Cerebrospinal fluid protein construction precise prediction of non-small cell lung cancer meningeal metastasis risk model

Abstract

The application discloses a method for constructing a precise prediction non-small cell lung cancer meningeal metastasis risk model based on cerebrospinal fluid proteins and application, and relates to the technical field of biological medicine.Combining Olink detection technology, aiming at the problems of NSCLC-LM late discovery and diagnosis difficulty, 46 NSCLC-LM patients and 40 nLM patients cerebrospinal fluid samples are collected, among which 59 cases are used as a training set to construct a risk prediction model, and 27 cases are used as a validation set to judge the model efficiency.The results show that 7 specific proteins ADA, NOS3, KDR, TGFB1, TNFRSF12A, MMP7 and CASP8 are involved in model construction, the training set AUC is 0.94, the validation set AUC is 1.00, the training set sensitivity is 0.96, the training set specificity is 0.88, the validation set sensitivity is 1.00, and the validation set specificity is 0.79.The application has great significance in the field of NSCLC-LM early diagnosis and screening.

Description

Technical Field

[0001] This invention belongs to the field of biomedical technology, and in particular relates to a method and application for constructing an accurate model for predicting the risk of meningeal metastasis in non-small cell lung cancer based on cerebrospinal fluid proteins. Background Technology

[0002] Meningeal metastases (LM) are a serious complication of advanced non-small cell lung cancer (NSCLC), with an insidious onset and extremely poor prognosis; untreated NSCLC patients with LM have a survival of only 4-6 weeks. The incidence of LM in ordinary NSCLC patients is 3%-5%, reaching as high as 9.4% in patients with epidermal growth factor receptor (EGFR) mutations. However, autopsy series show that the incidence of LM is approximately 20% or even higher, indicating insufficient early recognition and diagnosis of LM.

[0003] Currently, clinical diagnosis of LM mainly relies on clinical symptoms, cerebrospinal fluid cytology, and enhanced magnetic resonance imaging. However, leptomeningeal biopsy is highly invasive, difficult to perform, and rarely used in clinical practice. In recent years, some new diagnostic techniques for LM have emerged.

[0004] Clinical symptoms: The symptoms of LM generally include the following: If the brain is involved, nausea, vomiting, headache, drowsiness, gait difficulty, and cognitive changes may occur. In later stages, seizures, dysphagia, and mental changes may occur. If the nerves and spinal cord are involved, specific cranial nerve palsy, visual disturbances, diplopia, dysarthria, hearing loss, radicular neck and back pain, and cauda equina syndrome may occur. LM also frequently causes hydrocephalus, leading to increased intracranial pressure. Clinical manifestations of increased intracranial pressure include headaches that are more pronounced when lying down and awake, slowed cognitive function, ataxia, decreased vision, urinary incontinence, drowsiness, nausea, and vomiting. Because the symptoms of LM are not specific, differentiating other neurological diseases is necessary to determine whether it is NSCLC-LM, which is one of the important reasons for the late detection of NSCLC-LM.

[0005] Imaging examinations: Enhanced magnetic resonance imaging (MRI) of the head is an important tool for diagnosing Leydig Leyen (LM). MRI findings of this condition include lesions of the pia mater, dura mater, subependymal or superficial brain, enhancement of cranial nerves, and communicating hydrocephalus. Literature indicates that approximately 20% of LM patients have negative cerebrospinal fluid cytology. In such cases, imaging results can still confirm the diagnosis, but the imaging examination must be performed before lumbar puncture to prevent false-positive pia mater enhancement due to secondary inflammation caused by the puncture. Compared to MRI, computed tomography (CT) has lower sensitivity and is generally not used for evaluating LM. It is important to note that imaging examinations cannot replace pathological results. Some patients may have positive cerebrospinal fluid cytology but show no significant changes on enhanced head MRI; the final diagnosis still requires consideration of cerebrospinal fluid cytology results.

[0006] Tumor markers: Traditional serum protein markers are not specific to LM, but they can be abnormally elevated in LM patients and can be used in combination for risk assessment. CYFRA21-1 (cytokeratin 19 fragment) has high sensitivity in cerebrospinal fluid, but low sensitivity in plasma (approximately 47%). Elevated plasma levels are often associated with active extracranial metastases, indirectly indicating LM risk, but its specificity is insufficient and needs to be confirmed by imaging or cerebrospinal fluid testing. Elevated CEA (carcinoembryonic antigen) indicates a high systemic metastatic burden, which may be caused by LM, but false positives can occur in smokers.

[0007] Cerebrospinal fluid (CSF) cytology analysis: A positive CSF cytology result is the gold standard for diagnosing tumor leukemia (LM). However, CSF needs to be obtained through lumbar puncture. Although this examination is simple, low-cost, and has few serious complications, it is an invasive procedure with low adoption and success rates, and patient comfort is poor. Literature indicates that the sensitivity of the first CSF analysis is only 50%, while the sensitivity of the second analysis can increase to 75%-85%, thus requiring repeated sampling when necessary. It is also recommended to collect at least 10 ml of CSF for each cytology test to increase the chances of capturing malignant tumor cells. Positive malignant cells, high protein content, low glucose concentration, and increased lymphocytes are characteristic CSF findings in LM patients. It is worth noting that even if CSF cytology is negative, a significant increase in CSF tumor markers cannot rule out LM. Therefore, CSF cytology analysis also faces challenges in specimen collection and is prone to false negative results.

[0008] Other diagnostic techniques: With technological advancements, novel diagnostic techniques such as liquid biopsy are being applied to the diagnosis of NSCLC-LM. As one of the most important molecular markers in tumor prediction, circulating cell-free DNA (cfDNA) and circulating tumor DNA (ctDNA) permeate all aspects of tumor screening, diagnosis, and treatment. Currently, there are no reliable markers for the diagnosis of LM, and it is easily affected by the blood-brain barrier. The level of ctDNA in the plasma of LM patients is often low. Compared to blood, cerebrospinal fluid cell-free DNA (cfDNA) exhibits unique genetic characteristics of LM and can more accurately identify sensitive genes and drug-resistant mutations. Therefore, the detection of cerebrospinal fluid cfDNA and ctDNA has advantages. Bu Hui et al. enrolled cerebrospinal fluid and blood samples from 35 NSCLC patients with LM for gene sequencing and found that ctDNA was more easily detected in cerebrospinal fluid (P = 0.003). Wu Yilong et al. found that the detection rate of driver genes in cerebrospinal fluid ctDNA of NSCLC-LM patients was 100%, significantly higher than other samples. ctDNA may be an effective aid in the diagnosis of LM, prediction of efficacy, and guidance of treatment. Although ctDNA technology has significant value in disease diagnosis and treatment monitoring, its application also faces bottlenecks. For example, in terms of sample selection, the low sensitivity of ctDNA in plasma necessitates the collection of cerebrospinal fluid samples via lumbar puncture for testing. Furthermore, ctDNA cannot technically pinpoint the source of mutations.

[0009] Emerging Technology: Addressing the challenges of insidious onset, late detection, and diagnostic difficulties in NSCLC-LM, the research team plans to utilize Olink proteomics technology to improve the sensitivity and specificity of initial cerebrospinal fluid (CSF) analyses by detecting specific proteins in the patient's CSF. This will accurately determine the risk of CSF involvement and facilitate early detection and treatment. Olink proteomics is a groundbreaking proteomics technology that has gradually become an important component of precision medicine over the past decade, alongside advancements in genomics. Olink utilizes its unique Proximity Extension Assay (PEA) patented technology to achieve ultrasensitive protein detection through dual antibody recognition and nucleic acid signal amplification. This enables high-throughput, low-volume, high-sensitivity, and wide dynamic range detection of protein biomarkers, particularly excelling in detecting low-abundance proteins in bodily fluid samples. Specifically, compared to traditional enzyme-linked immunosorbent assays (ELISA), Olink technology achieves fg / ml sensitivity (e.g., the detection limit for IL-8 is 30 fg / ml), a 1000-fold improvement, while ELISA sensitivity is only pg / ml, easily missing low-abundance proteins. Sample volume is reduced by 50-100 times; Olink requires only 1 µL (serum / plasma), while ELISA requires 50-100 µL per indicator. Furthermore, Olink technology has multiplexing capabilities, detecting 92-3072 proteins in a single assay, and exhibits high specificity and resistance to interference, while ELISA only detects a single target, has a high risk of cross-reactivity, and is susceptible to interference from enzyme activity or heterophile antibodies. Overall, Olink, through its PEA technology innovation, solves the three major bottlenecks of proteomics: low abundance, multiplexing, and sample limitations, providing fundamental support for precision medicine.

[0010] However, NSCLC-LM is a serious complication of advanced lung cancer, characterized by a low incidence and rapid disease progression. Furthermore, most patients do not receive timely medical intervention due to the rapid disease progression, making sample collection a significant challenge. Additionally, there are few reports on the application of Olink proteomics technology in NSCLC-LM. Summary of the Invention

[0011] To address the shortcomings of existing technologies, this invention combines Olink detection technology to tackle the late detection and diagnostic difficulties of NSCLC-LM. Cerebrospinal fluid samples were collected from 46 NSCLC-LM patients and 40 nLM patients. 59 samples were used as the training set to construct a risk prediction model (containing 32 LM cases and 27 nLM cases), while 27 samples were used as the validation set (containing 14 LM cases and 13 nLM cases) to evaluate model efficacy. Results showed that seven specific proteins were involved in model construction, and the area under the curve (AUC) of all cohorts was above 0.9 (training set AUC = 0.94, validation set AUC = 1.00). Both sensitivity and specificity reached high levels (training set sensitivity = 0.96, specificity = 0.88; validation set sensitivity = 1.00, specificity = 0.79). This invention has high application value in the early diagnosis and screening of NSCLC-LM.

[0012] To achieve the above objectives, the present invention adopts the following technical solution:

[0013] One objective of this invention is to provide a combination of protein biomarkers, which includes ADA, NOS3, KDR, TGFB1, TNFRSF12A, MMP7, and CASP8.

[0014] The second objective of this invention is to provide the application of the above-described protein biomarker combination in the preparation of diagnostic products for non-small cell lung cancer leptomeningeal metastasis, early screening products, disease monitoring products, and / or drug efficacy monitoring products.

[0015] A third objective of this invention is to provide a screening method for combinations of the protein biomarkers described above, wherein the screening method includes the use of Olink proteomics technology.

[0016] Furthermore, the sample source in the screening method includes cerebrospinal fluid.

[0017] The fourth objective of this invention is to provide a product for the diagnosis, early screening, disease monitoring, and / or drug efficacy monitoring of non-small cell lung cancer leptomeningeal metastasis, wherein the product contains reagents for detecting the combination of the protein biomarkers described above.

[0018] Furthermore, the product also contains one or more pharmaceutically acceptable carriers or excipients.

[0019] Furthermore, the product form includes a reagent kit.

[0020] The fifth objective of this invention is to provide a model for the diagnosis, early screening, disease monitoring, and / or drug efficacy monitoring of non-small cell lung cancer leptomeningeal metastasis, wherein the method for constructing the model includes using the combination of the protein biomarkers described above.

[0021] Furthermore, the model is based on Olink proteomics technology.

[0022] Furthermore, the model formula is as follows:

[0023] Risk-score = 2.9532 + 0.0708 × ADA − 0.0386 × NOS3 + 0.0331 × KDR− 0.1447 × TGFB1 − 0.31 × TNFRSF12A − 0.0762 × MMP7 − 0.2271 × CASP8.

[0024] Compared with the prior art, the present invention has the following technical effects:

[0025] 1. This invention utilizes Olink technology to construct a diagnostic model for LM (Leukemia lesions) for the first time using plasma proteins. Compared with currently used diagnostic techniques, it has the following advantages:

[0026] Comparing Clinical Symptoms: From Subjective Experience to Objective Molecular Early Warning. Clinical symptoms have limitations; early symptoms of liver tumors (LM) (headache, cranial nerve palsy) are not specific and are easily confused with brain parenchymal metastases or paraneoplastic syndromes, with a misdiagnosis rate as high as 40%. Diagnostic models, on the other hand, can provide objective and accurate molecular numerical assessments of the potential risk of LM in the early stages of symptoms, increasing doctors' diagnostic confidence, helping patients to detect and treat the disease early, and avoiding repeated biopsies.

[0027] Comparative Study with Enhanced Cranial MRI: Overcoming Sensitivity and Micrometastasis Detection Bottlenecks. MRI has a sensitivity of only 70%-80% in diagnosing leptomeningeal metastasis (LM), and is insensitive to microfocal LM or diffuse leptomeningeal enhancement, leading to a high false-negative rate and creating a diagnostic bottleneck. Diagnostic models, however, can detect specific molecular proteins in cerebrospinal fluid, reflecting the intracranial microenvironment and revealing the potential presence of LM from a microscopic molecular perspective.

[0028] Compared with tumor markers: the sensitivity has been greatly improved. Traditional tumor markers only provide information on tumor burden and have no specificity for the diagnosis of LM. Moreover, their detection sensitivity is low. The sensitivity of plasma CYFRA21-1 for LM is only 47%, which is far lower than the sensitivity of diagnostic models.

[0029] Compared to cerebrospinal fluid cytology: the sensitivity is significantly higher than the gold standard. The positive rate of cerebrospinal fluid cytology on the first lumbar puncture is only 50%, often requiring multiple punctures to improve the positive rate. However, multiple punctures increase patient suffering, infection risk, and distrust. The diagnostic model only requires 1 ml of cerebrospinal fluid sample to determine the likelihood of LM. Although it cannot currently replace cerebrospinal fluid cytology as the gold standard for diagnosing LM, the combined diagnostic approach of both helps patients reduce the pain of repeated punctures.

[0030] Compared to liquid biopsy technologies such as ctDNA testing: supplementing the functional blind spots of genomics. ctDNA testing is expensive, requires at least 10 ml of cerebrospinal fluid, and its dynamic range often only covers high-abundance mutations, providing only a mutation spectrum. In contrast, diagnostic models are not only cheaper and require smaller sample sizes, but their detection range can cover low-abundance functional proteins and can also reflect the state of the tumor microenvironment, effectively supplementing the detection blind spots of genomics.

[0031] Overall, the advantages of cerebrospinal fluid models over traditional diagnostic methods and novel diagnostic techniques are high sensitivity, high accuracy, low sample requirements, and relatively low cost.

[0032] For similar detection technologies, the use of Olink technology in diagnostic models offers a significant improvement over ELISA for protein detection. It has higher sensitivity, requires less sample, can detect more proteins, has a wider dynamic range, and better detection specificity. It represents an iteration of ELISA and also helps upgrade clinical diagnostic models. Olink's underlying technological advantages have established its core position in the field of translational medicine.

[0033] 2. This invention, combining Olink detection technology, addresses the challenges of late detection and difficult diagnosis in NSCLC-LM. Cerebrospinal fluid samples were collected from 46 NSCLC-LM patients and 40 nLM patients. 59 samples were used as the training set to construct a risk prediction model (containing 32 LM cases and 27 nLM cases), while 27 samples were used as the validation set (containing 14 LM cases and 13 nLM cases) to evaluate model efficacy. Results showed that seven specific proteins were involved in model construction, and the area under the curve (AUC) of the cohort was above 0.9 (training set AUC = 0.94, validation set AUC = 1.00). Both sensitivity and specificity reached high levels (training set sensitivity = 0.96, specificity = 0.88; validation set sensitivity = 1.00, specificity = 0.79).

[0034] In summary, compared with existing technologies, this invention uses Olink to detect specific proteins in cerebrospinal fluid to aid in the diagnosis of leptomeningeal metastases in non-small cell lung cancer. Compared with commonly used pathological diagnoses in clinical practice, this diagnostic model significantly improves the success rate of diagnosis, avoids secondary biopsies, and reduces diagnostic time. Compared with commonly used ELISA technology, the Olink technology used in this diagnostic model can detect multiple targets simultaneously, increasing detection sensitivity by 1000 times. It requires smaller sample sizes, is less affected by non-specific binding interference, and has high result stability. Using a control trial as a comparison, the sensitivity of the initial cerebrospinal fluid cytology (gold standard) diagnosis was only 50%, while the sensitivity of this invention's model increased to 96%, achieving significant progress. Attached Figure Description

[0035] Figure 1 This is the LASSO-featured contraction path diagram in Embodiment 1 of the present invention;

[0036] Figure 2 This is the LASSO-Cross-Validation Error Plot in Embodiment 1 of the present invention;

[0037] Figure 3 The image shows the ROC curve of the cerebrospinal fluid protein prediction model in Embodiment 1 of the present invention. Specifically, it shows the diagnostic performance of the logistic regression model constructed based on seven key proteins (ADA, NOS3, KDR, TGFB1, TNFRSF12A, MMP7 and CASP8) on the training set and validation set.

[0038] Figure 4 This is a four-part table of the cerebrospinal fluid protein prediction model in Embodiment 1 of the present invention, specifically used to evaluate the sensitivity and specificity of the model on the training and validation sets. Detailed Implementation

[0039] The following examples are used to illustrate the present invention, but are not intended to limit the scope of the invention. Any modifications or substitutions made to the methods, steps, or conditions of the present invention without departing from the spirit and essence of the invention are within the scope of the invention. The reagents, products, and instruments used in the following examples are all commercially available, and the methods used in the examples, unless otherwise specified, are consistent with conventional methods.

[0040] The technical solution of the present invention will be further described in detail below with reference to the embodiments.

[0041] Example 1

[0042] I. Sample Collection

[0043] 1. Sample type: Cerebrospinal fluid

[0044] 2. Sample collection: (1) Collect at least 1 ml of cerebrospinal fluid into a polypropylene collection tube (store on ice or at 4°C for short periods). No protease inhibitors need to be added to the sample. (2) Centrifuge at ≥500×g for 10 minutes at 2-8°C to remove cells and insoluble substances. (3) Immediately transfer the supernatant to a clean tube. Samples tested or aliquoted on the same day should be stored at -80°C (multiple samples taken from the same site should be mixed together before aliquoting).

[0045] 3. Sample size: 46 cases in the leptomeningeal metastasis group (LM) and 40 cases in the non-leptomeningeal metastasis group, for a total of 86 cases. According to the collection time order, the first 59 cases were used as the training set to build the risk model, and the last 27 cases were used as the validation set to validate the model (NSCLC-LM is one of the serious complications of advanced lung cancer, characterized by a low incidence and rapid disease progression. Moreover, most patients do not receive timely medical intervention due to the rapid progression of the disease, which makes sample collection a great challenge).

[0046] II. Panel Detection

[0047] The Immuno-Oncology Panel was used, which included 92 immune checkpoint proteins (PD-L1, CTLA4, etc., as shown in Table 1 below).

[0048] Table 1. Details of 92 Immune Checkpoint Proteins

[0049]

[0050] III. Sample Testing

[0051] 1. Sample randomization: Use the RAND function in an Excel spreadsheet to output the value corresponding to each sample, and then arrange the values ​​in ascending order to randomly distribute the samples in the 96-well plate.

[0052] 2. Sample Sampling: According to the randomized sample distribution table, take 10-40 µL of each sample into a 96-well plate. External Control: Add 5 µL of mixed plasma sample to wells A12 and B12, add 5 µL of Negative Control (NC) to wells C12, D12 and E12, and add 5 µL of Interplate Control (IPC) to wells F12, G12 and H12.

[0053] 3. Hybridization Incubation: First, prepare the hybridization mixture. Thaw the reagents at room temperature, vortex to mix, and centrifuge. Take a 1.5ml centrifuge tube without nuclease and add 280 µL of Incubation Solution, 40 µL of Incubation Stabilizer, 40 µL of A-probes, and 40 µL of B-probes sequentially. Vortex to mix and centrifuge. Take an 8-tube strip and add 47 µL of hybridization mixture to each well. Take a 96-well plate, label it the incubation plate, and use an 8-channel pipette to add 3 µL of hybridization mixture to each well. Then, add 1 µL of sample to each well. After sealing with film, centrifuge at 400g for 1 minute, place it in a PCR instrument, set it to 4℃ with infinite pressure, and record the time it takes to be placed in the PCR instrument. The incubation time should be 16-24 hours.

[0054] 4. Extension and Amplification: Set the PCR program to 50℃ for 20 minutes; 95℃ for 5 minutes; 95℃ for 30 seconds, 54℃ for 1 minute, and 60℃ for 1 minute for 17 cycles; hold at 10℃. After setting, run the PCR instrument, then press the pause button to maintain the PCR instrument temperature at 50℃. Prepare the extension and amplification mixture: Take a 15 ml centrifuge tube and add 9385 µL of High Purity Water, 1100 µL of PEA Solution, 55 µL of PEA Enzyme, and 22 µL of PCR Polymerase in sequence. Vortex to mix and centrifuge. Remove the incubation plate from the PCR instrument and centrifuge. Using an 8-channel pipette, add 96 µL of the extension and amplification mixture to each well. Seal the tube, vortex to mix, centrifuge, and place it on a preheated PCR instrument at 50℃. Then resume the PCR program.

[0055] 5. Chip Pre-processing: Unpack the chip, remove it, and check the chip and syringe for damage. Remove the syringe, press down on the plunger hole, and inject the liquid into the syringe. Click the instrument's Target 96 option. After the instrument pushes out the sample stage, place the chip and cover plate in it. After the instrument automatically retracts into the sample stage, click Start to begin chip pre-processing.

[0056] 6. Execution Procedure: Take one 1.5 ml centrifuge tube and add 550 µL of Detection Solution, 230 µL of High Purity Water, 7.8 µL of Detection Enzyme, and 3.1 µL of PCR Polymerase sequentially. Vortex to mix and centrifuge. Then, take another tube and aliquot 95 µL into each well. Take one 96-well plate and add 7.2 µL of the mixture to each well. After the extension amplification program is complete, add 2.8 µL of product, vortex to mix, and set aside. Prepare the diluted sample plate, primer plate, and pre-treated chip. Add 5 µL of primer to the left side of the chip and 5 µL of diluted product to the right side, avoiding air bubbles. After adding samples, place the plate on the sample loading stage, set the run name, and click "Run".

[0057] 7. Data Analysis and Export: After the instrument is shut down, export the run name file with the extension "XXX.q100". Open the Olink NPX Signature 1.5.3.0 software, import "XXX.q100", set the reagent panel information, reagent version number, and sample name, and click "OK". The software will automatically analyze the data and output an NPX Excel document and a QC AR PDF document.

[0058] IV. Data Analysis

[0059] A strategy combining ensemble feature selection and logistic regression modeling was employed for data analysis. The specific analysis process is as follows:

[0060] 1. Data preprocessing: All samples were de-batch-effects and standardized before modeling.

[0061] 2. Input training set and validation set: Input the training set data obtained from the detection to construct the model, and input the validation set data to validate the model.

[0062] 3. LASSO Model Construction and Parameter Optimization:

[0063] In each training set, the LASSO logistic regression method is used for modeling. Five-fold cross-validation (5-fold CV) is employed to select the lambda value that minimizes the cross-validation error. This lambda value is then used to train the LASSO model across the entire training set, and applied to the validation set for predictions. The predicted probabilities are calculated, and the AUC value on the validation set is evaluated.

[0064] 4. Model stability assessment and feature selection:

[0065] Genes selected by the LASSO method in each round of modeling were counted, and their frequency of occurrence was summarized. Genes that appeared more than 50 times (i.e., >50%) in 100 modeling iterations were considered to have high stability and predictive contribution.

[0066] 5. Final Model Construction and Evaluation:

[0067] Using these stable trait genes as input variables, a multivariate logistic regression model was reconstructed across all samples to form the final risk score formula. Finally, based on this model, the predicted probability for each sample was calculated, an ROC curve was plotted, and the AUC value was calculated to evaluate the overall predictive performance of the final model.

[0068] V. Results Presentation

[0069] 1. Software and version information is shown in Table 2 below.

[0070] Table 2 Software and Version Information

[0071]

[0072] 2. Molecular models based on cerebrospinal fluid proteomics can accurately predict the risk of leptomeningeal metastasis in NSCLC patients.

[0073] This study used the Olink platform to perform proteomic analysis on cerebrospinal fluid samples from non-small cell lung cancer (NSCLC) patients, constructing a highly sensitive molecular diagnostic model for predicting the risk of leptomeningeal metastasis (LM). Modeling and evaluation were conducted using LASSO feature screening combined with logistic regression.

[0074] 3. Iterative construction of the LASSO model and screening of characteristic proteins.

[0075] The training set (containing 32 LM and 27 nLM examples) was modeled 100 times to identify the set of genes selected in each modeling iteration. In each iteration, features were screened using the LASSO method, selecting genes with a frequency >50% and building predictive models accordingly. Five-fold cross-validation was used to determine the optimal regularization parameter lambda (i.e., the lambda value with the minimum cross-validation error), and the AUC value of the model was calculated on the corresponding validation set to evaluate model performance. In the 100 modeling iterations, the frequency of each protein being selected as a model feature was counted to assess its stability under different data partitions. According to the LASSO results, ADA, NOS3, KDR, TGFB1, TNFRSF12A, MMP7, and CASP8 were selected in over 50% of the modeling iterations, indicating high stability and predictive contribution in the model, and were ultimately included in the model to construct the final predictive formula. The protein screening process is as follows: Figures 1-2 As shown, LASSO regression achieves feature selection and model shrinkage, ultimately retaining 7 key protein features to construct a prediction model.

[0076] 4. Key proteins possess a defined or hypothesized biological functional basis.

[0077] The seven key proteins ultimately included in the model—ADA, NOS3, KDR, TGFB1, TNFRSF12A, MMP7, and CASP8—form multidimensional associations in tumor angiogenesis, immune regulation, invasion and metastasis, and apoptosis, jointly influencing tumor occurrence, development, and immune escape, supporting their feasibility as biomarkers related to leptomeningeal metastasis.

[0078] ADA's main function is to catalyze the degradation of adenosine. The accumulation of adenosine in the tumor microenvironment can inhibit the activity of immune cells such as T cells. ADA can restore the immune-killing ability by degrading adenosine, making it a key enzyme in maintaining the balance of the immune system and directly regulating the activation and function of immune cells.

[0079] NOS3: Its main function is to produce nitric oxide (NO) in vascular endothelial cells, regulating vasodilation. NO has a "two-way" effect: low concentrations of NO can promote tumor angiogenesis and enhance tumor cell invasiveness; high concentrations of NO can inhibit tumor cell proliferation and even induce their death.

[0080] KDR is the main receptor for vascular endothelial growth factor (VEGF) and mediates angiogenesis signaling. VEGF binding to KDR strongly promotes tumor angiogenesis ("tumor angiogenesis switch"), making it a key target for the progression of advanced tumors (such as lung cancer and colorectal cancer); some tumor cells also abnormally express KDR, directly promoting their own proliferation.

[0081] TGFB1 is a multifunctional cytokine that regulates cell proliferation, differentiation, and tissue repair. It is a potent immunosuppressive factor that can inhibit the killing ability of T cells and NK cells (natural killer cells), and can also induce the generation of regulatory T cells (Tregs, which suppress immune responses), thereby weakening the body's immune surveillance against tumors.

[0082] TNFRSF12A: Also known as the "TWEAK receptor," it transmits signals after binding to the ligand TWEAK. It is highly expressed in various tumors (such as liver cancer and colorectal cancer). After binding to TWEAK, it can promote tumor cell proliferation, invasion, and angiogenesis, and can also enhance tumor resistance to chemotherapy.

[0083] MMP7: Its main function is to degrade the extracellular matrix (such as collagen and adhesion molecules). It is a "key tool" for tumor invasion and metastasis. By degrading the extracellular matrix, it helps tumor cells break through tissue barriers (such as the basement membrane) and enter the blood / lymphatic system for metastasis. It can also promote tumor angiogenesis and activate oncogenes. It is highly expressed in various tumors such as gastric cancer and lung cancer and is associated with poor prognosis.

[0084] CASP8 is a key initiator of the apoptosis (programmed cell death) pathway. It plays a core role in tumor suppression. Under normal circumstances, CASP8 can induce cancer cell death through the "exogenous apoptosis pathway." If CASP8 is mutated or missing, tumor cells can escape immune-mediated apoptosis, leading to tumorigenesis.

[0085] In summary, the seven proteins mentioned above work together in terms of biological function, with TGFB1 as the core hub, and form a functional network of "synergistic cancer promotion" and "antagonistic balance" around the four key biological processes of tumors: "tumor angiogenesis (KDR / NOS3), invasion and metastasis (MMP7 / TNFRSF12A), immune regulation (ADA / NOS3), and apoptosis (CASP8)". This provides important clues for the early identification and potential mechanisms of tumor lesions (LM).

[0086] 5. Model Building and Performance Evaluation

[0087] Based on the above screening results, we constructed a final logistic regression model including seven protein indicators (ADA, NOS3, KDR, TGFB1, TNFRSF12A, MMP7, and CASP8). The model formula is as follows:

[0088] Risk-score = 2.9532 + 0.0708 × ADA − 0.0386 × NOS3 + 0.0331 × KDR− 0.1447 × TGFB1 − 0.31 × TNFRSF12A − 0.0762 × MMP7 − 0.2271 × CASP8

[0089] Here, risk-score represents the risk score for predicting LM based on the olink protein detection results of plasma samples, and the coefficients represent the direction and magnitude of the influence of each protein expression level on the risk score. After constructing the final model, we further calculate the ROC curve based on the continuous risk score (score) output by logistic regression, as shown below. Figure 3 As shown in Figure A, the area under the ROC curve (AUC) was 0.943. The optimal cutoff value was determined using Youden's Index (sensitivity + specificity − 1) to achieve the best classification effect between LM and non-LM (nLM) patients. The final results showed that the optimal cutoff value was -0.289. At this threshold: Figure 4 As shown in Figure A, the sensitivity is 0.96, meaning the model can correctly identify 96% of LM patients; the specificity is 0.88, meaning the model can correctly exclude 88% of non-LM patients.

[0090] 6. Further verification

[0091] We further validated the risk model on the validation set (containing 14 LM cases and 13 nLM cases), and the results are as follows. Figure 3 B and Figure 4 As shown in Figure B. The results show that the risk model has an AUC of 1.00, a sensitivity of 1.00, and a specificity of 0.79 on the validation set. Overall, this suggests that the model has high accuracy and stability in distinguishing between patients with leptomeningeal metastases (LM) and those without (nLM).

[0092] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A combination of protein markers, characterized in that, The protein marker combination consists of ADA, NOS3, KDR, TGFB1, TNFRSF12A, MMP7 and CASP8.

2. Use of the protein marker combination of claim 1 in the preparation of a non-small cell lung cancer meningeal metastasis diagnosis product, early screening product and / or disease monitoring product.

3. A screening method of the protein marker combination according to claim 1, characterized by, The screening method comprises using Olink proteomics technology.

4. The screening method according to claim 3, characterized in that, The sample source in the screening method comprises cerebrospinal fluid.

5. A product for the diagnosis, early screening and / or monitoring of the condition of meningeal metastases of non-small cell lung cancer, characterized in that, The product comprises a reagent for detecting the protein marker combination of claim 1.

6. The product of claim 5, wherein, The product further comprises one or more pharmaceutically acceptable carriers or adjuvants.

7. The product of claim 6, wherein, The product form comprises a kit.

8. A method for constructing a model for the diagnosis, early screening and / or monitoring of the condition of non-small cell lung cancer meningeal metastasis, characterized by, The model construction method comprises using the protein marker combination of claim 1.

9. The method of constructing a model according to claim 8, wherein, The model construction method is based on Olink proteomics technology.

Images