Stomach cancer early-stage multi-molecular diagnosis model construction method based on plasma protein

Abstract

The invention discloses a plasma protein-based stomach cancer early-stage multi-molecular diagnosis model construction method, and aims to solve the problems that the existing stomach cancer early-stage diagnosis standard and minimally invasive or non-invasive method are imperfect, the sensitivity and specificity of the existing diagnosis molecular marker are low and the like, an Olink proteomics PEA technology is utilized and a specific detection panel is selected to construct a stomach cancer early-stage multi-molecular diagnosis model. And identifying the protein markers of gastric cancer, early gastric cancer and gastric precancerous lesions in a discovery queue (n = 88). In a discovery queue, a group of multi-protein molecular marker models consisting of 13 proteins are successfully identified by utilizing an LASSO regression model, and the multi-protein molecular marker models have good diagnostic performance for distinguishing gastric cancer, early gastric cancer and pregastric lesion and contrast in the discovery queue and verification pair. Furthermore, in a verification queue sample, the diagnosis model also shows good diagnosis performance, and the marker model is expected to become a powerful tool for early diagnosis of gastric cancer.

Description

Technical Field

[0001] The present invention relates to the field of gastric cancer diagnosis research, and in particular to a method for constructing an early multi-molecular diagnosis model for gastric cancer based on plasma proteins. Background Art

[0002] Gastric cancer (GC) is a major global health problem, with morbidity and mortality ranking fifth and fourth among cancer cases worldwide, respectively.

[0003] Most patients with early gastric cancer have no obvious symptoms and the existing early screening system is immature, resulting in a low early detection rate of gastric cancer in my country. Most patients are already in the middle and late stages when they are first diagnosed, missing the best treatment period, resulting in a greatly reduced survival rate, with a 5-year survival rate of less than 50%. If gastric cancer patients can be discovered, diagnosed and treated early, the survival rate of patients can be significantly increased to more than 90%. Therefore, early screening for gastric cancer has important clinical significance.

[0004] At present, endoscopic biopsy and serological screening are the two most commonly used techniques for early screening of gastric cancer. Although endoscopic biopsy is the gold standard for early GC diagnosis, its invasiveness, expensive nature, and diagnostic accuracy are highly dependent on the skills of endoscopists, which hinders its application in large-scale screening of early GC. Currently, serum carcinoembryonic antigen (CEA), carbohydrate antigen 19-9 (CA19-9), and carbohydrate antigen 72-4 (CA72-4) are still the most commonly used biomarkers in clinical applications. However, CEA and CA19-9 are widely present in various cancer types other than gastric cancer, with low detection sensitivity, and limited research on CA72-4, which affects their application in the diagnosis and screening of early gastric cancer. Therefore, there is an urgent need to develop new diagnostic markers for gastric cancer.

[0005] Individual plasma proteomes exhibit unique and stable protein profiles over a period of up to 2 years and have strong correlations with clinical chemistry, whereas other omics, such as transcriptomics, lipidomics, and metabolomics, vary greatly over time and have poor correlations with clinical chemistry. In addition, proteins ultimately play biological roles in the human body and are more closely related to disease phenotypes than nucleic acid molecules. Therefore, it is reasonable to use plasma proteomics to predict and detect various human diseases in the clinic. However, the wide range of protein concentrations in blood (>10 orders of magnitude) and the difficulty in capturing low-abundance disease-associated proteins pose challenges to proteomic diagnosis. Although most current GC proteomics studies are from mass spectrometry, the difficulty in identifying low-abundance proteins (ng / ml-fg / ml) and the large sample input limit its further application. Multiplexed proximity extension assays (PEA, Olink Proteomics) combine antibody-based immunoassays with quantitative PCR or next-generation sequencing to accurately detect thousands of proteins in as little as a few microliters of body fluid with high specificity, sensitivity, and a wide detection range (mg / ml-fg / ml). PEA technology has been widely used to find multiple biomarker groups and potential drug targets in various diseases. Therefore, the use of PEA technology will help develop gastric cancer diagnostic markers with higher performance and lower input.

[0006] The existing molecular markers, diagnostic criteria, and minimally invasive or non-invasive methods for early diagnosis of gastric cancer are imperfect. Conventional gastroscopy is highly invasive, expensive, and the accuracy of diagnosis is highly dependent on the skills of the endoscopist. In addition, the sensitivity and specificity of conventional gastric cancer screening markers are low. In addition, the common proteomics technology mass spectrometry is difficult to identify low-abundance proteins (ng / ml-fg / ml) and requires a large amount of sample input. Summary of the invention

[0007] The purpose of the present invention is to provide a method for constructing a plasma protein-based multi-molecular diagnostic model for early gastric cancer in response to the imperfections of existing early gastric cancer diagnostic standards and minimally invasive or non-invasive methods, as well as the low sensitivity and specificity of existing diagnostic molecular markers.

[0008] The objective of the present invention is achieved through the following technical solutions:

[0009] A method for constructing an early multi-molecular diagnostic model for gastric cancer based on plasma proteins, comprising the following steps:

[0010] S1. Collect whole blood, centrifuge to obtain plasma, freeze it in liquid nitrogen, store it at -80℃, and remove hemolyzed samples;

[0011] S2. Use Olink plasma proteomics PEA technology to detect plasma samples and obtain standardized protein expression values;

[0012] S3. Screening found differentially expressed proteins in gastric cancer and early gastric cancer in the cohort, and selected 26 proteins with significant differences between early GC (HGIN-I stage) and GC (I-IV stage) samples and normal control groups as input proteins for model construction;

[0013] S4. Using the LASSO regression model and cross-validation optimization model, a diagnostic model composed of 13 proteins was obtained, and its diagnostic performance in the training set and test set of the discovery cohort was evaluated. The model was applied to the validation cohort to evaluate its diagnostic performance in distinguishing gastric cancer, early gastric cancer, gastric precancerous lesions and controls.

[0014] Furthermore, the step S2 is specifically as follows: the collected plasma samples are detected using the Olink plasma proteomics PEA technology, 1 ml of plasma is mixed with 369 pairs of antibodies labeled with different DNA oligonucleotides, after the antibodies bind to the target protein, the oligonucleotides are extended, and quantified into a standardized protein expression (NPX) value using next generation sequencing.

[0015] Furthermore, step S3 is specifically as follows: in the discovery cohort, differentially expressed proteins in gastric cancer and early gastric cancer are screened by differential analysis, z-score normalization is used, and 26 proteins with significant differential protein intersections between early GC (HGIN-I stage) and GC (I-IV stage) samples and the normal group are selected as input proteins for model construction.

[0016] Furthermore, the step S4 is specifically as follows: using the LASSO regression model to examine the performance and stability values ​​of the input protein, fitting the model in the training set (90% samples), verifying it in the test set (10% samples), training the penalty parameter using the ten-fold cross-validation method, performing cross-validation, taking the median of the penalty parameter as the penalty parameter of the final model, and obtaining a molecular diagnostic model containing 13 proteins.

[0017] The beneficial effects of the present invention are:

[0018] (1) From the perspective of liquid biopsy, the use of a small amount of plasma samples for testing greatly reduces the invasiveness and harm of current conventional endoscopic testing methods to patients;

[0019] (2) Relying on Olink proteomics PEA technology to search for diagnostic markers for gastric cancer, this technology can accurately detect thousands of proteins in a few microliters of blood samples. The concentration of the detected target protein spans 10 orders of magnitude, and the lowest detection concentration can reach fg / ml. Compared with mass spectrometry technology, it has higher sensitivity and specificity, and lower sample input;

[0020] (3) The multi-molecular markers for gastric cancer diagnosis found by this method and the diagnostic model constructed have good diagnostic performance for patients with zoned gastric cancer, early gastric cancer, gastric precancerous lesions and controls. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is a flowchart of the steps of the embodiment of the method;

[0022] Figure 2 Schematic diagram of the 26 proteins intersected after comparing the early GC (HGIN-I stage) samples and GC (I-IV stage) samples with the normal group;

[0023] Figure 3 The model distinguishes gastric cancer (stage I-IV) from the control in the training set and test set;

[0024] Figure 4 This is a data comparison chart for the discovery cohort and validation cohort for distinguishing gastric cancer (stages I-IV) based on the model;

[0025] Figure 5 This is a data comparison chart of the discovery cohort and validation cohort for distinguishing early gastric cancer (HGIN-I) based on the model;

[0026] Figure 6 This is a data comparison chart of the discovery cohort and validation cohort for distinguishing gastric precancerous lesions (LGIN-HGIN) based on the model;

[0027] Figure 7 Schematic diagram of plasma protein levels and GC pathological progression response dose for all samples in the discovery cohort and validation cohort. DETAILED DESCRIPTION

[0028] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0029] This embodiment proposes a method for constructing an early multi-molecular diagnostic model for gastric cancer based on plasma proteins, the steps comprising: S1. collecting whole blood, centrifuging to obtain plasma, freezing it in liquid nitrogen, storing it at -80°C, and removing hemolyzed samples;

[0030] S2. Use Olink plasma proteomics PEA technology to detect plasma samples and obtain standardized protein expression values;

[0031] S3. Screening found differentially expressed proteins in gastric cancer and early gastric cancer in the cohort, and selected 26 proteins with significant differences between early GC (HGIN-I stage) and GC (I-IV stage) samples and normal control groups as input proteins for model construction;

[0032] S4. Using the LASSO regression model and cross-validation optimization model, a diagnostic model consisting of 13 proteins was obtained, and its diagnostic performance in the discovery cohort training set and test set was evaluated;

[0033] S5. The model was applied to the validation cohort to evaluate its diagnostic performance in differentiating gastric cancer, early gastric cancer, gastric precancerous lesions from controls.

[0034] The following is a specific implementation according to the steps of the above method:

[0035] Clinical data and plasma sample collection of gastric cancer and control investigators:

[0036] Using a retrospective analysis, two study cohorts were recruited. One was a discovery cohort, which included 100 gastric cancer cases and 100 controls who underwent gastrointestinal endoscopy. The other was a validation cohort, which included gastric cancer cases (n=50) and control donors who underwent gastrointestinal endoscopy (n=50). The inclusion criteria were as follows: (1) Chinese Han donors. (2) Aged between 40 and 85 years. (3) Cases that met the gastric cancer diagnosis and treatment guidelines of NCCN and CSCO. (4) Cases with gastrointestinal endoscopy and pathology reports. The exclusion criteria were as follows: (1) Cases with other cancers. (2) Cases who received radiotherapy and chemotherapy. (3) Cases with very serious chronic diseases. (4) Cases with infectious diseases. (5) Plasma samples with hemolysis. After reviewing the clinical data and pathological reports, the discovery cohort that met the above inclusion and exclusion criteria included 48 gastric cancer patients (stages I-IV) and 6 patients with gastric precancerous lesions (low-grade intraepithelial neoplasia, LGIN and high-grade intraepithelial neoplasia, HGIN), while the validation cohort included 24 gastric cancer patients and 4 gastric precancerous lesions. The histological classification followed the WHO. The TNM classification was determined according to the 8th edition of the American Joint Committee on Cancer (AJCC). During the same period, control donors who underwent gastrointestinal endoscopy were recruited. The inclusion criteria were consistent with (1) and (2) above. The exclusion criteria were as follows: (1) Donors with cancer. (2) Donors with HP-positive, intestinal metaplasia, atrophic gastritis, and gastric polyps. Others were the same as (3)-(5) above. After reviewing the gastrointestinal endoscopy reports, the discovery cohort and validation cohort included 34 or 22 non-cancer donors, respectively. The relevant statistical information of the included study cohorts is shown in Table 1.

[0037] Table 1

[0038]

[0039] Among them, LGIN is low-grade intraepithelial neoplasia and HGIN is high-grade intraepithelial neoplasia.

[0040] Collect whole blood into a collection tube containing EDTA anticoagulant and centrifuge at 1000-2000g for 10 minutes in a refrigerated centrifuge (2-8°C). After centrifugation, immediately transfer the upper plasma to a clean and sterile 1.5ml cryotube, freeze it in liquid nitrogen, and then store it in a -80°C refrigerator for a long time. Samples that show severe hemolysis during the collection process will be discarded.

[0041] Screening of multi-molecular markers of gastric cancer plasma proteins and construction, optimization and validation of molecular diagnostic models:

[0042] The collected plasma samples were tested using Olink Plasma Proteomics' PEA technology. 1 ml of plasma was mixed with 369 pairs of antibodies labeled with different DNA oligonucleotides. When the antibody binds to the target protein, the oligonucleotide is brought into proximity and extended by DNA polymerase. They are then quantified as surrogate markers for specific proteins using next-generation sequencing. To normalize the data and reduce intra- and inter-assay variations, the data are reported as normalized protein expression (NPX) values ​​(units on a log2 scale).

[0043] In the discovery cohort, differentially expressed proteins in gastric cancer and early gastric cancer were screened through differential analysis. The protein data were then normalized using z-score, and 26 proteins that intersected with proteins that were significantly different between early GC (HGIN-I stage) samples and GC (I-IV stage) samples and the normal group were selected as input proteins for model construction to establish a diagnostic model that is applicable to both early gastric cancer and gastric cancer patients in the discovery cohort.

[0044] The performance and stability of the input protein were investigated using the LASSO regression model. The main features of the model were obtained by fitting the LASSO regression model on the training set (90% of the samples) and further verifying it on the test set (10% of the samples). The penalty parameter (λ) was trained using a ten-fold cross-validation method. In order to obtain robust model parameters, 500 ten-fold cross-validation steps were performed, and the λ of the final model was taken from the median of 500 λ values. Thus, a molecular diagnostic model containing 13 proteins was obtained, and the specific model parameters are shown in Table 2 below.

[0045] Table 2

[0046]

[0047] The overall diagnostic performance of the model was evaluated by the AUC of the ROC curve as well as the sensitivity, specificity, and positive predictive value. The model had good diagnostic performance in distinguishing gastric cancer (stage I-IV) from controls in both the training set and the test set, with AUC greater than 0.999, and sensitivity, specificity, and positive predictive value all equal to 1.000. Figure 3 And as shown in Table 3.

[0048] Table 3

[0049]

[0050] The optimal molecular diagnostic model obtained in the discovery cohort was applied to the validation cohort, and the overall diagnostic performance of the model in the two cohorts was evaluated by the AUC of the ROC curve as well as the sensitivity, specificity, and positive predictive value. It was finally found that the model had good diagnostic efficacy in distinguishing gastric cancer, early gastric cancer, and gastric precancerous lesions in both the discovery cohort and the validation cohort. Figure 4-6 shown.

[0051] The specific diagnostic performance is shown in Table 4. The diagnostic performance in the discovery cohort was gastric cancer (stage I-IV): AUC 0.994, sensitivity 1.000, specificity 0.979, positive predictive value 0.971, early gastric cancer (HGIN-I): AUC 0.998, sensitivity 1.000, specificity 1.000, positive predictive value 1.000, gastric precancerous lesions (LGIN-HGIN): AUC 0.990, sensitivity 0.971, specificity 1.000, positive predictive value 1.000; The diagnostic efficacy in the validation cohort was as follows: gastric cancer (stage I-IV): AUC 0.930, sensitivity 1.000, specificity 0.875, positive predictive value 0.880; early gastric cancer (HGIN-I): AUC 0.818, sensitivity 0.909, specificity 0.778, positive predictive value 0.909; gastric precancerous lesions (LGIN-HGIN): AUC 0.977, sensitivity 0.955, specificity 1.000, positive predictive value 1.000.

[0052] Table 4

[0053]

[0054] The potential role of plasma protein multi-molecular markers discovered in prognosis:

[0055] A dose-response analysis of plasma protein levels and GC pathological progression was performed on all samples in the discovery cohort and validation cohort to further explore the prognostic value and reliability of the proteins identified in the resulting diagnostic panel. All proteins in the diagnostic panel met the criteria of a Q value less than 0.05, such as Figure 7This suggests that the proteins identified in the diagnostic panel have high reliability and mediate the pathological progression of GC and are also potential prognostic markers.

[0056] The multi-molecular markers for gastric cancer diagnosis found in this embodiment and the diagnostic model constructed have good diagnostic performance for patients with zoned gastric cancer, early gastric cancer, gastric precancerous lesions and controls. It was found that the AUC in the cohort was greater than 0.990, the sensitivity was greater than 0.971, the specificity was greater than 0.979, and the positive predictive value was greater than 0.971. The AUC in the verification cohort was greater than 0.818, the sensitivity was greater than 0.909, the specificity was greater than 0.778, and the positive predictive value was greater than 0.88. Relying on Olink proteomics PEA technology to find gastric cancer diagnostic markers, the technology can accurately detect thousands of proteins in a few microliters of blood samples, and the detection target protein concentration spans 10 orders of magnitude. The minimum detection concentration can reach fg / ml, which has higher sensitivity and specificity than mass spectrometry technology, as well as lower sample input. And from the perspective of liquid biopsy, the use of trace plasma samples for detection greatly reduces the invasiveness and harm of current conventional endoscopic detection methods to patients. It is beneficial to solve the problems of existing molecular markers, diagnostic criteria and minimally invasive or non-invasive methods for early diagnosis of gastric cancer, the high invasiveness and high cost of conventional gastroscopy, the high reliance on the skills of endoscopists for diagnostic accuracy, and the low sensitivity and specificity of conventional gastric cancer screening markers. In addition, it solves the problem that the commonly used proteomics technology mass spectrometry is difficult to identify low-abundance proteins (ng / ml-fg / ml) and requires a large amount of sample input.

[0057] The above is only a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, and should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications and environments, and can be modified within the scope of the concept described herein through the above teachings or the technology or knowledge of the relevant field. The changes and modifications made by those skilled in the art shall not deviate from the spirit and scope of the present invention, and shall be within the scope of protection of the claims attached to the present invention.

Claims

1. A method for constructing a multi-molecular diagnostic model for early gastric cancer based on plasma proteins, characterized in that the steps include: S1. Collect whole blood, centrifuge to obtain plasma, freeze it in liquid nitrogen, store it at -80℃, and remove hemolyzed samples; S2. Use Olink plasma proteomics PEA to detect plasma samples and obtain standardized protein expression values; S3. Screening found differentially expressed proteins in gastric cancer and early gastric cancer in the cohort, and selected 26 proteins with significant differences between early GC (HGIN-I stage) and GC (I-IV stage) samples and normal control groups as input proteins for model construction; S4. Using the LASSO regression model and cross-validation optimization model, a diagnostic model consisting of 13 proteins was obtained.

2. The method for constructing an early multi-molecular diagnostic model for gastric cancer based on plasma proteins according to claim 1, characterized in that: The step S2 is specifically as follows: using Olink plasma proteomics PEA to detect the collected plasma sample, 1 ml of plasma is mixed with 369 pairs of antibodies labeled with different DNA oligonucleotides, after the antibody binds to the target protein, the oligonucleotide is extended, and quantified into a standardized protein expression (NPX) value using next generation sequencing.

3. The method for constructing an early multi-molecular diagnostic model for gastric cancer based on plasma proteins according to claim 1, characterized in that: The step S3 is specifically as follows: in the discovery cohort, differentially expressed proteins in gastric cancer and early gastric cancer are screened by differential analysis, z-score normalization is used, and 26 proteins with significant differential protein intersections between early GC (HGIN-I stage) and GC (I-IV stage) samples and the normal group are selected as input proteins for model construction.

4. The method for constructing an early multi-molecular diagnostic model for gastric cancer based on plasma proteins according to claim 1, characterized in that: The step S4 is specifically as follows: using the LASSO regression model to examine the performance and stability values ​​of the input protein, fitting the model in the training set, verifying it in the test set, training the penalty parameter using a ten-fold cross-validation method, performing cross-validation, taking the median of the penalty parameter as the penalty parameter of the final model, and obtaining a molecular diagnostic model containing 13 proteins.

Images