune response will be active because the tumor's defense mechanism has not yet been activated. Therefore, by utilizing the characteristics of the activated state of the immune response when the tumor is small, the method can accurately detect cancer at an early stage.

[0311] In fact, Fig.18 The ability of tumor-associated immune cell signatures to detect small tumors was demonstrated. It can be observed that the tumor-associated immune cell signature can accurately classify early-stage cancers ( Fig.18 , a), and can distinguish early-stage and non-cancer samples with a sensitivity of 100% and a specificity of 98.3% ( Fig.18 , b). When analyzing the principal components that contribute to accurate diagnosis of early cancer, PC1 was identified as the main factor for early cancer diagnosis ( Fig.18 , c). Next, the accuracy of the model was analyzed using the ROC curve, where the area under the curve was 1.00 ( Fig.18 , d). Fig.18 , e) and as shown in Table 5, the accuracy, specificity, and sensitivity of tumor tissue origin classification of small tumors were 100%.

[0312] Table 5. Accuracy of machine learning models for multiple cancer diagnosis of small tumors

[0313]

[0314] Fig.19The applicability of tumor-associated immune cell characteristics for the diagnosis of multiple cancers was demonstrated. The accuracy of tumor-associated immune cells alone in the diagnosis of multiple cancers exceeded 95% (Table 6), the specificity exceeded 98% (Table 7), and the sensitivity exceeded 96% (Table 8). By combining the immune cell characteristics of T cells, NK cells, B cells, dendritic cells, and macrophages, the performance of tumor-associated immune cells based on multiple cancer diagnostic tests reached 100% (Table 6). Fig.19 ), eliminating false negatives.

[0315] Table 6. Accuracy of using immune cell signatures for diagnosis of various cancers.

[0316]

[0317]

[0318] Table 7. Specificity of using immune cell signatures for diagnosis of various cancers.

[0319] Cancer Type T cells Macrophages NK cells Dendritic cells Breast 98.9 100 99.7 100 lung 100 99.7 98.3 99.7 Colorectal 97.3 98.8 95.7 95.8 kidney 100 100 100 100 Uterus 97.9 99.2 99.2 100 Glioblastoma 97.9 99.2 99.2 98.9 Astrocytoma 97.1 90.5 100 100 Oligodendroglioma 96.2 98.3 98.3 100 nose 100 98.9 98.9 100 Ovaries 99.3 100 100 98.5 testis 97.8 100 100 100

[0320] Table 8. Sensitivity of using immune cell signatures for diagnosis of various cancers.

[0321]

[0322]

[0323] Example 8 - Cancer Diagnosis Using Tumor-Associated Neutrophils

[0324] Materials and methods

[0325] Cell culture

[0326] Neutrophils (Stem Cell Technologies) (HL60 cells) are a human cell line derived from peripheral blood lymphocytes. Cells were cultured in RPMI 1640 + 2mM glutamine + 10-20% fetal bovine serum (FBS). Cancer cell lines of breast cancer (MDA-MB231), non-small cell lung cancer (H69 AR) and colon adenocarcinoma (colo205) were obtained from ATCC. Breast cancer cell lines were maintained in DMEM containing 10% FBS. Lung cancer and colon cancer cell lines were maintained in RPMI 1640 containing 10% FBS. All cells were cultured at 37°C, 5% CO2. Fig. 20 As shown in the schematic diagram of the experimental design in , Raman spectral signatures were obtained from healthy neutrophils and neutrophils relevant to preclinical models of cancer.

[0327] Inflammation is considered a hallmark of cancer 42The complex interaction of immune cells within the tumor microenvironment enables ovarian cancer (OC) cells to rapidly colonize and multiply throughout the organ. Neutrophils are classic inflammatory immune cells that account for 50-70% of circulating white blood cells. 43 By interacting with the tumor microenvironment, neutrophils form an important link between inflammation and cancer development 44 Currently, it is believed that tumor-associated neutrophils (TANs) exhibit anti-tumor properties in the early stages of tumorigenesis, while TANs in the late stages of tumorigenesis (when chronic inflammation has developed) exhibit tumor-promoting functions such as proliferation, invasion, and dissemination, as well as immunosuppression. 45 Similar to cancer cells that exhibit heterogeneity, neutrophil phenotype and function appear to fluctuate greatly as tumors grow, suggesting the presence of neutrophil heterogeneity. 44 In addition, neutrophils are short-lived cells with a lifespan of 24 hours; therefore, monitoring neutrophils can reveal the state of the tumor in real time. Typically, the formation of pro-tumor or anti-tumor neutrophils is driven by the transition of cells from a quiescent state to an active state. 46 Therefore, the presence of a tumor may be reflected in the biochemical and metabolic signature of activated neutrophils. In fact, as shown below, accurate cancer diagnosis may be possible by analyzing signatures of TAN functional plasticity.

[0328] The study showed that molecular profiles of ovarian tumor-associated neutrophils (TANs), which interact with ovarian cancer and ovarian cancer stem cells, provide diagnostic information for detecting different stages of cancer. Neutrophils play an important role in ovarian cancer immune surveillance 47 , and in response to signals from the OC tumor microenvironment, neutrophils are reprogrammed to promote or suppress tumors 49 . This editing of tumor cells induces metabolic changes in neutrophils, which results in altered Raman profiles. To detect these altered signals in neutrophils, an ultrasensitive sensor capable of detecting low levels of TAN in circulation is required. To this end, an in-house neutrophil analyzer was developed as a scalable liquid biopsy platform. Raman spectroscopy (SERS) was used as the detection method ( Fig. 20). SERS is a versatile spectroscopic technique characterized by its ability to detect analytes in circulation in a non-destructive and rapid manner. Two types of neutrophil Raman spectra were obtained using the SERS technique, namely, the uninteracted Raman spectra and the ovarian cancer-associated Raman spectra for the detection of ovarian cancer. These phenotypic Raman subtypes of neutrophils form the basis for the detection of ovarian cancer using the ANN algorithm. Here, neutrophils (HL-60) were co-cultured with ovarian cancer (OVCAR) for 1, 6, 12, and 24 hours to obtain the TAN spectra. Uninteracted HL-60 neutrophils were used as a control, representing healthy neutrophils. In addition, HL-60 were stimulated with TGF-b and all-trans retinoic acid, respectively, to obtain the pro-tumor and anti-tumor Raman spectra of neutrophils. These tumor spectra help distinguish between early and late stages of ovarian cancer. To validate the pro-tumor and anti-tumor spectra, the study used PDL-1 antibodies for correlation. These validated and confirmed Raman light spectra were then used as input training data for early and late stages of OC. To identify different neutrophil signatures, a machine learning classifier was used to extract tumor interaction signals from Raman data of patient blood. This was achieved by using an artificial neural network (ANN) to separate the spectra into healthy and OC-related spectra. After training, classification was successfully achieved, and the model had a sensitivity of 90% and a specificity of 100% on the test data (area under the curve (AUC) = 1.00).

[0329] In the TME of OC, a large number of immune cell infiltrations can be seen 50,51 The Cancer Genome Atlas (TCGA) of 430 ovarian cancer samples revealed the presence of neutrophils in addition to other immune cells 52 The data were obtained by CIBERSORT analysis of 22 leukocyte subsets. Fig.21 , a) shows a heat map of immune cell infiltration found in ovarian cancer, Fig.21 , b) Pie chart showing the proportion of immune cells (neutrophils found in OC). Together, these data suggest that neutrophils interact with ovarian tumor cells. Neutrophils are the most common white blood cell and are terminally differentiated effector cells. 53 Neutrophil activation of cytotoxic immunity 54 or NO production to stop the growth of primary tumors. As tumors progress, neutrophils signal through ROS, RNS, and proteases to promote tumorigenesis 55 Thus, neutrophils exhibit metabolic adaptations when they move from the circulation to intratumoral sites. 56 To obtain the unique signature of each immune cell, Raman maps were acquired using nanosensors synthesized in-house ( Fig.21, c). Immune cells were isolated from peripheral blood using a commercially available kit, and 5 μl of the isolated immune cells were dropped onto the sensor and the Raman spectrum was taken. Fig.21 , d) shows the Raman spectra of neutrophils, dendritic cells, monocytes, natural killer cells, and T cells. Changes in the intensity of proteins, lipids, and nucleic acids within immune cells were observed from the spectra. The spectral differences between the Raman spectra of different white blood cell nuclei are very subtle and difficult to detect with the naked eye. In comparison, lymphocytes (NK cells, T cells) have much larger nuclei that almost fill the entire cell, while neutrophils have smaller nuclei with only small lobes. 57-59 Therefore, there are significant differences in the nucleic acid vibrational bands between neutrophils and lymphocytes. Furthermore, because neutrophils are granulocytes, protein contributions dominate in the neutrophil spectrum, while carotenoid bands dominate in the lymphocyte spectrum. 60,61 .

[0330] In order to obtain the metabolic phenotype of tumor-associated neutrophils (TANs), HL60 (a human leukemia cell line from ATCC) was differentiated into neutrophils. These neutrophils were co-cultured with OVCAR (an ovarian cancer cell line from ATCC) to obtain ovarian tumor-associated neutrophils. Neutrophils that did not interact were used as controls. Next, surface enhanced Raman spectroscopy (SERS) was used to analyze the metabolic phenotypes present in controls and TANs. SERS is a non-destructive technique widely used to analyze biological samples. SERS identifies the biochemical composition within these cells, which can be used to identify biochemical changes in neutrophils of different phenotypes in the TME. Fig. 22 , A(i) and A(ii) show schematic diagrams of uninteracted neutrophils and ovarian tumor-associated neutrophils on the nanosensor, respectively. Fig. 22 , B(i) and B(ii) show the SERS spectra of non-interacting and ovarian tumor-associated neutrophils. In a stable state, circulating neutrophils are not stimulated by external stimuli. Therefore, these cells are in a resting state, and the neutrophils used as controls in this experiment mimic these resting neutrophils. Several prominent peaks corresponding to proteins, lipids, and nucleic acids can be seen in the resting neutrophil spectrum. The main contributors to the resting Raman spectrum come from cellular biomacromolecules, such as 672cm -1 Corresponding to the CS stretching mode of cystine and cytosine, 1003 cm -1 Corresponds to phenylalanine, 2918 cm -1 Corresponding to asymmetric stretching of lipids and proteins.

[0331] Since SERS generates high-dimensional data points, it is necessary to reduce the dimensionality of the data to extract the salient features of TANs. To this end, principal component analysis (PCA) has been shown to be a powerful dimensionality reduction technique applicable to SERS. Fig. 22 , D is a biplot showing two distinct clusters of non-interacting neutrophils and TANs. From the PCA loading plot ( Fig. 22 , E) It can be observed that TAN is characterized by a decrease in lipid contribution and an increase in nucleic acid contribution. Similar results were obtained using hierarchical clustering analysis ( Fig. 22 , F).

[0332] The ovarian cancer tumor microenvironment (TME) is a heterogeneous population that includes ovarian cancer cells, cancer stem cells, immune cells, endothelial cells, and fibroblasts ( Fig.23 , A) 62 When neutrophils infiltrate the ovarian cancer TME, the phenotype shifts depending on their interaction with cancer cells or cancer stem cells ( Fig.23 , B and C). This shift in metabolic phenotype occurs within a day, resulting in neutrophil populations with different properties. To capture the shift in metabolic phenotype, Raman spectra of neutrophils interacting with cancer and with cancer stem cells were measured at different time periods. Immediately after interaction (0 hours), the Raman spectra changed significantly, implying that neutrophils undergo significant functional plasticity in a short period of time. Since neutrophils have a short lifespan (24 hours), it is necessary to dynamically engage the TME in a relatively short period of time.

[0333] Metabolic changes in cancer and CSCs were analyzed at different time points, including 0, 6, 12, and 24 hours. Neutrophils interacting with cancer cells had an altered carbohydrate profile corresponding to 1122 cm -1 (COC) polysaccharide and 1112cm -1 These findings suggest that significant glycolytic reprogramming occurs at different time points ( Fig.23 , D). Similar reprogramming was seen in neutrophils that interacted with cancer stem cells ( Fig.23 , E). These phenotypes can be attributed to the nutrient deficiencies commonly found in the tumor microenvironment. Therefore, neutrophils must compete with cancer and cancer stem cells for glucose energy sources. In addition, lipid and fatty acid metabolism in neutrophils interacting with cancer and CSCs was analyzed. Unlike the glycolysis profile, the lipid profiles between cancer- and CSC-associated neutrophils showed drastic changes between 0-24 hours ( Fig.23, D and E). While TAN showed a gradual increase in lipids from 0 to 24 hours, the lipid profile of CSC-associated neutrophils decreased at 24 hours, corresponding to 1745 cm in normal tissues. -1 (phospholipids), 1750cm -1 Next, we analyzed the protein-related metabolic changes in cancer- and CSC-associated neutrophils ( Fig.23 D and E). It was observed that the protein content of cancer-associated neutrophils gradually increased from 0 to 24 hours, while the protein content of CSC-associated neutrophils increased significantly twice at 6 hours and 24 hours.

[0334] Since Raman spectroscopy generates multidimensional data, features can be effectively identified by PCA. Fig.23 , F shows the changes in neutrophil Raman signals upon interaction with ovarian cancer cells and CSCs by PCA. Significant differences were observed between cancer-associated neutrophils and CSCs at 0, 6, 12, and 24 h. Clear clustering was seen at all time points observed, indicating that neutrophils acquire heterogeneous phenotypes depending on the interacting cells and the duration of their interaction.

[0335] In this study, the immune checkpoint protein programmed death ligand-1 (PD-L1) was used as an indicator of cancer interaction. PD-L1 is expressed in neutrophils when they interact with tumors. 63 , so the presence of PD-L1 signaling in neutrophils should be consistent with tumor interaction ( Fig.24 , A). To this end, PD-L1 spectra were taken and corresponding peaks were identified in early and late neutrophils. Fig.24 , A can be observed, the PDI-1 peak is located at 848cm -1 、931cm -1 、1335cm -1 and 1463cm -1 These PD-L1 peaks were recorded and identified in early and late neutrophils. To stimulate early and late neutrophils, neutrophils were cultured in vitro with all-trans retinoic acid and TGFβ, respectively. 64 . SERS spectra were taken ( Fig.24 , C), and PD-L1 peaks were identified in both early and late neutrophils ( Fig.24 , B). It can be observed from the figure that late neutrophils show a more prominent PD-L1 peak compared with early neutrophils. Further univariate analysis of early and late neutrophils was performed using Student's t-test, corresponding to 546 cm -1 、677cm -1、853cm -1 、1004cm -1 、1452cm -1 、1589cm -1 and 1672cm -1 ( Fig.24 , D). All analyses showed statistically significant differences between early and late neutrophils (****P<0.0001).

[0336] like Fig.25 As shown in , B and Table 9, the ANN model has a sensitivity of 90% and a specificity of 100% for diagnosing OC. This deep learning-based diagnostic model showed a classification accuracy of 95% ( Fig.25 , D), the area under the curve is 1.000, and the precision is 1.000. As shown in the ANN scatter plot, clear clustering was observed between the healthy group and the cancer group ( Fig.25 , E). The overlap of the cancer subset in the healthy cluster indicates the misclassification of one sample, resulting in a sensitivity of 90% ( Fig.25 , B).

[0337] Table 9. Neural network model accuracy for OC detection.

[0338] Model AUC CA Sensitivity Specificity Accuracy Recall Neural Networks 1.000 0.950 0.900 1.000 1.000 0.900

[0339] To further validate the above results, we developed an ANN classification model using blood from healthy and OC patients to classify OC and healthy on both training and test datasets. This yielded 100% sensitivity and 100% sensitivity and 100% overall classification accuracy ( Fig.24 , F and Table 10). These results validate the ANN-based diagnostic model. Scatter plot ( Fig.25 , G) shows different clustering with 100% accuracy.

[0340] Table 10. Neural network model accuracy for OC detection using patient data.

[0341] Model AUC CA Sensitivity Specificity Accuracy Recall Neural Networks 1.000 1.000 1.000 1.000 1.000 1.000

[0342] After successfully predicting OC diagnosis, the diagnostic model was tested for early OC detection. To classify early and late OC, SERS data of early neutrophils and late neutrophils were used as training data. In the test dataset, blood from patients with early OC and late OC was used. Neural network scatter plot ( Fig.26, E) shows clear clustering between early and late OC. Homogeneous clusters (early and late) indicate high precision and accuracy in the classification process. There are 100 neurons in the hidden layer and a total of 200 iterations, which yields 90% sensitivity and 100% specificity ( Fig.26 , B). The overall classification accuracy peaks at 90% ( Fig.26 , D), and the area under the curve (AUC) was 1.000. These results suggest that early detection of OC can be achieved using neutrophils as a biomarker and adopting a machine learning-based diagnostic model.

[0343] Example 9 - Immune cell DNA as a diagnostic marker for cancer diagnosis

[0344] Immune cells have been shown to undergo disease-selective activation through selective promoters that are expressed primarily in diseased cells (Ronald, JA et al., 2015). Disease-selective activation leads to changes in the DNA expression profile of immune cells, including changes in DNA methylation status and changes in chemical structure. By detecting changes in circulating immune cell DNA, these inherent molecular changes can serve as surrogate markers for cancer diagnosis.

[0345] Studies have shown that after T cells interact with breast cancer stem cells, dramatic changes occur at the molecular level and surface receptor level. It has been confirmed that T cells can induce the expression of stemness genes and genes related to dedifferentiation. 65 . T cells stimulate enhanced malignancy by bringing about changes at the DNA and cellular levels in tumors, leading to excessive tumor progression. The collective molecular changes in the DNA of T cells that interact with breast cancer stem cells provide important information about the individual disease state. Here, T cell DNA was validated as a biomarker for breast cancer diagnosis by isolating T cells from healthy PBMCs (peripheral blood mononuclear cells) and co-culturing them with breast cancer stem cells (BCSCs). The T cells after interaction with BCSCs were isolated, and the DNA of these T cells was isolated for further SERS analysis. The SERS spectra of the DNA of healthy T cells and T cells interacting with BCSCs were acquired using nanosensors to study the changes in the DNA of T cells at the molecular level. The spectra of healthy T cells and T cells interacting with BCSCs are shown in Figure 2. Fig. 27 , as shown in a, the figure shows 1098cm- 1 、838cm- 1 and 1276cm- 1 The differences in the positions and intensities of the peaks at and (corresponding to DNA phosphorylation, adenine & thymine, and cytosine peaks, respectively) are all related to DNA changes.

[0346] In addition, machine learning was applied to the DNA SERS data to study the entire set of spectra, allowing the differences between healthy T cells and BCSC-associated T cells to be determined. Principal component analysis was performed, and the scatter plots were shown as Fig. 27 , as shown in b), which shows that the spectral data points of healthy T cell DNA and BCSC-associated T cell DNA are distinguishable from each other except for brief overlapping regions. The minimal overlap is attributed to common DNA features such as DNA proteins, phosphates, and nitrogenous bases. These differences are observed as an overall spectrum, contributed by all components in the genomic DNA. To further investigate the specific peaks that contribute to these changes observed in the PCA analysis, the principal component loading plots of principal components PC1, PC2, PC3, and PC4 were analyzed. The PC loading plots are shown in Fig. 27 , c). From the PC loading diagram, the peaks that contribute significantly to the change are 887, 1563, 1648, 1098, 1276, and 1376 cm -1 . 66,67 The peaks observed in the loading plot correspond to A&T, C, PO2, G, amino acid vibrations, and phosphodiester bridges, respectively.

[0347] This study emphasizes the use of liquid biopsies (peripheral blood) to detect circulating tumor-associated signals through cell-free DNA present in plasma. To verify the feasibility of using liquid biopsies to reflect tumor signals, the similarity of signals between patient tumor DNA, BCSCAT DNA, and matched patient plasma (i.e., cell-free DNA) was studied. DNA was isolated from patient tumor samples and T cells interacting with breast cancer tumors, and Raman spectra of the isolated DNA and plasma were recorded. The data were analyzed to determine the similarity scores of the three categories of samples. Fig. 27 , a) shows a schematic diagram of the studies conducted to validate biomarkers in liquid biopsies. Fig.28 , b)-d) It can be seen that serum and BCSC-related T cell DNA ( Fig.28 , b), serum and tumor DNA ( Fig.28 , c) and BCSC-associated T cell DNA and tumor DNA ( Fig.28 , d), indicating that cell-free DNA in plasma is a useful biomarker in cancer diagnosis.

[0348] The tumor microenvironment (TME) plays a key role in tumor occurrence, development, and metastasis. T cells account for a large proportion of the breast TME and play contrasting roles, such as controlling tumor growth and promoting tumor growth by exerting immunosuppressive activity. Infiltrating T cells in breast cancer stem cells interact with tumor cells and exhibit breast cancer-related characteristics. Studies have found that the phenotypes and abundance of different types of T cells are significantly different in different histological grades of breast cancer.68-70 They have demonstrated increased expression of FOXP3 and CD152, which are associated with regulatory T cell function, which is responsible for the immunosuppressive activity. Furthermore, in higher grades, increased levels of FOXP3+ Tregs are directly associated with poor prognosis and estrogen receptor-negative status, making diagnosis and treatment more challenging. 71

[0349] Since there are significant and distinct variations in the phenotype and abundance of T cells of different classes, a heterogeneous population of the primary breast cancer cell line BT 474 was used to model four sizes of tumor organoids in vitro, such as Fig.29 , as shown in a). Heterogeneous cells enriched in cancer stem cells were counted and seeded in Aggrewell plates to form tumor organoids of known sizes. The cell seeding densities were 10 million, 1 million, 10,000 and 1,000 cells per well, respectively, and were seeded in Aggrewell plates supplemented with serum-free medium to promote the formation of tumor organoids. The organoid cultures were then treated with T cells isolated from healthy PBMCs for 24 hours to enable interaction between T cells and breast cancer TME. The T cells were then isolated, and DNA was isolated from the T cells that interacted with tumor cultures of different sizes. The SERS spectra of DNA were recorded using a nanosensor, with peaks such as Fig.29 , b). The DNA optical spectrum shows peaks at 820, 1087, 1318, 1570 and 1622 cm -1 , corresponding to the peaks of citric acid, PO2-phosphodiester bond symmetric stretching, adenine ring stretching, adenine and guanine, and thymine and cytosine, respectively. 67,72,73 like Fig.29 , c), these peaks corresponding to DNA peaks were observed in T cells interacting with breast cancer tumors of different sizes.

[0350] Fig.29 , four significant peaks are shown separately in b) to show that as the size of the breast cancer tumors interacting with T cells decreases, the peak intensity also decreases. The smallest tumors still show a positive effect on T cells, which can be observed in the form of peak intensity and DNA corresponding wave number ( Fig.29, e)-i). In contrast, these peaks were not observed in the DNA of healthy or non-interacting T cells, clearly demonstrating that T cells undergo genomic-level changes when interacting with tumors, independent of cell size and number in the TME. These subtle differences between the DNA of healthy T cells and T cells interacting with tumors can be used as biomarkers for breast cancer detection. Since the smallest tumors can induce changes in T cell phenotype and genotype, this biomarker is a reliable and accurate biomarker for early diagnosis and prognosis of breast cancer.

[0351] Next, the in vitro model generated for early diagnosis of breast cancer was applied and its ability to detect the level of cancer cell differentiation in early tumors was evaluated. Fig.30 It can be observed that the immune cell DNA-based method for early diagnosis of breast cancer can achieve 93.6% sensitivity and 100% specificity ( Fig.30 , d). It can detect cancer stage with a sensitivity of 99% and a specificity of 86.8% ( Fig.30 , i). Similarly, the cell differentiation level of early breast cancer can be detected with a sensitivity of 98.2% and a specificity of 91.3% ( Fig.30 , m).

[0352] Example 10 - Immune Cell DNA for Detection of Cancer Metastasis

[0353] The tumor immune microenvironment is a key link in the development and metastasis of breast cancer. For example, studies have shown that tumor-infiltrating lymphocytes are associated with a better prognosis in breast cancer. 74 In addition, tumor-infiltrating immune cells are readily available markers for breast cancer prognosis. 75,76 Therefore, these immune cell signatures, especially the immunogenomic signatures, may provide a comprehensive and reliable method to determine the prognosis and metastasis of breast cancer.

[0354] By co-culturing naive T cells with breast cancer stem cell-enriched spheroids generated from in vitro models (MDA-MB-231, MDA-MB 361), a unique immunogenomic signature for breast cancer was developed. Breast cancer stem cell-enriched spheroids were used to explain the tumor heterogeneity observed in tumor samples. DNA isolated from T cells co-cultured with breast cancer stem cell spheroids was used to obtain Fig.31 , The immunogenomic SERS features presented in b). Fig.31 , Representative SERS spectra presented in b) show characteristic peaks associated with DNA. In addition, a peak at 831 cm was observed compared to DNA isolated from naive T cells. -1 OPO bond at 1173cm -1 Visual differences in cytosine residues at 77Due to the high dimensionality of SERS data 78 , and the algorithm can reduce the impact of noise on clustering, so the robust local linear embedding algorithm is also used for clustering analysis. Fig.31 ,c) It can be observed that the immune genomic signature associated with breast cancer forms a unique cluster that does not overlap with the immune genomic signature of naive T cells. This observation suggests that the genomic signature of immune cells changes when they encounter tumor cells.

[0355] Furthermore, PC loading analysis pinpointed specific SERS peaks that contribute to differences in breast cancer immunogenomic signatures ( Fig.31 , d). In addition, correlation analysis was performed to investigate the contribution of heterogeneity to the unique immunogenomic signature of breast cancer, which was divided into two distinct clusters, representing the presence of genotypic heterogeneity in both non-cancer and breast cancer-associated immunogenomic signatures ( Fig.31 , f). Comparison of F1 scores showed that breast cancer had a statistically significant (p<0.0001) lower score, confirming the differences between the immune genomic signatures ( Fig.31 , e).

[0356] Breast cancer is one of the most phenotypically diverse cancers, characterized by different responses to treatment, pathological features, and survival. 79 Histologically, breast cancer is characterized by the expression of surface receptors such as estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2). 80,81 Specifically, the prognosis and aggressiveness of breast cancer are highly correlated with HER2 expression 82 . In addition, HER2 expression is routinely assessed to decide on treatment options. Therefore, it was investigated whether the immunogenomic signatures of breast cancer subtypes differ. To this end, the genomic signatures of breast cancer stem cell-enriched spheres generated from in vitro models of HER2-negative breast cancer (MDA-MB-231) and HER2-positive breast cancer (MDA-MB361) co-cultured with naive T cells were analyzed ( Fig.32 , a).

[0357] Fig.32 , b) Representative SERS signatures of breast cancer subtypes, showing characteristic peaks associated with DNA molecules. Due to the high-dimensional nature of the SERS signature and its close relationship with the analyte, a multidimensional scaling algorithm was used to precisely locate the peak positions that can distinguish breast cancer subtypes ( Fig.32 , e). According to the analysis, the wavelength at 786 cm-1 that contributes to the difference in SERS characteristics is -1 、1208cm -1 、1282cm -1 and 1421cm -1The key peak positions at , can be attributed to DNA bases, the ring breathing mode of A / T DNA bases, cytosine, adenine / guanine bases, 77 In addition, the normalized SERS intensities of the peak positions mentioned above were compared ( Fig.32 , c), a statistically significant difference was observed between HER2-positive breast cancer and HER2-negative breast cancer (p<0.0001).

[0358] Furthermore, linear correlation analysis showed that the peak position was 1208 cm -1 、1282cm -1 、1421cm -1 The normalized intensity of 786cm -1 The peak at shows a negative correlation ( Fig.32 , d), consistent with the literature reports 76,83,84 .

[0359] HER2-positive breast cancer is one of the most aggressive cancers. In its early stages, it can metastasize to multiple organs, including the brain, liver, and lymph nodes, and has a poor prognosis. 87 In addition, the study pointed out that the main reason for clinical ineffectiveness in the treatment of HER2-positive breast cancer is spatial and temporal heterogeneity. 88–90 Specifically, almost 50% of patients diagnosed with HER2-positive breast cancer will develop brain metastases. 91,92 , which significantly reduces the overall survival of patients. In addition, the prognosis of HER2-positive breast cancer is highly correlated with the immune landscape, especially the infiltration of immune cells. Therefore, we evaluated whether the immune genomic features of metastatic breast cancer are unique.

[0360] To obtain an immunogenomic signature specific for HER2-positive metastatic breast cancer, we analyzed the genomic signatures of naive T cells co-cultured with breast cancer stem cell-enriched spheroids generated from in vitro models of HER2-positive primary breast cancer (BT 474) and HER2-positive metastatic breast cancer (MDA-MB 361). Fig.33 , a). Fig.33 The SERS spectrum in b) shows characteristic peaks related to DNA that are consistent with those reported in the literature. 77,94 .from Fig.33 ,It can also be observed in b) that the normalized intensity of metastatic breast cancer is significantly higher than that of primary breast cancer.,In addition, graph network analysis was performed to analyze the differences between primary and metastatic breast cancer. Fig.33, The network analysis presented in c) shows that the immunogenomic signatures of primary and metastatic breast cancers are significantly different without any overlap. In addition, these analyses show the overall connectivity of each sample with metastatic breast cancer. No mutual relationships were observed outside the clusters.

[0361] Next, highly connected nodes were selected to determine how connectivity can be distinguished between primary and metastatic breast cancers. Finally, centrality indices were compared to examine statistical significance. 95 The SERS peak is located at 725 cm -1 、1258cm -1 and 1287cm -1 , corresponding to the ring breathing modes of thymine / cytosine, adenine, and cytosine, respectively 77,96,97 Comparison of normalized peak intensities revealed significant differences between primary and metastatic breast cancers ( Fig.33 , d), indicating that the immune genomic signature can distinguish HER2-positive primary breast cancer from metastatic breast cancer. The study highlights the importance of HER2 status in metastasis and its impact on clinical prognosis 98,99 Taken together, these results suggest that the uniqueness of the immunogenomic signature of HER2-positive metastatic breast cancer could significantly open up the possibility of using these markers as predictive biomarkers for the management of HER2-positive metastatic breast cancer.

[0362] Immunogenomic signatures obtained using in vitro breast cancer models provided preliminary data that were then extended to clinical samples for use as a clinical diagnostic test. Therefore, in assessing the translatability of the immunogenomic signature of metastatic breast cancer to clinical samples, the immunogenicity signatures obtained from in vitro breast cancer models were compared with clinical samples from patients diagnosed with metastatic and primary breast cancer.

[0363] A hierarchical clustering algorithm was used to assess the prevalence of metastatic immune genomic signatures in DNA isolated from tissue biopsy samples. Hierarchical cluster analysis is a robust algorithm that analyzes multiple factors to classify samples based on similarities and differences and captures tumor heterogeneity. 100 . Tissue biopsy DNA represents the tumor microenvironment and its heterogeneity. The immunogenomic signature is mainly determined by the influence and interaction of immune cells in the tumor microenvironment 101 In addition, the immune escape mechanism of breast cancer is regulated by the tumor microenvironment through the formation of an immunosuppressive state. 102,103 Therefore, to evaluate the validity of an immunogenomic signature of metastatic breast cancer, it is critical to establish similarities with the genomic signature of the tumor microenvironment.

[0364] from Fig.34, it can be noted in a) that the metastatic immunogenomic signature has a high degree of similarity with a subset of tissue biopsy samples, which may be attributed to the heterogeneity of tissue biopsy samples. Subsequently, the immunogenomic signature of non-metastatic breast cancer was compared with DNA from tissue biopsy samples. Fig.34 ,c) It can be observed that there is a high degree of difference between the samples, resulting in the formation of two different clusters.,In addition, a metastatic immune genomic signature could be identified, which could serve as an effective marker for the detection and prognosis of metastatic breast cancer.

[0365] Similarly, the presence of metastatic immunogenomic signatures in serum was analyzed, which is crucial for applying metastatic immunogenomic signatures to liquid biopsies. Several studies have noted that metastatic breast cancer has a different immune phenotype and, in some cases, has a negative impact on prognosis and patient survival. 76,79,81,99,104 However, the contribution of immunogenomic signatures to circulating tumor DNA remains largely unexplored, and thus their applicability as liquid biopsy biomarkers remains unclear. Therefore, the metastatic immunogenomic signature was compared with the similarity of serum samples isolated from patients diagnosed with metastatic breast cancer ( Fig.34 , b).

[0366] Due to the high dimensionality and complexity of the data, network analysis was used to measure the similarity and effectiveness of the metastatic immune genomic signatures. Fig.34 , b) it can be observed that the metastatic immunogenomic signature and the serum signature obtained from patients diagnosed with metastatic breast cancer form a single, highly interconnected network. In contrast, the network formed by the immunogenomic signature of non-metastatic breast cancer is significantly different from the signature of the serum obtained from patients diagnosed with metastatic breast cancer ( Fig.34 , d), with a high degree of disconnection. In addition, to assess the proximity between samples, a centrality measure was used, with an observed value of 0.9, confirming the similarity between metastatic serum samples and metastatic immune genomic signatures. In contrast, the centrality measure between metastatic serum samples and non-metastatic immune genomic samples was 0, showing a distinction between the sample sets. This suggests that immune genomic alterations play a key role in cancer progression and metastasis and can be assessed non-invasively.

[0367] In this study, by utilizing the circulating components of the immunogenomic signature associated with breast cancer metastasis, the goal was to expand the immunogenomic signature to predict the potential of primary breast cancer to metastasize at an early stage, assess the phenotype of breast cancer, and detect metastatic sites. A key factor influencing the metastatic potential of primary tumors is the infiltration of immune cells. For example, T cells indirectly promote breast cancer invasion and metastasis by activating epidermal growth factor signaling and influencing the phenotype of tumor-associated macrophages. 105Likewise, immunosuppressive cells such as CD4(+)CD25(+)Tregs are able to migrate and are often co-opted by tumors to suppress immune destruction. 106–108 Furthermore, it is well known that in order to form distant metastases, cancer cells infiltrate blood vessels, migrate to distant sites, and interact with circulating immune cells. 109 Circulating cancer cells interact with CD8 T cells and inhibit their effector functions 110 . The above studies have shown that the immune system plays a key role in the formation of distant metastasis. Although the immune system plays a vital role in the progression and metastasis of cancer, the composition of immune cells in the tumor microenvironment and circulation is relatively low. Therefore, deconvolution analysis was performed to separate the immune cell composition in the genomic signatures of tissue biopsy DNA and serum samples.

[0368] Tissue biopsies measure genomic signatures that reflect every cell in a sample. However, the heterogeneity of bulk tissue complicates the interpretation and quantification of the contribution of immune cells to the tumor microenvironment. 111 Therefore, it is critical to understand the composition of cell types in bulk tissue biopsy samples through deconvolution. 112 .

[0369] Using a quasi-supervised learning strategy, we obtained training data for deconvolution using synthetic mixtures by controlling the percentages of tumor DNA, immune cell DNA, and epithelial cell DNA to account for the broader heterogeneity of clinical samples. Fig.35 Representative spectra of the synthetic mixtures are presented in a) and b), which show characteristic peaks associated with DNA molecules. In addition, the differences in peak intensity can be visually observed when the composition changes. Subsequently, linear regression analysis was performed to determine the relationship between different immune cell DNA compositions and SERS peak intensity ( Fig.35 , c), which is critical to determine whether SERS spectral data can be used for deconvolution of cell types in heterogeneous samples. The characteristic peaks associated with DNA molecules showed a linear relationship with the different percentages of immune cell DNA composition. Therefore, a linear regression algorithm was used to evaluate the performance of the deconvolution algorithm using synthetic mixture data. When estimating the cell type proportions, it was observed that the difference between the actual and predicted values ​​of the percentage of each type was small ( Fig.35 ,e). In addition, the RMSE is kept within 10%, verifying the superior performance of the proposed algorithm.

[0370] Next, the algorithm was extended to clinical tissue and liquid biopsy samples to assess its applicability to complex samples. Finally, the algorithm was trained using synthetic mixture data and validated using an independent cohort of clinical samples ( Fig.35, f). The results showed that the percentage of immune cell components in tissue biopsies was higher than that in liquid biopsy samples, which may be attributed to the extremely low amount of tumor-associated cell-free DNA in circulation. In addition, it demonstrated the ultra-high sensitivity of the nanosensor, which can detect the presence of even 0.005% of tumor-associated immune genomic features in circulation.

[0371] When analyzing tumor-associated immune genomic signatures in tissue biopsies, the percentage of immune cells increased significantly with tumor stage, with clear differences in early-stage (T1) metastatic samples. In addition, significant differences in immune cell levels were observed between HER2-positive and -negative breast tumors ( Fig.35 , f). This confirms the uniqueness and potential applicability of the metastatic immunogenomic signature in breast cancer diagnosis and clinical management, including the assessment of metastatic potential, breast cancer phenotype, and potential metastatic sites.

[0372] To evaluate the validity of the metastatic immunogenomic signature, a simple machine learning algorithm was applied to independent cohorts of serum samples from patients diagnosed with metastatic breast cancer (n=20) and non-metastatic breast cancer (n=20). The machine learning algorithm was trained using the metastatic immunogenomic signature generated from in vitro samples. Fig.36 It can be observed that the prediction of tumor metastasis by blood samples has a high accuracy. In addition, the specificity and sensitivity of this classification are 97.2% and 100% respectively ( Fig.36 , middle figure).

[0373] The clinical impact of a false-positive result from a liquid biopsy is enormous, as it may lead to a misdiagnosis or affect treatment options. 113 . Therefore, reducing false positives is crucial for detection for clinical management of cancer. Traditional liquid biopsy detection methods suffer from high false positive rates, which is one of the main problems hindering the clinical implementation of liquid biopsy. In this study, the false positive rate was reduced to <0.1% using metastatic immune genomic features. Next, the spectral features that contribute to the high accuracy prediction of breast cancer metastasis were studied. The principal components (PC1 and PC2) provide the maximum variance between sample groups, thereby accurately predicting cancer metastasis. The ubiquitous SERS peak in PC1 and PC2 is located at 725cm -1 、1258cm -1 and 1287cm -1 , corresponding to the ring breathing modes of thymine / cytosine, adenine, and cytosine, respectively 77,96,97 .

[0374] Determining the phenotype of early breast cancer is critical to determining the likelihood of metastasis and possible metastatic sites. Studies have shown that HER2 expression levels promote an aggressive phenotype and provide a selective advantage for metastasis during tumor initiation. 114 In this study, a simple machine learning algorithm was applied that showed excellent performance in differentiating HER2-positive metastatic versus HER2-negative metastatic breast cancer samples ( Fig.36 ) between the specificity and sensitivity. The common -1 、1282cm -1 and 1421cm -1 The peak position at is the strongest predictor for accurate classification of breast cancer phenotype.

[0375] Metastatic HER2-positive breast cancer is one of the most aggressive cancers 91 . In addition, the incidence of brain metastases in patients with HER2-positive breast cancer is higher than that in patients with HER2-negative breast cancer. In addition, screening for brain metastases is not currently recommended due to the lack of data supporting an overall survival benefit. 115 . Therefore, a liquid biopsy test to assess the likelihood of developing brain metastases could potentially transform the clinical management of brain metastases. The high specificity of the metastatic immune genomic signature was used to validate whether it could be used to distinguish intracranial metastatic disease from other tissue metastases. By applying a simple machine learning algorithm, the model could accurately distinguish the presence of brain metastases with a sensitivity and specificity of 98.5% and 96.7%, respectively ( Fig.36 ).

[0376] Example 11 - Detecting cancer using methylation signals of tumor-associated NK cells

[0377] Using the nanosensor synthesized in-house, 5 μL of DNA sample was placed on the nanosensor to obtain the Raman signature of healthy immune cell (NK cell) DNA and colorectal tumor interacting NK cell DNA, and the SERS light spectrum at a wavelength of 785 nm ( Fig.37 , A). Surface functionalization (-OH, C-OH, CO, C=O) of the 3D nanoprobes is beneficial for improving the interaction between cells and the sensor surface. With the help of Raman spectroscopy, nitrogenous base pairs, adenine, guanine, thymine, and cytosine can be found next to the phosphate backbone. 27,28 In this study, SERS spectra of DNA isolated from in vitro natural killer cells (NK-92) interacting with colorectal cancer cells (Colo-205) and natural killer cells that did not interact with healthy tumors (control NK-92) were captured to identify genomic differences between them ( Fig.37, B). NK cells have characteristic nucleic acid Raman spectral bands with A, T, C, G and phosphate backbones, which are completely unrelated to tumor cells. Prominent peaks include 745 cm -1 (adenine (A), thymine (T) and cytosine (C) ring breathing), 1485cm -1 and 1577cm -1 (contributions from A and guanine (G)), and 785 cm from the OPO backbone -1 、1090cm -1 and 1115cm -1 .1090cm -1 The PO2 peak at 1.30 nm was determined as an internal standard, which helps to eliminate variations in absolute intensity measurements. The PO2 band was used as an internal benchmark because DNA structural damage has little effect on it. The results were compared using a two-tailed Student's t-test ( Fig.37 , C), highlighted spectral points showing statistically significant differences between tumor-associated NKDNA and NK DNA (750 cm -1 、781cm -1 、1010cm -1 and 1424cm -1 *0.05, **0.01, and ***0.001, respectively). The contribution of each peak to the spectral difference was determined by subtracting the healthy NK DNA spectrum from the spectrum of NK cell DNA interacting with tumor ( Fig.37 ). The spectral intensity of NK cell DNA associated with colorectal tumors is slightly different from that of NK cells not associated with tumors. As confirmed by the SERS spectra, NK cell activity is significantly altered at the genomic level in the presence of colorectal cancer cells. 29 The spectral changes were caused by these intracellular changes driven by NK DNA.

[0378] NK cells are epigenetically regulated in the tumor microenvironment. The epigenetic state of NK cells either contributes to immune clearance or promotes cancer progression. Among epigenetic modifications, DNA methylation is an important epigenetic change that determines cell properties. DNA methylation is an indispensable epigenetic modification that determines cell properties. The DNA methylation signal is unique due to its chemical stability and the insertion of methyl groups at the fifth position of cytosine and the sixth position of adenine. In the tumor microenvironment, NK cells develop a unique methylation profile that can be used to detect the presence of cancer.

[0379] The ability of the nanosensor to detect changes in DNA molecules allowed for further investigation into the ability to assess changes in methylation in NK cells. Prior to measuring the DNA methylation status of NK cells, methylated and unmethylated DNA standards were analyzed using Raman spectroscopy. Using the 5-mC standard control, an optimal standard curve was generated based on the optical density at 450 nm ( Fig.38 , A). Then, a linear correlation model was established using the Raman peak intensity and methylation percentage. Raman spectra of 5mC and methylated DNA ( Fig.38 , C) shows different degrees of methylation, 700 cm -1 Up to 1800cm -1 The signal in the region with the increase in the standard curve was calculated using a linear fit (R 2 = 0.9619) and the equation y = 0.2018x + 0.004 ( Fig.38 , B), calculate the methylation percentage in the DNA sample. Fig.38 , E shows that among the many DNA features, at 765cm -1 、924cm -1 、1020cm -1 、1310cm -1 and 1555cm -1 Five unique peaks were identified at 4 nt, corresponding to adenine and cytosine ring breathing patterns. These peaks were considered methylation-specific markers because changes in peak intensity and peak shift can be attributed to methylation-induced structural modifications. The relative spectral intensity of the methylation markers of NK cells interacting with colorectal tumors was elevated compared to the DNA of NK cells that did not interact with the tumor. Using the methylation-specific Raman markers, multivariate analysis using PCA obtained two separate clusters showing that healthy NK cell DNA and tumor-interacting NK cell DNA ( Fig.38 , F). The difference in the peaks also implies the insertion of a methyl group at the position of the cytosine nucleotide. To validate the in vitro studies, methylation markers were examined in patient blood samples with patient DNA samples from these experiments. As expected, colorectal cancer patients showed elevated methylation of immune cells ( Fig.38 , G).

[0380] Given that the changes in DNA molecules and methylation-related changes in DNA in NK cells that interacted with colon tumors (colo-205 co-cultured with NK-92) were different from those in NK cells that did not interact with colon tumors (control NK-92), the next step was to test whether the identified markers could produce the same results in patient samples. To this end, SERS spectra of 20 clinical samples obtained from the Ontario Tumor Board (OTB) were obtained. DNA was extracted from the peripheral blood of cancer patients diagnosed by clinical manifestations and histopathological diagnosis. The patient groups had colorectal cancer (n=7), and healthy (n=10) volunteers. After dropping 5μL of patient samples on the sensor, Raman spectra of 17 samples were collected. Fig.39 Figures A and B show the Raman spectra of the patient samples, respectively, with the genomic DNA molecules and methylation markers highlighted. To perform PCA, the collective Raman spectra of the molecules and methylation markers were used. The PCA results show that the patient samples are clearly clustered into two different groups ( Fig.39 , D). Subsequently, the coefficient of each variable in the first component is plotted against the coefficient in the second component, and the peaks that cause the difference are obtained from the plotted loading diagram ( Fig.39 , E). Raman spectra of samples from colorectal cancer patients and healthy volunteers were compared using heat maps ( Fig.39 , F). Each cell in the heat map is color-coded according to the spectral intensity value. The heat map clearly illustrates the difference between the two sample types.

[0381] Using DNA structure and methylation peaks, an exploratory approach based on principal component analysis was applied to the Raman spectra of NK cell DNA (CNKDNA) that interacted with colorectal tumors in vitro and NK cell DNA (NK DNA) that did not interact with tumors. The first two principal components (PCs) were selected, where two distinct clusters of CNKDNA (right) and NKDNA (left) can be seen ( Fig.40 , A). By examining the loading of each variable on the first and second principal components, the variables responsible for sample clustering can be identified, such as Fig.40 , B). This enables the identification of variables responsible for sample clustering. The features that contribute the most variance are identified and then used to distinguish cancer individuals from healthy individuals.

[0382] After completing the initial phase of feature extraction with the help of PCA, a PLS-DA model was developed to predict the class membership of newly collected observations. PLS-DA is a supervised classification technique that looks for a direct association between the variables under study and the predicted response. The expected response for the two classes (n=2, healthy and cancer) is either 0 or 1. The samples were divided into a calibration dataset and a test dataset, allowing the development of two different classification models. PLSDA was trained using the DNA structure and methylated Raman peaks of the patient's blood using Venice blind cross validation. The trained model was used to develop a cancer classification system. Using the sample distribution of the calibration predictions generated by the PLS model, a threshold of 0.5 was determined to be suitable for classifying the data into their respective classes. Samples above this threshold were classified as cancer, while samples below this threshold were considered healthy. The results of the PLS-DA calibration model are shown in Fig.40 , C). The results of the analysis were a predictive value of 1.0, an accuracy of 100%, a precision of 100%, a specificity of 100%, and a sensitivity of 100%. The area under the curve was equal to 1, and the coefficient of determination (R 2 ) is 0.99, and the predicted coefficient of determination is 0.96. Fig.40 , C).

[0383] In addition to PLSDA, artificial neural networks (ANN) were used to show that the predictions were independent of the algorithm used. Due to their adaptability and autonomous learning capabilities, they are widely used in the identification and diagnosis of cancer. An ANN is a multi-layered network structure consisting of many neurons connected to each other according to predetermined rules. It was trained using data from cell cultures and cross-validated (80:20 split). The model had a sensitivity and specificity of 100% for predicting healthy colorectal tissue and colorectal tumors. To calculate the loss of a machine learning model, the training loss and validation loss ( Fig.40 , D). The accuracy of the model can be evaluated by comparing the predicted data with the actual data. The loss of the model indicates whether the model is effective after each iteration, while the accuracy of the model can measure the efficiency of the model execution.

[0384] Example 12 - Characteristics of GBM-associated NK cells for cancer diagnosis

[0385] Materials and methods

[0386] Cell lines and co-culture assays

[0387] Nk-92 cells were obtained from ATCC (American Type Culture Collection, USA). The cell source is peripheral blood of non-Hodgkin's lymphoma. NK 92 cells are similar to circulating NK cells in morphology and function. Alpha minimum essential medium containing 0.2mM inositol, 0.1mM 2-mercaptoethanol, 0.02mM folic acid, 12.5% ​​horse serum and 12.5% ​​fetal bovine serum without nucleosides is used to maintain cultured cells. IL-2 is added at a concentration of 150IU / ml to ensure full proliferation. A-172 glioblastoma cancer cell line was obtained from ATCC and grown in DMEM medium containing 10% FBS. Both cell lines were cultured at 37°C and 5% CO2. NK-92 cells were co-cultured with three breast cancer cell lines using a Transwell device (pore size 0.4m, Corning, Lowell, Massachusetts). NK-92 cells were inoculated in the upper insert plate of a 24-well Transwell plate.

[0388] Isolation and identification of CIV

[0389] NK-92 cells were grown in the prepared medium for 24 h. Cells were stimulated with IL-2 and a total exosome isolation kit (Invitrogen) was used. Briefly, the supernatant was collected and incubated with the reagents in the kit overnight at 2-8°C. The vesicles were then recovered by centrifugation at a standard centrifugation speed of 10,000xg for 60 min. The particles were then redissolved in water for immediate use. For TEM analysis, CIV samples were adsorbed on metal grids covered with a lace-like membrane and fixed with a fixative to maintain shape and prevent osmotic damage. To improve low contrast, negative staining was performed with uranyl acetate contrast agent.

[0390] SERS data, collection, preprocessing and analysis

[0391] SERS spectra were acquired using a Renishaw invia confocal spectrometer and a Leica DMI 6000 epifluorescence microscope. A Raman shift of 520 cm -1 The spectrometer was calibrated with a silicon wafer. Crystal violet (CV) of different concentrations was prepared. To collect spectra of CV, NK cells, and CIV, 5 μl of the corresponding sample was placed directly on the sensor. The spectra of cells and extracellular vesicles were collected three times with a collection time of 10 s and the average was taken. The excitation wavelength was 785 nm and the spectral resolution was 1 cm -1 Prior to analysis, the raw data were preprocessed with SpectraGryph 1.2 for cosmic ray removal, smoothing (Savitsky-Golay), and baseline correction (adaptive).

[0392] Statistical analysis

[0393] Where applicable, multivariate and univariate analyses were performed. Multivariate statistics were performed using Solo Eigenvector and univariate analyses were performed using Prism Graphpad 9.0. All data are presented as mean ± SD. For comparison of individual variables between the two groups, Student's t-test was used. Pearson's test was used for correlation analysis. Principal component analysis (PCA) was used for multivariate analysis. PCA showed outliers (Hotelling T2 test) and variance in the sample by taking all the features in the sample, and the component score plot showed the correlation within the sample. PLSDA was performed to distinguish categorical variables. Confidence ellipses with a confidence level of 95% were used to classify the groups. 80% of the original data in the sample were used as training data and 20% were used for model prediction purposes. The calibration model was cross-validated using calibration error (training data) and cross-validation error (test data), and the prediction performance of the model was checked using the cross-validation data. The root mean square of the cross-validation RMSECV showed the quality of the model. The performance of the model was confirmed by the sensitivity, specificity, receiver operating characteristic curve (ROC) and area under the curve (AUC) of the training, cross validation and test sets. The importance of the variable (VIP) in the PLSDA score was calculated by the weighted sum of the squared correlations between the PLS-DA components and the original variables. The weights were obtained based on the percentage change of the PLS-DA components. Artificial neural networks (ANN) were applied to the Raman peaks derived from PCA using Keras, Numpy and Sklearn in Python. Generative adversarial networks (GAN) were used for data augmentation.

[0394] This example demonstrates that the molecular profile of circulating immune cell vesicles that interact with the glioblastoma tumor microenvironment (TME) provides diagnostic information for the detection of glioblastoma. NK cells play an important role in glioblastoma immune surveillance and respond to signals from GBM. NK cells are reprogrammed to promote or suppress tumors. This editing of tumor cells induces metabolic changes in NK cells, resulting in alterations in proteins, lipids, and nucleic acids within the cells. In addition, these altered metabolites are selectively packaged into NK cell vesicles, generating circulating immune vesicles (CIVs) with unique metabolic profiles. In order to detect CIVs, an ultrasensitive sensor capable of detecting low levels of analytes in the circulation is required. For this purpose, surface enhanced Raman spectroscopy (SERS) was used as a detection method. SERS is a versatile spectroscopic technique characterized by the non-destructive and rapid detection of analytes in the circulation. As Fig.41 As shown, two phenotypic spectra were obtained using SERS - the original CIV spectrum and the GBM-related CIV spectrum. These phenotypic subtypes of CIV form the basis for GBM detection in this example. Fig.41The graphical abstract in illustrates the diagnostic technique. In order to identify different CIV signatures, a machine learning classifier was used to obtain tumor-related signals from the Raman data. This was achieved by separating the spectra into healthy spectra and spectra associated with GBM using partial least squares discriminant analysis (PLS-DA). NK-92 cells were co-cultured with A172 glioblastoma cells to obtain GBM-CIV, and their Raman data were used as input training data, while healthy CIV was used as a control. Successful classification was obtained after training, and the model provided 100% sensitivity and 100% specificity on the test data (AUC = 1.0).

[0395] Characterization of NK cell-derived immune vesicles

[0396] Since activated NK cells are known to produce greater numbers of EVs, CIVs from parental NK-92 cells stimulated with interleukin-2 (IL-2) were generated (Pirzadeh Z, et al., 2014). Fig.42 , D) Isolated EVs were characterized by nanoparticle tracking analysis (NTA) and electron microscopy. NTA measures the size of larger vesicles (microvesicles) and smaller vesicles (exosomes) by a digital camera that records the Brownian motion of individual particles and calculates the diameter of the particles using the Stokes-Einstein equation (Himstedt R, et al., 2017). The size range of NK-92 cell-derived microvesicles was 150-450 nm, and the size range of NK-92 cell-derived exosomes was 50-100 nm, which is similar to the size of microvesicles and exosomes described elsewhere (He S, et al., 2019; Pirzadeh Z, et al., 2014). Fig.42 , D) A snapshot of the NTA video is also shown. The vesicles appear as electron-dense dark outer membranes and electron-transparent bright inner masses. Larger vesicles appear to have a cauliflower morphology due to protein aggregation around the EVs (Shibue T and Weinberg RA, 2017). Overall, the results of NTA and TEM indicate that the size range of the vesicles isolated in this example is 50-450nm.

[0397] Biochemical similarity of immune cell vesicles to parental NK cells using SERS analysis

[0398] The NK cell (NKC) spectrum is compared with the CIV spectrum to determine whether the CIV exhibits similar properties to the parental NK cells for diagnostic purposes ( Fig.42 ). The SERS spectra of NKC and CIV were analyzed to understand their biochemical compositions. Fig.42 , C show the spectra of NKC and CIV, respectively, and their tentative assignments are shown in Table 11.

[0399] Table 11. SERS assignments of NK cells (NKC) and NK cell-derived extracellular vesicles (NKEV).

[0400]

[0401] The NK cell spectrum clearly shows information about proteins, lipids, carbohydrates, and nucleic acids. Proteins such as amide III (1224 cm -1 ), Amide I (1632cm -1 )、Tryptophan(1359cm -1 ), tyrosine (859cm -1 ) at 678 cm corresponding to the ring breathing mode in DNA bases. -1 The 789 cm corresponding to the OPO stretching in DNA -1 Analysis of the CIV optical spectra showed that they have a similar biochemical composition to the parental cells, such as Fig.42 , c)vi). The spectral composition of CIV is mainly attributed to the phospholipid bilayer rich in cholesterol nucleic acids and cargo proteins. Therefore, the spectrum shows a vibration band at 1300 cm -1 (δCH2 twist, phospholipid swing) and 719cm -1 , corresponding to the symmetric stretching vibration of the choline group N+(CH3)3, which is characteristic of phospholipids. 669cm -1 The nucleic acid peak at 720 cm corresponds to the contribution of thymine and guanine in DNA / RNA, OPO stretches DNA phosphodiester, -1 (DNA), 644cm -1 The protein peak at 761 cm corresponds to the CC distortion mode of tyrosine. -1 The respiratory tryptophan ring at 951 cm -1 (α-helix of protein), 1224 cm -1 (amide III protein), and its band at 701 cm -1 (Cholesterol), 1255cm -1 lipids) and 1300cm -1The lipids at correspond to CH2 twists and sways (lipids). In order to reduce the dimensionality of the data and make it easier to interpret, principal component analysis (PCA) was performed. PCA compresses the high-dimensional data of SERS into a 3D graph by selecting the maximum changes within the data set and plotting along the component direction. Although PCA is performed to show the maximum variance, here PCA is used to show the similarity between spectra. The first three components (PC1, PC2, and PC3) are selected to include at least 70% of the variance. In the PCA score graph with 3 components, NKC and CIV samples are closely clustered, confirming that the composition of cells and vesicles is almost identical. In addition, in order to find key similarities between the spectra, chord diagrams were drawn to intuitively see the similarities in the biochemical components of the cell and EV spectra. Fig.42 , c)vi) shows the connections between CIV and parental NK cells and selected principal components, the lower half of the circle represents immune cells and their components. The connections shown in blue are the biochemical similarities of CIV with the first three principal components, the connections shown in red (NK cells) show similar connections with the first three principal components, and the metabolites are divided into different clusters, indicating that the metabolic characteristics have changed.

[0402] Compared with immune vesicles derived from non-interacting NK cells, immune vesicles derived from GBM-associated NK cells The SERS spectra show significant differences

[0403] To understand the behavior of NK cells in the tumor microenvironment, it is important to understand how NK cells identify, differentiate, and kill cancer cells. Therefore, their cellular metabolic expression varies depending on the cell type they interact with. Example 2 demonstrates that NK cells exhibit different SERS spectra when interacting with tumor cells due to the differentiated culture of cancer and undifferentiated cancer stem cells. Both resting and activated NK cells release extracellular vesicles (EVs) containing parent metabolites. The metabolites inside CIVs are mainly proteins, lipids, nucleic acids, sugars, and vitamins, representing the metabolic state of the parent cells. The metabolic state of the parent NK cells significantly affects the content of the EVs they produce. This is evident from the unique SERS spectra obtained from GBM-CIV ( Fig.43 ). The complete spectrum assignment can be found in Table 12. To obtain the characteristic spectrum of NK cell-derived extracellular vesicles associated with glioblastoma, NK-92 cells were exposed to A-172 glioblastoma cells for 24 hours. After interaction, the cell supernatant was collected and centrifuged to obtain EVs, named GBM-CIV. The supernatant of non-interacted NK-92 cells was used as a control to obtain EVs, named CIV.

[0404] Table 12. SERS assignments of GBM-CIV.

[0405] <![CDATA[Peak (cm -1 )]]> T tentative SERS assignment 535 Cholesterol ester 787 RNA 813 C'5-OPO-C'3 phosphodiester bands in RNA 844 Tyrosine (Fermi resonance of fundamental frequency and harmonics) 932 Backbone CC, α-helix of protein 1003 Phenylalanine, CC backbone 1064 Lipid backbone CC stretching 1094 DNA 1142 v(CN), protein (protein distribution) 1240 RNA 1235 Amide III 1281 Amide III and CH2 wiggling vibrations from the glycine backbone and proline side chains 1397 C==O symmetrical stretching 1447 CH2 deformation (protein vibration) - a marker of protein concentration 1482 Guanine and adenine ring breathing patterns in DNA bases 1672 Amide I band (C==O stretch connected to NH bend) 2887 Compared with CH3, lipids, and fatty acids 2945 CH vibrations in lipids and proteins 3015 CH of lipids

[0406] In a nutrient-poor environment, NK cells compete with GBM tumor cells for glucose and amino acids, exhibiting the "Warburg phenomenon" and shifting metabolism to aerobic glycolysis (Ames E, et al., 2015; Tallerico R, et al., 2013; Butler KT, et al., 2018). Lactate and pyruvate are metabolic indicators of glycolysis, and the levels of lactate and pyruvate in NK cells increase after metabolic reprogramming. Consistent with this, metabolic profiles become increasingly important as cells transform into effector cells. 866cm -1 and 835cm -1 The SERS features at 340 Å correspond to lactate and pyruvate, respectively, and can be used to detect glycolytic reprogramming that produces different SERS metabolic features. In GBM CIVs, the concentrations of lactate and pyruvate metabolites were significantly lower than those in healthy CIVs (p<0.0001).

[0407] In addition, by comparing 1125cm -1 The intensity at 866 cm -1 (lactic acid) and 835cm -1 The intensities at (pyruvate) can be used to assess the relative changes in these metabolites relative to glucose. The regression curves show a strong positive correlation between pyruvate-glucose and lactate-glucose (R 2 =0.9)( Fig.43 , C ). Overall, these findings suggest that glycolytic reprogramming distinguishes GBM CIVs from healthy CIVs, which maintain the increased energy demands induced by nutrient deprivation at the tumor site.

[0408] Fatty acid metabolism also plays a role in the development of different CIV phenotypes. Lipids are the main components of CIV. However, in the TME, GBM cells release a large amount of fatty acids into the environment, which interferes with the activity of NK cells. Intracellular lipid accumulation reduces the cytotoxic function of NK cells (Guo S, et al., 2021). In addition, changes in lipid content can affect signal transduction pathways, membrane fluidity and transport (Meza Ramirez CA, 2021). Changes in immune responses are due to functional changes, which are related to changes in phenotype. The lipid contribution in GBM CIV and healthy CIV was studied using SERS. The vibration is located at 3015cm -1 、2945cm -1 、1142cm -1 and 1064cm -1 , corresponding to the CH band, the asymmetric vibration of CH2, and the backbone CC stretching of fatty acids and lipids (acyl chains) ( Fig.43 , G).

[0409] When NK cells are exposed to tumor cells for a long time, their immune response is altered. Activation and inhibition of NK cells are caused by a complex metabolism that requires a high degree of protein regulation. Several protein receptors and molecules (natural cytotoxicity receptors - NCR, cytotoxic proteins, CD16, CD56, NKp44, NKG2D, PDL1) are transferred into extracellular vesicles in the form of dissolved particles. These proteins can be detected on the surface or in the cytoplasm of CIVs. The secondary structure of proteins and the relative content of different aromatic amino acids are probed using SERS. 1447cm -1 The SERS peaks at 1672 cm indicate protein concentration, and the most significant change is observed at 1672 cm -1 Amide I at 1235 cm -1 Amide III at 1003 cm -1 Phenylalanine ( Fig.43 , H). While amide I and phenylalanine were significantly reduced in GBM-CIV, an increase in the amide III peak was seen in GBM-CIV. In addition, the entire SERS spectrum was analyzed to visualize the complete metabolic profile in the form of PCA. The PCA score curve was used to explore lipids, proteins, and nucleic acids. PC1, PC2, and PC3 explain Fig.43 , the maximum variance in DF. The data within the same cluster are close to each other, showing highly reproducible results. Fig.43 Clear differences in the lipid and protein composition of EV packaging were revealed. Fig.43 Results of univariate (two-tailed Student's t-test) and multivariate analysis (PCA) are shown, demonstrating distinct signatures. In conclusion, the results suggest that the metabolic signature of GBM-CIV can be used as a biomarker for GBM diagnosis.

[0410] Validation of GBM interactions

[0411] Classically, programmed death ligand (PDL-1) has been found in tumors interacting with infiltrating immune cells as a way to evade detection. Since NK cells only express PDL-1 when interacting with tumors, PDL-1 expression in NK cells can be used as a reliable indicator of cancer involvement. PDL-1 is also secreted into vesicles after being endocytosed. Therefore, PDL-1 expression in GBM-CIV can be used to confirm the results of SERS. PDL-1 expression in GBM-CIV was compared using SERS spectra of GBM-CIV and healthy CIV. Fig.44 , A) shows the Raman characteristic spectrum of PDL-1. The prominent peak appears at 846 cm -1 (tyrosine Fermi resonance), 855cm -1 (tyrosine v(CC)), 952cm -1、1083cm -1 (CN stretching mode of protein), 1125cm -1 (CC stretching mode of protein) and 1464 cm -1 (Fermi interaction (bending CH2 and stretching CH2)). The relative expression of PDL-1 in GBM-associated CIV and healthy CIV was highly significant ( Fig.44 , F; two-tailed t test, p < 0.0001 and R 2 =0.99), and was further supported by PCA and biplot ( Fig.44 , D).

[0412] GBM tumor diagnosis using GBM-associated circulating immune vesicles

[0413] A classification model for GBM diagnosis was developed by applying supervised partial least squares discriminant analysis (PLS-DA) to measure the difference between the two phenotypes of CIV. PLS-DA was chosen because it is a powerful tool for chemometrics when the number of wavelengths exceeds the number of samples. PLS-DA combines partial least squares regression analysis and discriminant analysis classification to build a mathematical model to assign unknown samples to different categories (0 = healthy and 1 = GBM). Fig.45 , (A) shows the PLS-DA flow chart, describing the process of PLS ​​regression model combined with discriminant analysis to predict glioblastoma. Two types of spectral data were used as input training data: (1) GBM-CIV (cancer) and (2) CIV (healthy). As with the unsupervised data clustering, supervised PLS-DA showed similar clustering. The PLS-DA score curve distinguished CIV phenotypes with no overlap ( Fig.45 , (B)(ii)). To find the variables that lead to the greatest separation, the VIP score was calculated. The variable projection importance (VIP) score measures the importance of each variable in the PLS model projection to discover key variables. In this model, variables with a VIP score of more than 1.4 are considered important. Fig.45 , (B)(iii) shows that the model predicts several amino acids, lipids, and sugars as important variables. This suggests that metabolites involved in NK cell metabolism may play an important role in determining the diagnostic profile.

[0414] In order to evaluate the performance of the classification model and obtain the total number of latent variables (LV), Venetian crossover blind injection was used as a cross-validation (CV) method. In addition, the original data was divided into two independent datasets, 20% of the original data was classified as validation data, and the remaining (80% of the data) was grouped as calibration data and used to build the model.

[0415] Subsequently, the model was applied to the validation data and tested. Serum samples of GBM-positive and healthy patients were used as external validation of the input test data. Table 13 shows the output of the model in the form of a confusion matrix (Matthew correlation coefficient = 1.000). 0.42 was the threshold selected to classify samples as GBM or healthy to evaluate the performance of the prediction model using the ROC curve AUC (AUC, 0.9). Table 14 shows the statistical results obtained by the PLS-DA model, showing a high R 2 value (0.99) and a low error rate (RMSEP 0.02). Table 15 shows the classification accuracy of the model. Accuracy is determined by the ratio of successful predictions to the total number of samples. The optimized PLS-DA model correctly predicted 10 GBM and 10 healthy samples out of 20 samples. The 100% sensitivity and 100% specificity obtained from the ROC for GBM detection are promising for future detection. Metabolic signatures do not rely on a single biomarker, but rather provide collective changes that occur in the disease, eliminating the possibility that random changes in a single metabolite are associated with the disease. The diagnostic model takes into account collective spectral data, eliminating the need to associate a single biomarker for cancer detection.

[0416] Table 13. Model confusion matrix for GBM diagnosis.

[0417]

[0418] Table 14. Statistical results of the PLS-DA model performance for GBM diagnosis using serum samples.

[0419] calibration Cross Validation predict <![CDATA[R 2 ]]> 0.998296 0.985529 0.998296 error RMSEC:0.0206396 RMSECV:0.0629593 RMSEP:0.0206396 Sensitivity 1.000 1.000 1.000 Specificity 1.000 1.000 1.000 Classification Error 0 0 0

[0420] Table 15. Cross-validation confusion matrix for GBM diagnosis using serum samples.

[0421]

[0422] Example 13 - Using NK cell extracellular vesicles to differentiate astrocytomas, oligodendrogliomas, and glioblastomas tumor

[0423] The term glioma encompasses astrocytomas, oligodendrogliomas, and glioblastomas. GBM is a high-grade (grade IV) tumor, while astrocytomas and oligodendrogliomas are low-grade (grade II) tumors. Since GBM tumors have been shown to be able to train NK cells, it was assessed whether low-grade gliomas such as astrocytomas and oligodendrogliomas can also train NK cells to determine whether this would lead to metabolic changes in tumor-experienced NK cells, which would be reflected in the vesicles of NK cells, resulting in CIVs with unique metabolic profiles. To this end, tumor cells were isolated from patients diagnosed with grade II astrocytomas and grade II oligodendrogliomas. The isolated tumor cells were co-cultured with NK cells, and after interaction, the supernatant was collected and centrifuged (according to the manufacturer's protocol) to produce EVs, which were subsequently named astrocytoma-associated CIVs and oligodendroglioma-associated CIVs, respectively. Fig.46 A and Fig.46 The respective Raman spectra are shown in B. These Raman spectra were then compared with GBM-associated CIV to determine whether NK-CIV could distinguish between low-grade and high-grade tumors.

[0424] Fig.46 C shows the principal component analysis using the first two principal components, showing that Raman spectroscopy can clearly distinguish low-grade (astrocytoma vs oligodendroglioma), as well as low-grade and high-grade (glioblastoma) tumors. To further identify the key spectral data points that characterize low-grade and high-grade tumors, spectral data points were obtained from the PC1 and PC2 loadings of the PCA. The PC1 loadings, namely 776, 973, 1080, 1172, 1337, 1443, and 1665 cm -1 The spectral data points of , contribute most to distinguishing low-grade oligodendroglioma from high-grade GBM ( Fig.46 D), was selected as a candidate biomarker. It is noteworthy that the spectral data points associated with proteins (973, 1172, 1337 and 1655 cm -1 ) have positive values, while the spectral data points associated with lipids (776, 1080, and 1443 cm -1 ) has negative values. The above analysis leads to the selection of these seven spectral data points as potential differentiation markers for distinguishing oligodendroglioma from GBM. Similarly, the PC2 loadings, i.e., 884, 859, 1250, 1400 and 781, 1182, 1223, 1424 cm -1 The spectral data points of , contribute most to distinguishing low-grade astrocytoma from high-grade GBM ( Fig.46 E), was selected as a possible biomarker to differentiate astrocytoma from GBM. Spectral data points associated with the protein have positive values ​​(884, 859, 1250, 1400 cm-1 ), while the spectral data points associated with nucleic acids have negative values ​​(781, 1182, 1223, 1424 cm -1 ).

[0425] Example 14 - Detection of difficult-to-detect ovarian cancer using extracellular vesicles from tumor-associated T cells

[0426] We obtained Raman spectroscopic signatures of extracellular vesicles (EVs) from healthy T cells and T cells relevant to a preclinical model of ovarian cancer. Fig.47 , A). Heat map ( Fig.47 , B) and principal component analysis ( Fig.47 , C) both show clear and discrete clusters of healthy and ovarian cancer-associated T cell EVs. Cross-validation of preclinical models using patient plasma ( Fig.47 , D), and demonstrated that the model can classify samples as ovarian cancer with 60% sensitivity and 100% specificity ( Fig.47 , F).

[0427] Example 15 - Early diagnosis of brain cancer using tumor-associated T cells

[0428] Raman spectral signatures were obtained from T cells associated with preclinical models of early-stage and advanced brain cancer ( Fig.48 These showed distinct clustering of early and late-stage brain cancers in preclinical models ( Fig.48 , C).

[0429] Example 16 - Application of tumor-associated immune cells in distinguishing between benign and malignant cancers

[0430] T cells play a key role in recognizing tumor cells, including benign ones. Specifically, T cells have been shown to infiltrate benign brain tumors such as meningiomas and may play a role in controlling tumor growth. T cells can also induce an immune response that can recognize and destroy nearby tumor cells. T cells respond to benign tumors by recognizing specific tumor-associated antigens and attracting other immune cells to the tumor site. By studying the immune response of T cells to benign brain tumors, in this study, meningiomas were distinguished from other malignant gliomas with high accuracy, such as Fig.49 The confusion matrix predicted by ANN is shown in Table 16.

[0431] Table 16. Confusion matrix for ANN prediction of benign or malignant tumors.

[0432]

[0433] Although the applicant's teachings described herein are combined with various embodiments for illustrative purposes, this does not mean that the applicant's teachings are limited to these embodiments, as the embodiments described herein are only examples. On the contrary, the applicant's teachings described and shown herein encompass various alternatives, modifications and equivalents without departing from the embodiments described herein, the general scope of which is defined in the appended claims.

[0434] References cited in the Chinese version of the manual

[0435] Ames E.et al., "NK cells preferentially target tumor cells with acancer stem cell phenotype," The Journal of Immunology, vol.195, no.8, pp.4010–4019, 2015.

[0436] Butler KT, Davies DW, Cartwright H., Isayev O., and Walsh A., "Machinelearning for molecular and materials science," Nature, vol.559, no.7715, pp.547–555, 2018.

[0437] El-Said WA, Kim T.-H., Kim H., and Choi J.-W., "Analysis of Intracellular State Based onControlled 3D Nanostructures Mediated SurfaceEnhanced Raman Scattering," PLOS ONE, vol.6, no.2, p.e15836, Feb.2011, doi:10.1371 / journal.pone.0015836.

[0438] Gentles, AJ et al. The prognostic landscape of genes and infiltrating immune cells across human cancers. Nat. Med. 21, 938–945 (2015).

[0439] Guo S.,Popp J.,and Bocklitz T.,“Chemometric analysis in Ramanspectroscopy from experimental design to machine learning–based modeling,”Nature protocols,vol.16,no.12,pp.5426–5459,2021.

[0440] Himstedt R.et al.,“Localized surface plasmon resonances of variousnickel sulfidenanostructures and Au–Ni3S2 core–shell nanoparticles,”Chemistryof Materials,vol.29,no.17,pp.7371–7377,2017.

[0441] Ichimura,T.et al.Non-label immune cell state prediction using Ramanspectroscopy.Sci.Rep.6,1–7(2016).

[0442] Jiang S.,Munker R.,and Andreeff M.,“Bcl-2 is expressed in humannatural killer cells and isregulated by interleukin-2,”Nat.Immun.vol.15,no.6,pp.312–7,1997 1996.

[0443] Kornberg M.D.,“The immunologic Warburg effect:Evidence andtherapeutic opportunities in autoimmunity,”Wiley Interdiscip Rev Syst BiolMed,vol.12,no.5,pp.e1486–e1486,Sep.2020,doi:10.1002 / wsbm.1486.

[0444] Kresovich,J.K.et al.“Prediagnostic Immune Cell Profiles and BreastCancer.”JAMA Netw.Open 3,1–14(2020).

[0445] López-Soto A.,Gonzalez S.,Smyth M.J.,and Galluzzi L.,“Control ofmetastasis by NK cells,”Cancer cell,vol.32,no.2,pp.135–154,2017.

[0446] Luna J.I.,Grossenbacher S.K.,Murphy W.J.,and Canter R.J.,“NaturalKiller Cell Immunotherapy Targeting Cancer Stem Cells,”Expert opinion onbiological therapy,vol.17,no.3,p.313,2017.

[0447] Meza Ramirez C.A.,Greenop M.,Ashton L.,and Rehman I.U.,“Applicationsof machinelearning in spectroscopy,”Applied Spectroscopy Reviews,vol.56,no.8–10,pp.733–763,2021.

[0448] Movasaghi,Z.,Rehman,S.&Rehman,I.U.Raman spectroscopy of biologicaltissues.Appl.Spectrosc.Rev.42,493–541(2007).

[0449] Myers,J.A.and Miller J.S.,“Exploring the NK cell platform for cancerimmunotherapy,”Nature reviews Clinical oncology,vol.18,no.2,pp.85–100,2021.

[0450] Pirzadeh Z.,Pakizeh T.,Miljkovic V.,Langhammer C.,and Dmitriev A.,“Plasmon–interband coupling in nickel nanoantennas,”Acs Photonics,vol.1,no.3,pp.158–162,2014.

[0451] Ronald,J.A.,Chuang,H.Y.,Dragulescu-Andrasi,A.,Horia,S.S.&Gambhira,S.S.Detecting cancers through tumor-activatable minicircles that lead to adetectable blood biomarker.Proc.Natl.Acad.Sci.U.S.A.112,3068–3073(2015).

[0452] Salmon,H.&Donnadieu,E.Within tumors,interactions between T cells andtumor cells are impeded by the extracellular matrix.Oncoimmunology 1,992(2012).

[0453] Savas,P.et al.Clinical relevance of host immunity in breast cancer:from TILs to the clinic.Nat.Rev.Clin.Oncol.13,228–241(2016).

[0454] Sheetz K.E.and Squier J.,“Ultrafast optics:Imaging and manipulatingbiological systems,”Journal of Applied Physics,vol.105,no.5,p.2,2009.

[0455] Shen Y.,Yue J.,Xu W.,and Xu S.,“Recent progress of surface-enhancedRaman spectroscopy for subcellular compartment analysis,”Theranostics,vol.11,no.10,pp.4872–4893,2021,doi:10.7150 / thno.56409.

[0456] Shibue T.and Weinberg R.A.,“EMT,CSCs,and drug resistance:themechanistic link and clinical implications,”Nature reviews Clinical oncology,vol.14,no.10,pp.611–629,2017.

[0457] Tallerico R.et al.,“Human NK cells selective targeting of coloncancer–initiating cells:a role for natural cytotoxicity receptors and MHCclass I molecules,”The Journal of Immunology,vol.190,no.5,pp.2381–2390,2013.

[0458] Varn,F.S.,Tafe,L.J.,Amos,C.I.&Cheng,C.Computational immune profilingin lung adenocarcinoma reveals reproducible prognostic associations withimplications for immunotherapy.Oncoimmunology 7,1–10(2018).

[0459] Wellenstein,M.D.&de Visser,K.E.Cancer-Cell-Intrinsic MechanismsShaping the Tumor Immune Landscape.Immunity 48,399–416(2018).

[0460] Wong D.M.,Bol’shakov A.A.,and Russo R.E.,“Laser induced breakdownspectroscopy,”in Encyclopedia of Spectroscopy and Spectrometry,2010,pp.1281–1287.

[0461] Xu Y.et al.,“Killer immunoglobulin-like receptors / human leukocyteantigen class-I,a crucial immune pathway in cancer,”Annals of translationalmedicine,vol.8,no.5,2020.

[0462] Zhao,W.et al.Deconvolution of cell type-specific drug responses inhuman tumor tissue with single-cell RNA-seq.Genome Med.13,(2021).

[0463] References cited in the instructions

[0464] 1.Waldman,A.D.,Fritz,J.M.&Lenardo,M.J.A guide to cancerimmunotherapy:from T cell basic science to clinical practice.Nat Rev Immunol20,651–668(2020).

[0465] 2.Ren,H.,Cao,K.&Wang,M.A Correlation Between DifferentiationPhenotypes of Infused T Cells and Anti-Cancer Immunotherapy.Frontiers inImmunology 12,(2021).

[0466] 3.Andreatta,M.et al.Interpretation of T cell states from single-celltranscriptomics data using reference atlases.Nat Commun 12,2965(2021).

[0467] 4.Whitaker,K.Earlier diagnosis:the importance of cancer symptoms.TheLancet Oncology 21,6–8(2020).

[0468] 5.Reading,J.,Foster,K.,Joshi,K.&Chain,B.Tracking down tumor-specificT cells.Cancer Cell 40,351–353(2022).

[0469] 6.Wang,Q.et al.Genetic associations of T cell cancer immune response-related genes with T cell phenotypes and clinical outcomes of early-stagelung cancer.J Immunother Cancer 8,e000336(2020).

[0470] 7.Guo,G.et al.Immune cell concentrations among the primary tumormicroenvironment in colorectal cancer patients predicted by clinicopathologiccharacteristics and blood indexes.Journal for ImmunoTherapy of Cancer 7,1–13(2019).

[0471] 8.Wu,S.Y.,Fu,T.,Jiang,Y.Z.&Shao,Z.M.Natural killer cells in cancerbiology and therapy.Molecular Cancer 19,1–26(2020).

[0472] 14.Nakamura,K.&Smyth,M.J.Immunoediting of cancer metastasis by NKcells.Nat Cancer 1,670–671(2020).

[0473] 15.Mantovani,A.,Allavena,P.,Marchesi,F.&Garlanda,C.Macrophages astools and targets in cancer therapy.Nat Rev Drug Discov 21,799–820(2022).

[0474] 16.Cendrowicz,E.,Sas,Z.,Bremer,E.&Rygiel,T.P.The Role of Macrophagesin CancerDevelopment and Therapy.Cancers(Basel)13,1946(2021).

[0475] 17.Kumari,N.&Choi,S.H.Tumor-associated macrophages in cancer:recentadvancements in cancer nanoimmunotherapies.Journal of Experimental&Clinical CancerResearch 41,68(2022).

[0476] 18.Poh,A.R.&Ernst,M.Targeting Macrophages in Cancer:From Bench toBedside.Frontiers in Oncology 8,(2018).

[0477] 19. K.,Clevers,H.&Dost,A.F.The role of macrophages in non-small cell lungcancer and advancements in 3D co-cultures.eLife 12,e82998(2023).

[0478] 20.Condeelis,J.&Pollard,J.W.Macrophages:Obligate partners for tumorcell migration,invasion,and metastasis.Cell 124,263–266(2006).

[0479] 21.Conway,E.M.et al.Macrophages,inflammation,and lung cancer.AmericanJournal ofRespiratory and Critical Care Medicine 193,116–130(2016).

[0480] 24.Duan,Z.&Luo,Y.Targeting macrophages in cancer immunotherapy.SigTransductTarget Ther 6,1–21(2021).

[0481] 25.Dendritic cells in cancer immunology and immunotherapy|NatureReviews Immunology.https: / / www.nature.com / articles / s41577-019-0210-z.

[0482] 26.Del Prete,A.et al.Functional Role of Dendritic Cell Subsets inCancer Progression andClinical Implications.Int J Mol Sci 21,3930(2020).

[0483] 27.Veglia,F.&Gabrilovich,D.I.Dendritic cells in cancer:the rolerevisited.Curr OpinImmunol 45,43–51(2017).

[0484] 28.Lee,Y.S.&Radford,K.J.The role of dendritic cells in cancer.Int RevCell Mol Biol348,123–178(2019).

[0485] 29.Zanna,M.Y.et al.Review of Dendritic Cells,Their Role in ClinicalImmunology,andDistribution in Various Animal Species.Int J Mol Sci 22,8044(2021).

[0486] 30.Wang,H.et al.Metabolic labeling and targeted modulation ofdendritic cells.NatureMaterials(2020)doi:10.1038 / s41563-020-0680-1.

[0487] 31.Engelhard,V.et al.B cells and cancer.Cancer Cell 39,1293–1296(2021).

[0488] 32.Shang,J.,Zha,H.&Sun,Y.Phenotypes,Functions,and Clinical RelevanceofRegulatory B Cells in Cancer.Front Immunol 11,582657(2020).

[0489] 35.Hao,D.et al.The Single-Cell Immunogenomic Landscape of B andPlasma Cells inEarly-Stage Lung Adenocarcinoma.Cancer Discovery 12,2626–2645(2022).

[0490] 36.Wang,S.et al.Tumor-infiltrating B cells:their role and applicationin anti-tumorimmunity in lung cancer.Cell Mol Immunol 16,6–18(2019).

[0491] 37.Monzon,F.A.,Medeiros,F.,Lyons-Weiler,M.&Henner,W.D.Identificationof tissueof origin in carcinoma of unknown primary with a microarray-basedgene expression test.Diagnostic Pathology 5,3(2010).

[0492] 38.Nguyen,L.,Van Hoeck,A.&Cuppen,E.Machine learning-based tissue oforiginclassification for cancer of unknown primary diagnostics using genome-wide mutation features.Nat Commun 13,4013(2022).

[0493] 39.Miao,Y.et al.Identifying cancer tissue-of-origin by a novelmachine learning methodbased on expression quantitative trait loci.FrontOncol 12,946552(2022).

[0494] 40.Visvader,J.E.Cells of origin in cancer.Nature 469,314–322(2011).

[0495] 41.Xia,D.et al.Minimalist approaches to cancer tissue-of-originclassification by DNAmethylation.Modern Pathology 33,1874–1888(2020).

[0496] 42.Murata,M.Inflammation and cancer.Environmental Health andPreventive Medicine 23,50(2018).

[0497] 43.Treffers,L.W.,Hiemstra,I.H.,Kuijpers,T.W.,van den Berg,T.K.&Matlung,H.L.Neutrophils in cancer.Immunol Rev 273,312–328(2016).

[0498] 44.Mishalian,I.,Granot,Z.&Fridlender,Z.G.The diversity of circulatingneutrophils incancer.Immunobiology 222,82–88(2017).

[0499] 45.Uribe-Querol,E.&Rosales,C.Neutrophils in Cancer:Two Sides of theSame Coin.JImmunol Res 2015,983698(2015).

[0500] 46.Fridlender,Z.G.et al.Polarization of Tumor-Associated NeutrophilPhenotype byTGF-β:“N1”versus“N2”TAN.Cancer Cell 16,183–194(2009).

[0501] 47.Xiong,S.,Dong,L.&Cheng,L.Neutrophils in cancer carcinogenesis andmetastasis.JHematol Oncol 14,173(2021).

[0502] 48.Dissemond,J.et al.Activated neutrophils exert antitumor activityagainst humanmelanoma cells:reactive oxygen species-induced mechanisms andtheir modulation bygranulocyte-macrophage-colony-stimulating factor.J InvestDermatol 121,936–938(2003).49.Wang,T.et al.Tumour-activated neutrophils ingastric cancer foster immunesuppression and disease progression through GM-CSF-PD-L1 pathway.Gut 66,1900–1911(2017).

[0503] 50.Lei,X.et al.Immune cells within the tumor microenvironment:Biological functions androles in cancer immunotherapy.Cancer Letters 470,126–133(2020).

[0504] 51.Li,X.et al.Immune Cell Infiltration Landscape of Ovarian Cancer toIdentify Prognosisand Immunotherapy-Related Genes to AidImmunotherapy.Frontiers in Cell andDevelopmental Biology 9,(2021).

[0505] 52.Gentles,A.J.et al.The prognostic landscape of genes andinfiltrating immune cellsacross human cancers.Nature medicine 21,938–945(2015).

[0506] 53.Kumar,S.&Dikshit,M.Metabolic Insight of Neutrophils in Health andDisease.FrontImmunol 10,2099(2019).

[0507] 54.Li,Y.et al.The regulatory roles of neutrophils in adaptiveimmunity.CellCommunication and Signaling 17,147(2019).

[0508] 55.Zhao,H.et al.Inflammation and tumor progression:signaling pathwaysand targetedintervention.Sig Transduct Target Ther 6,1–46(2021).

[0509] 56.Nicolás- J. Adrover,J.M.&Hidalgo,A.Neutrophils inhomeostasis,immunity,and cancer.Immunity 46,15–28(2017).

[0510] 57.Ramoji,A.et al.Toward a spectroscopic hemogram:Ramanspectroscopicdifferentiation of the two most abundant leukocytes fromperipheral blood.Analyticalchemistry 84,5335–5342(2012).

[0511] 58.Pully,V.,Lenferink,A.T.&Otto,C.Time-lapse Raman imaging of singlelivelymphocytes.Journal of Raman spectroscopy 42,167–173(2011).

[0512] 59.Sijtsema,N.,Otto,C.,Segers-Nolten,G.M.,Verhoeven,A.&Greve,J.ResonanceRaman microspectroscopy of myeloperoxidase and cytochrome b558 inhuman neutrophilicgranulocytes.Biophysical journal 74,3250–3255(1998).

[0513] 60.Puppels,G.,Garritsen,H.,Kummer,J.&Greve,J.Carotenoids located inhumanlymphocyte subpopulations and natural killer cells by Ramanmicrospectroscopy.Cytometry:The Journal of the International Society forAnalytical Cytology 14,251–256(1993).

[0514] 61.van Manen,H.-J.,Kraan,Y.M.,Roos,D.&Otto,C.Intracellular chemicalimaging ofheme-containing enzymes involved in innate immunity using resonanceRaman microscopy.The Journal of Physical Chemistry B 108,18762–18771(2004).

[0515] 62.Baci,D.et al.The Ovarian Cancer Tumor Immune Microenvironment(TIME)as Targetfor Therapy:A Focus on Innate Immunity Cells as TherapeuticEffectors.Int J Mol Sci 21,3125(2020).

[0516] 63.Yajuk,O.et al.The PD-L1 / PD-1 Axis Blocks Neutrophil Cytotoxicityin Cancer.Cells 10,1510(2021).

[0517] 64.Zhang,S.et al.Berberine Maintains the Neutrophil N1 Phenotype toReverse CancerCell Resistance to Doxorubicin.Frontiers in Pharmacology 10,(2020).

[0518] 65.Stein,R.G.et al.Cognate Nonlytic Interactions between CD8+T Cellsand BreastCancer Cells Induce Cancer Stem Cell–like Properties.CancerResearch 79,1507–1519(2019).

[0519] 66.Surface-Enhanced Raman Spectroscopy of DNA|Journal of the AmericanChemicalSociety.https: / / pubs.acs.org / doi / 10.1021 / ja800023j.

[0520] 67.De Gelder,J.,De Gussem,K.,Vandenabeele,P.&Moens,L.Referencedatabase ofRaman spectra of biological molecules.Journal of RamanSpectroscopy 38,1133–1147(2007).

[0521] 68.Ghebeh,H.et al.FOXP3+Tregs and B7-H1+ / PD-1+T lymphocytes co-infiltrate thetumor tissues of high-risk breast cancer patients:Implicationfor immunotherapy.BMC Cancer8,57(2008).

[0522] 69.Bates,G.J.et al.Quantification of regulatory T cells enables theidentification ofhigh-risk breast cancer patients and those at risk of laterelapse.J Clin Oncol 24,5373–5380(2006).

[0523] 70.Abo-Elenein,A.,Elgohary,S.E.&Hashish,A.Significance ofImmunoregulatory TCells in Different Stages of Breast Cancer Patients.8.

[0524] 71.Leong PP,Mohammad R,Ibrahim N,Ithnin H,Abdullah M,Davis WC,Seow HF(2006).Phenotyping of lymphocytes expressing regulatory and effector markersin infiltrating ductalcarcinoma of the breast.Immnol Lett.102:229–36.-GoogleSearch.https: / / www.google.com / search?q=Leong+PP%2C+Mohammad+R%2C+Ibrahim+N%2C+Ithnin+H%2C+Abdullah+M%2C+Davis+WC%2C+Seow+HF(2006).+Phenotyping+of+lymphocytes+expressing+regulatory+and+effector+markers+in+infiltrating+ductal+carcinoma+of+the+breast.+Immnol+Lett.+102%3A229%E2%80%9336.&rlz=1C1CHBF_enCA899CA899&oq=Leong+PP%2C+Mohammad+R%2C+Ibrahim+N%2C+Ithnin+H%2C+Abdullah+M%2C+Davis+WC%2C+Seow+HF(2006).+Phenotyping+of+lymphocytes+expressing+regulatory+and+effector+markers+in+infiltrating+ductal+carcinoma+of+the+breast.+Immnol+Lett.+102%3A229%E2%80%9336.&aqs=chrome..69i57.284j0j7&sourceid=chrome&ie=UTF-8.

[0525] 72.Wang,Y.,Wei,Z.,Zhang,Y.&Chen,Y.Direct detection of DNA using 3Dsurfaceenhanced Raman scattering hotspot matrix.ELECTROPHORESIS 40,2104–2111(2019).

[0526] 73.Barhoumi,A.,Zhang,D.,Tam,F.&Halas,N.J.Surface-Enhanced RamanSpectroscopyof DNA.ACS Publications http: / / pubs.acs.org / doi / full / 10.1021 / ja800023j(2008)doi:10.1021 / ja800023j.

[0527] 74.Zhou,J.&Zhong,Y.Breast cancer immunotherapy.Cell Mol Immunol 1,247–255(2004).

[0528] 75.Ali,H.R.,Chlon,L.,Pharoah,P.D.P.,Markowetz,F.&Caldas,C.PatternsofImmuneInfiltration in Breast Cancer and Their Clinical Implications:A Gene-Expression-BasedRetrospective Study.PLOS Medicine 13,e1002194(2016).

[0529] 76.Wang,S.et al.Clinical significance and immunogenomic landscapeanalyses of theimmune cell signature based prognostic model for patients withbreast cancer.Brief Bioinform22,bbaa311(2021).

[0530] 77.Movasaghi,Z.,Rehman,S.&Rehman,I.U.Raman spectroscopy of biologicaltissues.Applied Spectroscopy Reviews 42,493–541(2007).

[0531] 78.Xing,X.,Du,S.&Wang,K.Robust Hessian Locally Linear EmbeddingTechniques forHigh-Dimensional Data.Algorithms 9,36(2016).

[0532] 79.Rivenbark,A.G.,O’Connor,S.M.&Coleman,W.B.Molecular andCellularHeterogeneity in Breast Cancer.Am J Pathol 183,1113–1124(2013).

[0533] 80.Godoy-Ortiz,A.,Alba-Bernal,A.,Pascual,J.,Comino-Méndez,I.&Alba,E.Unveilingthe Potential ofLiquid Biopsy in HER2-Positive Breast CancerManagement.Cancers 14,587(2022).

[0534] 81.Lopes,C.et al.HER2 Expression in Circulating Tumour Cells Isolatedfrom MetastaticBreast Cancer Patients Using a Size-Based MicrofluidicDevice.Cancers 13,4446(2021).82.Pupa,S.M.et al.HER2 Signaling and BreastCancer Stem Cells:The Bridge behindHER2-Positive Breast Cancer Aggressivenessand Therapy Refractoriness.Cancers(Basel)13,4778(2021).

[0535] 83.Li,X.et al.Uncovering the Subtype-Specific MolecularCharacteristics of Breast Cancerby Multiomics Analysis of Prognosis-Associated Genes,Driver Genes,Signaling Pathways,and Immune Activity.FrontCell Dev Biol 9,689028(2021).

[0536] 84.Fan,L.,Ma,L.-X.,Zhou,P.&Shao,Z.-M.Atlas of immune cellinfiltration in breastcancer—high M2 macrophage and low native Bcellproportions are associated with poorsurvival.Annals of Breast Surgery 6,(2022).

[0537] 88.Croker,A.k&Allan,A.l.Cancer stem cells:implications for theprogression andtreatment of metastatic disease.Journal of Cellular andMolecular Medicine 12,374–390(2008).

[0538] 89.Sheridan,C.et al.CD44+ / CD24-Breast cancer cells exhibit enhancedinvase properties:An early step necessary for metastasis.Breast CancerResearch 8,1–13(2006).

[0539] 90.Turner,N.H.&Di Leo,A.HER2 discordance between primary andmetastatic breastcancer:Assessing the clinical impact.Cancer TreatmentReviews 39,947–957(2013).

[0540] 91.Zimmer,A.S.,Van Swearingen,A.E.D.&Anders,C.K.HER2-positive breastcancerbrain metastasis:A new and exciting landscape.Cancer Rep(Hoboken)5,e1274(2022).

[0541] 92.Sammons,S.,Van Swearingen,A.E.D.,Chung,C.&Anders,C.K.Advances inthemanagement of breast cancer brain metastases.Neuro-Oncology Advances 3,v63–v74(2021).93.Stanton,S.E.&Disis,M.L.Clinical significance of tumor-infiltrating lymphocytes inbreast cancer.J Immunother Cancer 4,59(2016).

[0542] 94.Lin,D.et al.Label-free liquid biopsy based on blood circulatingDNA detection usingSERS-based nanotechnology for nasopharyngeal cancerscreening.Nanomedicine:Nanotechnology,Biology,and Medicine 22,102100(2019).

[0543] 95.Papachristou,N.et al.Network Analysis of the MultidimensionalSymptom Experienceof Oncology.Sci Rep 9,2258(2019).

[0544] 96.Bell,S.E.J.&Sirimuthu,N.M.S.Surface-Enhanced Raman Spectroscopy(SERS)forSub-Micromolar Detection of DNA / RNA Mononucleotides.Journal of theAmerican ChemicalSociety 128,15580–15581(2006).

[0545] 97.Ganesh,S.,Venkatakrishnan,K.&Tan,B.Quantum scale organicsemiconductors forSERS detection of DNA methylation and geneexpression.Nature Communications 11,(2020).98.Martínez-Sáez,O.&Prat,A.Currentand Future Management of HER2-PositiveMetastatic Breast Cancer.JCO OncologyPractice 17,594–604(2021).

[0546] 99.Fabi,A.et al.HER2 Protein and Gene Variation between Primary andMetastatic BreastCancer:Significance and Impact on Patient Care.ClinicalCancer Research 17,2055–2064(2011).

[0547] 100.Liu,M.et al.Classification using hierarchical clustering oftumor-infiltrating immunecells identifies poor prognostic ovarian cancerswith high levels of COX expression.ModPathol 22,373–384(2009).

[0548] 101.Wang,T.et al.Immunogenomic Landscape in Breast CancerRevealsImmunotherapeutically Relevant Gene Signatures.Front Immunol 13,805184(2022).

[0549] 102.Spranger,S.Mechanisms of tumor escape in the context of the T-cell-inflamed and thenon-T-cell-inflamed tumor microenvironment.InternationalImmunology 28,383–391(2016).103.Horton,B.L.et al.Lack of CD8+T cell effectordifferentiation during priming mediatescheckpoint blockade resistance in non–small cell lung cancer.Science Immunology 6,eabi8800.

[0550] 104.Yang,J.,Xu,J.,E,Y.&Sun,T.Predictive and prognostic value ofcirculating bloodlymphocyte subsets in metastatic breast cancer.Cancer Med 8,492–500(2019).

[0551] 105.DeNardo,D.G.et al.CD4+T Cells Regulate Pulmonary Metastasis ofMammaryCarcinomas by Enhancing Protumor Properties of Macrophages.Cancer Cell16,91–102(2009).

[0552] 106.Karavitis,J.et al.Regulation of COX2 Expression in Mouse MammaryTumor CellsControls Bone Metastasis and PGE2-Induction of Regulatory T CellMigration.PLOS ONE 7,e46342(2012).

[0553] 107.Janssen,L.M.E.,Ramsay,E.E.,Logsdon,C.D.&Overwijk,W.W.Theimmunesystem in cancer metastasis:friend or foe?Journal for ImmunoTherapy ofCancer 5,79(2017).

[0554] 108.Olkhanud,P.B.et al.Breast Cancer Lung Metastasis RequiresExpression of ChemokineReceptor CCR4 and Regulatory T Cells.Cancer Research69,5996–6004(2009).

[0555] 109.de Kruijf,E.M.et al.NKG2D ligand tumor expression and associationwith clinicaloutcome in early breast cancer patients:an observationalstudy.BMC Cancer 12,24(2012).110.Coffelt,S.B.et al.IL-17-producingγδT cellsand neutrophils conspire to promote breastcancer metastasis.Nature 522,345–348(2015).

[0556] 111.Kuhn,A.,Thu,D.,Waldvogel,H.J.,Faull,R.L.M.&Luthi-Carter,R.Population-specific expression analysis(PSEA)reveals molecular changes indiseased brain.Nat Methods 8,945–947(2011).

[0557] 112.Kang,K.et al.CDSeq:A novel complete deconvolution method fordissectingheterogeneous samples using gene expression data.PLOS ComputationalBiology 15,e1007510(2019).

[0558] 113.Hu,Y.et al.False-positive plasma genotyping due to clonalhematopoiesis.ClinicalCancer Research 24,4437–4443(2018).

[0559] 114.Freudenberg,J.A.et al.The role of HER2 in early breast cancermetastasis and theorigins of resistance to HER2-targeted therapies.Exp MolPathol 87,1–11(2009).

[0560] 115.Cardoso,F.et al.5th ESO-ESMO international consensus guidelinesfor advanced breastcancer(ABC 5).Annals of Oncology 31,1623–1649(2020).

Claims

1. A method for determining one or more cancer characteristics in a subject, the method comprising: a. measuring the value of one or more cancer characteristics of at least one tumor-associated immune cell (TAIC) or TAIC-derived biomarker in a fluid sample obtained from a subject to obtain a test profile; b. comparing the test spectrum with one or more reference spectra; c. wherein the similarity or difference between the test profile and the one or more reference profiles indicates one or more cancer characteristics; and optionally, d. providing a cancer assessment for the subject based on the one or more cancer characteristics.

2. The method of claim 1, wherein the cancer characteristic is: presence or absence of cancer, cancer type, cancer stage, cancer grade, primary cancer, metastatic cancer, cancer metastatic potential, molecular phenotype of cancer, benign cancer, malignant cancer, cancer origin tissue, cancer metastasis location, or a combination thereof.

3. The method according to claim 1 or 2, wherein the at least one tumor-associated immune cell comprises: a T cell, a NK cell, a macrophage, a neutrophil, a dendritic cell or a B cell, or a combination thereof.

4. The method of any one of claims 1-3, wherein the one or more TAIC-derived biomarkers comprise metabolic changes, activation state, or surface phenotype of the at least one tumor-associated immune cell.

5. The method of any one of claims 1-3, wherein the one or more TAIC-derived biomarkers comprise one or more extracellular vesicles from at least one tumor-associated immune cell.

6. The method of any one of claims 1-3, wherein the one or more TAIC-derived biomarkers comprise cell-free nucleic acid of the at least one tumor-associated immune cell.

7. The method of claim 6, wherein the cell-free nucleic acid is: cell-free DNA, miRNA, IncRNA, mRNA, or histone-associated DNA.

8. The method of claim 7, wherein the cell-free DNA is molecularly modified by one or more of methylation, oxidation, phosphorylation, and acetylation.

9. The method of claim 7, wherein the one or more TAIC-derived biomarkers comprise the structural and molecular composition of the cell-free DNA.

10. The method of any one of claims 1-3, wherein the one or more TAIC-derived biomarkers comprise the DNA methylation status of the at least one tumor-associated immune cell.

11. The method of any one of claims 1-10, wherein the one or more cancer traits are measured by Raman spectroscopy.

12. The method of any one of claims 1-3, wherein the one or more cancer traits are measured by DNA next generation sequencing, PCR, repertoire sequencing, immunosequencing, single cell RNA sequencing, DNA methylation sequencing, unique molecular identifiers (UMIs), light microscopy, immunohistochemistry, super-resolution microscopy, flow cytometry, or mass cytometry.

13. The method according to any one of claims 1-12, wherein the reference profile is derived from one or more reference TAICs.

14. The method of claim 13, wherein the reference TAIC is: a T cell, a NK cell, a macrophage, a neutrophil, a dendritic cell or a B cell, or a combination thereof.

15. The method of claim 13, wherein the reference TAIC is: a cancer cell-associated T cell, a cancer stem cell-associated T cell, or a tumor-associated stem cell-like T cell.

16. The method of claim 13, wherein the reference TAIC is: cancer cell-associated NK cells, cancer stem cell-associated NK cells, or tumor-associated stem cell-like NK cells.

17. The method of claim 13, wherein the reference TAIC is: a cancer cell-associated macrophage, a cancer stem cell-associated macrophage, or a tumor-associated stem cell-like macrophage.

18. The method of claim 13, wherein the reference TAIC is: cancer cell-associated neutrophils, cancer stem cell-associated neutrophils, or tumor-associated stem cell-like neutrophils.

19. The method of claim 13, wherein the reference TAIC is: a cancer cell-associated dendritic cell, a cancer stem cell-associated dendritic cell, or a tumor-associated stem cell-like dendritic cell.

20. The method of claim 13, wherein the reference TAIC is: a cancer cell-associated B cell, a cancer stem cell-associated B cell, or a tumor-associated stem-like B cell.

21. The method according to any one of claims 1-20, wherein the fluid is: plasma, serum, urine, feces, mucus or cerebrospinal fluid.

22. The method of any one of claims 1-21, wherein providing a cancer assessment comprises: Provides cancer type, cancer location, cancer stage, cancer grade, cancer metastasis potential, or cancer treatment effectiveness.

23. The method of any one of claims 1-21, wherein providing a cancer assessment comprises: Provide a prognosis for a patient, provide an early diagnosis of cancer, determine whether a tumor is benign or malignant, determine whether a tumor is primary or metastatic, determine whether a tumor has metastatic potential, determine the progression of cancer, determine lymph node metastasis of cancer, determine clinical metastasis of cancer, predict patient survival, provide a prognosis for a patient, determine the presence of an aggressive brain cancer, provide the location of a tumor, monitor for cancer recurrence during or after treatment, and / or determine the presence of minimal residual disease.

24. The method of claim 23, wherein when the tumor is benign, determining whether the tumor has malignant potential.

25. The method of any one of claims 1-24, wherein the cancer is a brain cancer, optionally a glioblastoma, astrocytoma, or oligodendroglioma.

26. The method of any one of claims 1 to 24, wherein the cancer is: ovarian cancer, brain cancer, bladder cancer, breast cancer, colon cancer, esophageal cancer, stomach cancer, liver cancer, intestinal cancer, lung cancer, head and neck cancer, rectal cancer, prostate cancer, pancreatic cancer, thyroid cancer, cervical cancer, melanoma, nasopharyngeal cancer, testicular cancer or uterine cancer.

27. The method of any one of claims 1-26, wherein the cancer is an adult cancer or a pediatric cancer.

28. A method of determining one or more cancer characteristics in a subject, the method comprising: a. separating a volume of liquid from a liquid sample of the patient, wherein the volume of liquid comprises at least one tumor-associated immune cell (TAIC) or a TAIC-derived biomarker; b. adding at least a portion of the volume of liquid to a nanosensor comprising nanoparticles configured to capture at least one TAIC or TAIC-derived biomarker and amplify a signal emitted by the at least one TAIC or TAIC-derived biomarker during Raman spectroscopy; c. performing Raman spectroscopy on a volume of liquid on the nanosensor to produce a sample Raman spectrum having an amplified signal, the amplified signal indicating the presence of at least one TAIC or TAIC-derived biomarker on the nanosensor; d. processing the sample Raman spectrum using data from a template Raman spectrum having a cancer signature to detect whether the sample contains one or more cancer signatures; and optionally e. providing a cancer assessment for the subject based on the detected one or more cancer characteristics.

29. The method of claim 28, wherein one or more cancer signatures of the sample are detected based on determining which correlation values ​​are greater than a correlation threshold, the correlation values ​​being obtained by correlating the amplified signal of the sample Raman spectrum with a template Raman spectrum from a known cancer sample having the cancer signature.

30. The method of claim 28, wherein one or more cancer characteristics of the sample are detected by: (i) performing feature extraction on the Raman sample spectrum data to extract feature values; (ii) performing classification by applying the feature values ​​to at least one set of classification models determined for at least one TAIC or TAIC-derived biomarker to detect one or more cancer signatures; and (iii) providing a cancer assessment by combining each detected cancer signature, Template Raman spectra from known cancer samples were used to determine the classification model.

31. The method of claim 30, wherein feature extraction is performed using principal component analysis, multivariate curve resolution analysis, or a combination thereof.

32. The method according to claim 30 or 31, wherein the classification model comprises: partial least squares discriminant analysis (PLSDA), support vector machine discriminant analysis (SVMDA), artificial neural network analysis (ANN) TSNE and random forest classification.

33. The method of any one of claims 28-32, wherein the cancer characteristic is: cancer type, cancer stage, cancer grade, cancer metastasis, cancer metastasis potential, or a combination thereof.

34. The method of any one of claims 28-33, wherein the at least one tumor-associated immune cell comprises: a T cell, a NK cell, a macrophage, a neutrophil, a dendritic cell, or a B cell, or a combination thereof.

35. The method of any one of claims 28-34, wherein the one or more TAIC-derived biomarkers comprise one or more extracellular vesicles from at least one tumor-associated immune cell.

36. The method of any one of claims 28-34, wherein the one or more TAIC-derived biomarkers comprise cell-free nucleic acid of at least one tumor-associated immune cell.

37. The method of claim 36, wherein the cell-free nucleic acid is: cell-free DNA, miRNA, IncRNA, mRNA, or histone-associated DNA.

38. The method of claim 37, wherein the cell-free DNA is molecularly modified by one or more of methylation, oxidation, phosphorylation, and acetylation.

39. The method of claim 37, wherein the one or more TAIC-derived biomarkers comprise the structural and molecular composition of cell-free DNA.

40. The method of any one of claims 28-34, wherein the one or more TAIC-derived biomarkers comprise DNA methylation of at least one tumor-associated immune cell.

41. The method of any one of claims 28-40, wherein the fluid is: plasma, serum, urine, feces, mucus, cerebrospinal fluid or other body fluid.

42. The method of any one of claims 28-41, wherein providing a cancer assessment comprises: Provides cancer type, cancer location, cancer stage, cancer grade, cancer metastasis potential, or cancer treatment effectiveness.

43. The method of any one of claims 28-42, wherein providing a cancer assessment comprises: Provide a prognosis for a patient, provide an early diagnosis of cancer, determine whether a tumor is benign or malignant, determine whether a tumor is primary or metastatic, determine whether a tumor has metastatic potential, determine the progression of cancer, determine lymph node metastasis of cancer, determine clinical metastasis of cancer, predict patient survival, provide a prognosis for a patient, determine the presence of an aggressive brain cancer, provide the location of a tumor, monitor for cancer recurrence during or after treatment, and / or determine the presence of minimal residual disease.

44. The method of claim 43, wherein when the tumor is benign, determining whether the tumor has malignant potential.

45. The method of any one of claims 28-44, wherein the cancer is a brain cancer, optionally a glioblastoma, astrocytoma, or oligodendroglioma.

46. ​​The method of any one of claims 28-45, wherein the cancer is: ovarian cancer, brain cancer, bladder cancer, breast cancer, colon cancer, esophageal cancer, stomach cancer, liver cancer, intestinal cancer, lung cancer, head and neck cancer, rectal cancer, prostate cancer, pancreatic cancer, thyroid cancer, cervical cancer, melanoma, nasopharyngeal cancer, testicular cancer or uterine cancer.

47. The method of claim 25, wherein the sample Raman spectrum includes a second amplified signal, the second amplified signal indicating the presence of a second biomarker on the nanosensor, and the method further comprises: (i) performing further data processing on the sample Raman spectrum to compare the second amplified signal with the second template Raman spectrum to determine the correlation between the sample Raman spectrum and the second template Raman spectrum, wherein the template Raman spectrum has a known cancer feature; as well as (ii) providing a diagnosis of cancer in the patient based on the correlation between the sample Raman spectrum and the template Raman spectrum, and a second correlation between the sample Raman spectrum and the second template Raman spectrum.

48. A method for generating reference tumor-associated immune cells (TAICs) for cancer assessment, the method comprising co-culturing one or more naive immune cells with a tumor and / or cancer cell line until the one or more reference tumor-associated immune cells (TAICs) express one or more cancer traits.

49. The method of claim 45, wherein the one or more primary immune cells comprise: T cells, NK cells, macrophages, neutrophils, or a combination thereof.

50. The method of claim 45 or 46, wherein the cancer assessment comprises: Provide the type of cancer, the location of the cancer, the stage of the cancer, the grade of the cancer, the metastatic potential of the cancer, or the effectiveness of treatment of the cancer.

Images