Method for constructing anti-programmed death protein-1 monoclonal antibody treatment response prediction model

Abstract

The application provides a kind of anti-programmed death protein-1 monoclonal antibody treatment response prediction model construction method, belongs to the construction method technical field of prediction model.The method comprises: obtaining non-small cell lung cancer patient blood sample;Extract peripheral blood mononuclear cell of patient blood sample before and after anti-programmed death protein-1 monoclonal antibody treatment;In turn, the steps of including protein extraction, reduction alkylation, enzymolysis are handled, and sample peptide is obtained;Basic characteristic analysis and function annotation are carried out, and the protein spectrum and modification spectrum of peripheral blood mononuclear cell of anti-programmed death protein-1 monoclonal antibody treatment non-small cell lung cancer patient are obtained;Using machine learning, biomarker is screened;Monitoring is carried out, and anti-programmed death protein-1 monoclonal antibody treatment response prediction model is obtained.It can predict the anti-programmed death protein-1 monoclonal antibody treatment effect of non-small cell lung cancer patient.

Description

Technical Field

[0001] This invention relates to the technical field of predictive model construction methods, and in particular to a method for constructing a predictive model for anti-programmed death protein-1 monoclonal antibody therapy response. Background Technology

[0002] In recent years, the discovery of immune checkpoints such as cytotoxic T-lymphocyte-associated protein-4 (CTLA-4) and anti-programmed death protein-1 (PD-1) has propelled the development of tumor immunotherapy. Immune checkpoint inhibition (ICI) therapy, by releasing the immunosuppressive effects of T cells, elicits a powerful anti-tumor immune response, fundamentally changing the treatment paradigm for cancer. Currently, immune checkpoint inhibitors targeting cytotoxic T-lymphocyte-associated protein-4 and anti-programmed death protein-1 monoclonal antibodies and their ligand (PD-L1) are increasingly used to treat various types of tumors. However, their therapeutic response is rarely studied. Summary of the Invention

[0003] In view of this, the present invention provides a method for constructing a predictive model for anti-programmed death protein-1 monoclonal antibody therapy response, which can predict the anti-programmed death protein-1 monoclonal antibody therapy response in non-small cell lung cancer patients, thereby predicting the efficacy of anti-programmed death protein-1 monoclonal antibody therapy in non-small cell lung cancer patients, and thus is more suitable for practical use.

[0004] To achieve the above objectives, the technical solution for constructing a predictive model for anti-programmed death protein-1 monoclonal antibody therapy provided by this invention is as follows:

[0005] The method for constructing a predictive model for anti-programmed death protein-1 monoclonal antibody therapy provided by this invention includes the following steps:

[0006] Obtain samples from non-small cell lung cancer patients, wherein the patients are those who have received anti-programmed death protein-1 monoclonal antibody treatment;

[0007] Peripheral blood mononuclear cells were extracted from the patient samples before and after anti-programmed death protein-1 monoclonal antibody treatment. For each patient sample, there was a pair of samples: a pre-treatment subsample and a post-treatment subsample.

[0008] Peripheral blood mononuclear cells from the patient sample before and after anti-programmed death protein-1 monoclonal antibody treatment were subjected to steps including protein extraction, reductive alkylation, and enzymatic digestion to obtain sample peptides.

[0009] The sample peptides are processed to obtain processed sample peptides.

[0010] Basic feature analysis and functional annotation were performed on the processed sample peptides to obtain the protein profile and modification profile of peripheral blood mononuclear cells from patients treated with anti-programmed death protein-1 monoclonal antibody.

[0011] Based on the protein and modification profiles of peripheral blood mononuclear cells from patients treated with the anti-programmed death protein-1 monoclonal antibody, machine learning was used to screen for biomarkers of response to anti-programmed death protein-1 monoclonal antibody treatment.

[0012] Biomarkers for the response to anti-programmed death protein-1 monoclonal antibody therapy in the non-small cell lung cancer patient samples were monitored to obtain a predictive model for the response to anti-programmed death protein-1 monoclonal antibody therapy.

[0013] The emotion recognition method provided by this invention can also be further implemented using the following technical measures.

[0014] Preferably, in the step of obtaining non-small cell lung cancer patient samples, wherein the patients are those treated with anti-programmed death protein-1 monoclonal antibody,

[0015] The inclusion criteria for the non-small cell lung cancer patient sample included: patients aged 18 years or older, patients clinically diagnosed with non-small cell lung cancer, patients with complete clinical information, and patients receiving anti-programmed death protein-1 monoclonal antibody therapy for the first time.

[0016] The exclusion criteria for the non-small cell lung cancer patient samples included: the patient died during treatment with anti-programmed death protein-1 monoclonal antibody, or the patient died during treatment with anti-programmed death protein-1 monoclonal antibody.

[0017] In the process of obtaining sample peptides from peripheral blood mononuclear cells of the patient sample before and after anti-programmed death protein-1 monoclonal antibody treatment, the steps include protein extraction, reductive alkylation, and enzymatic digestion. The protein extraction includes the following steps:

[0018] Peripheral blood mononuclear cell samples were taken from the liquid nitrogen container, with a cell count of 2 × 10⁻⁶. 6 After thawing in a 37°C water bath, centrifuge at 1200 rpm for 3 minutes and discard the frozen solution.

[0019] Prepare a 1× medium-efficiency radioimmunoprecipitation lysis buffer, wherein the volume ratio of the radioimmunoprecipitation lysis buffer to the protease inhibitor is 99:1.

[0020] Add radioimmunoprecipitation lysis buffer according to the amount of peripheral blood mononuclear cells, wherein each 1×10 6 Add 200 μL of radioimmunoprecipitation lysis buffer to each cell.

[0021] Two homogenization beads were added for lysis. Homogenization was performed once, with a frequency of 65 Hz and a total of 2 grinding cycles. The cycle was 45 seconds on, 10 seconds off, and incubated on ice for 10 minutes. This process was repeated twice.

[0022] Centrifuge at 4℃ and 12000rpm for 10min, then remove the supernatant;

[0023] Accurately pipette 200 μg of protein solution, add 3 times the volume of ice-cold acetone, and freeze at -80°C for 1 hour.

[0024] After 1 hour, remove the protein and centrifuge at 12,000 rpm for 10 minutes at 4°C. Discard the upper acetone layer and add 500 μL of 80% ice acetone to the lower protein layer. Vortex to mix and centrifuge again under the same conditions. Discard the supernatant.

[0025] Add 200 μL of 50 mM ammonium bicarbonate solution, sonicate for 10 min, and incubate on ice to rehydrate the protein;

[0026] Add 2 μL of 5 mM dithiothreitol to bring the final concentration to 0.05 mM, vortex to mix, then centrifuge at 1000 rpm for 1 min and in a 37°C metal bath for 30 min.

[0027] Add 6 μL of 5 mM iodoacetamide to bring the final concentration to 0.15 mM, vortex to mix, then centrifuge at 1000 rpm for 1 min, and incubate in a 55°C metal bath for 30 min.

[0028] Trypsin hydrolysis: Add 10 μL of 0.5 μg / μL sequencing-grade trypsin to make the protein to sequencing-grade trypsin volume ratio 40:1.

[0029] Vortex to mix, then centrifuge at 1000 rpm for 1 min, digest in a metal bath at 37℃ with shaking at 900 rpm for 12 h, and add 5 μL of sequencing-grade trypsin the next day, digest again under the same conditions for 2 h.

[0030] After 2 hours, the sample was removed, and 2 μL of 0.1% formic acid was added. The reaction was terminated by vortexing for 3 seconds and then frozen at -80°C to obtain the sample peptide.

[0031] Preferably, the sample peptides are processed to obtain processed sample peptides, specifically including the following steps:

[0032] An octadecyl solid-phase extraction column is installed on a solid-phase extraction device, and a centrifuge tube is placed below the column to collect waste liquid.

[0033] Activation: Add 500 μL of 100% acetonitrile to activate the octadecyl solid-phase extraction column, repeat twice;

[0034] Equilibration: Add 500 μL of 0.1% formic acid to wash away residual acetonitrile, repeat twice;

[0035] Add sample: Add the peptide to an octadecyl solid-phase extraction column, collect the eluent, repeat twice and then discard the eluent;

[0036] Desalting: Wash the column with 500 μL of 0.1% formic acid, repeating 2-3 times to remove as much salt as possible;

[0037] Elution of peptides: Add 200 μL of 70% acetonitrile to elute the peptides, repeat twice, and collect the eluent;

[0038] After desalting, peptides were quantified using a peptide concentration assay kit.

[0039] Turn on the low-temperature freeze dryer, set the pressure to 150 kPa, and the cold trap temperature to -56℃ for pre-cooling. After the temperature drops to the predetermined value, place the sample, turn on the centrifuge and vacuum valve in sequence, and close the atmospheric valve. Vacuum concentrate and dry the 400 μL eluent collected by desalting for 3 hours.

[0040] After complete lyophilization, 50 μL of 0.1% formic acid prepared with mass spectrometry water was added to dissolve the lyophilized peptides. After thorough vortexing, the mixture was centrifuged at 12,000 rpm for 10 min at 4 °C. 45 μL of the supernatant was injected into high-performance liquid chromatography for fractionation to obtain the processed sample peptides.

[0041] Preferably, the fractionation by high-performance liquid chromatography specifically includes the following steps:

[0042] 100× Stock Solution: Dilute mass spectrometry grade concentrated ammonia to a concentration of 2M. Take 6.7 mL of 2M ammonia solution and add it to 43.3 mL of mass spectrometry water. Adjust the pH to 10.5 to prepare 50 mL of 100× stock solution and store at 4°C for later use.

[0043] Aqueous buffer A: Add 10 mL of chromatographic grade acetonitrile to 485 mL of mass spectrometry-grade water, then add 5 mL of prepared 100× stock solution to make a 500 mL mixture of 97% mass spectrometry-grade water, 2% chromatographic grade acetonitrile, and 1% 100× stock solution.

[0044] Organic phase buffer B: Add 2 mL of mass spectrometry-grade water to 196 mL of chromatographic grade acetonitrile, then add 2 mL of prepared 100x stock solution to make a 200 mL mixed solution of 1% mass spectrometry-grade water, 98% chromatographic grade acetonitrile, and 1% 100x stock solution.

[0045] 10% Methanol: Dissolve 10 mL of chromatographic grade methanol in 90 mL of water for mass spectrometry to prepare 10% methanol solution for later use.

[0046] After the required solution is prepared, tighten the cap and sonicate for 15 minutes to remove air bubbles from the liquid.

[0047] Under the conditions of column temperature 40℃, sample tray temperature 7℃, and pressure 4000psi, the power supply, column oven, fraction collector, computer, and UV monitor were turned on sequentially. After replacing with freshly prepared mobile phase, wet perfusion, rinsing the injector, cleaning the injection needle, cleaning the pump head, and UV monitoring were performed sequentially, acquiring absorption peaks at 214nm and 280nm wavelengths.

[0048] After injecting 45 μL of sample into the chromatographic column via an autosampler, gradient elution was started while collecting the fractions. The fraction collector was programmed to collect one tube per minute for a total of 56 minutes, collecting 56 tubes. The first tube was discarded, leaving 55 tubes. These tubes were labeled and then combined in a specific order to form the final 20 fractions. The fractions were then centrifuged, concentrated, and lyophilized to obtain the processed sample peptides.

[0049] Preferably, basic feature analysis and functional annotation are performed on the processed sample peptides to obtain the protein profile and modification profile of peripheral blood mononuclear cells from patients treated with anti-programmed death protein-1 monoclonal antibody. This specifically includes the following steps:

[0050] The raw mass spectrometry data was searched using proteome discovery software (Proteome Discoverer 2.4 was used in this patent). Twenty fractions from each sample were combined and matched against a human protein database downloaded from https: / / www.uniprot.org / to obtain protein identification and quantification. The database search used the SEQUEST algorithm with the following parameters: the number of amino acids constituting the peptide must be greater than or equal to 6, a maximum of 2 missed cleavage sites are allowed, the mass tolerance of the precursor ion is set to -10ppm to -10ppm, and the mass tolerance of the precursor ion fragment ions is set to -20ppm to -20ppm.

[0051] After the results identified by Protein Discover 2.4 are exported, the data are merged to generate a matrix. The protein quantification value is standardized using the formula: protein abundance / (sample average protein / average protein abundance of all samples) to eliminate inter-sample protein quantification errors caused by sample volume, instrument operation, etc.

[0052] Based on the results identified by Protein Discover 2.4, after removing contaminating proteins, peptide, protein, and precursor ion mass tolerance distribution analyses were performed to assess the quality of the mass spectrometry data. Basic characteristic analysis, annotation from functional databases including Gene Ontology and the Kyoto Encyclopedia of Genes and Genomes databases, and quantitative analysis were also conducted on the identified proteins, including overall differential analysis of identified proteins, screening of differentially expressed proteins, and expression pattern clustering analysis.

[0053] For each sample, 20 raw files obtained from mass spectrometry were analyzed using an open search engine (OpenDelta was used in this patent). A human protein database downloaded from a protein database (https: / / www.uniprot.org / ) was used for searching and matching. Known modifications were systematically annotated using an existing modification database (https: / / www.unimod.org / ), and unknown modifications were searched and annotated using MSFragger software. Specifically, acetylation at the N-terminus of the protein and oxidation on methionine were set as variable modifications, allowing for open searches of both known and unknown modifications. The molecular weight deviation of precursor ions was set to -10ppm to 10ppm, and the molecular weight deviation range of fragment ions was set with a mass tolerance of -20ppm to 20ppm.

[0054] The analysis results of the open search engine (OpenDelta is used in this patent) are calculated as (number of peptide matching maps of a certain modification at a certain protein site) / (total number of peptide matching maps of the unmodified, modified and other arbitrary modifications of the protein site), in order to remove the influence of protein abundance.

[0055] Based on the identification results of OpenDelta, potential biomarkers were screened after removing in vitro modifications, including alkylation modifications introduced by pretreatment and oxidation modifications of methionine.

[0056] Preferably, in the step of screening for biomarkers of response to anti-programmed death protein-1 monoclonal antibody treatment based on the protein and modification profiles of peripheral blood mononuclear cells from patients treated with the anti-programmed death protein-1 monoclonal antibody using machine learning, the screening criteria include:

[0057] The quantitative values ​​of each protein in two compared sample groups were subjected to a T-test, and the corresponding P-values ​​were calculated as significance indicators. The criteria for screening differentially expressed proteins were a fold change > 2 and a P-value < 0.05.

[0058] The quantitative values ​​of each modification in two compared sample groups were subjected to rank-sum tests, and the corresponding P values ​​were calculated as significance indicators. The false discovery rate method was used to correct the P values, and the fold difference >2 and P value <0.05 were used as the criteria for screening differentially expressed proteins.

[0059] Random forest was used for simulation screening to further explore candidate biomarkers that could be used to predict the benefit of anti-programmed death protein-1 monoclonal antibody therapy. In the random forest analysis, 50,000 trees were constructed using the R package, with a maximum of 5 terminal nodes per tree. To evaluate the effectiveness of the model, subject survival characteristic curves were plotted, and the area under the curve (AUC) of the 95% confidence interval was calculated.

[0060] Preferably, in the step of screening for biomarkers of response to anti-programmed death protein-1 monoclonal antibody treatment based on the protein and modification profiles of peripheral blood mononuclear cells from patients treated with the anti-programmed death protein-1 monoclonal antibody using machine learning, the analytical methods include:

[0061] Protein expression pattern clustering analysis: Two methods, heatmap and pseudo-temporal sequence analysis based on fuzzy mean clustering (MFUZZ), were used to perform clustering analysis on differentially expressed proteins;

[0062] Functional enrichment analysis based on gene ontology and the Kyoto Encyclopedia of Genes and Genomes: Functional and pathway enrichment analysis of differentially expressed proteins was performed using the ClueGO plugin of Cytoscape 3.8.1 software;

[0063] Protein interaction network analysis: The target protein is imported into the STRING database to obtain protein interaction relationships, and the results are imported into Cytoscape 3.8.1 software to draw a protein interaction network diagram. The corresponding core modules and core nodes are obtained by using Cytoscape's internal plugins Mcode and cytoHubba, respectively.

[0064] Preferably, in the step of screening for biomarkers of response to anti-programmed death protein-1 monoclonal antibody treatment based on the protein and modification profiles of peripheral blood mononuclear cells from patients treated with the anti-programmed death protein-1 monoclonal antibody using machine learning, the verification method includes:

[0065] The protein pretreatment for the validation cohort includes protein extraction, reduction, alkylation, enzymatic digestion, and desalting;

[0066] Proteomics: Parallel reaction monitoring methods are established by selecting peptides of target proteins according to the following criteria: 1) Only protein-specific peptides are selected; 2) There are no miscleavage sites in the peptides; 3) There are no variable modification sites in the peptides, such as oxidation and alkylation modifications; 4) Peptides that meet the above requirements are selected based on their high response values; 5) At least one and a maximum of three protein-specific peptides are selected for each protein.

[0067] Modification group: Establish parallel reaction monitoring method for target modified peptides: 1) No miscleavage sites in the peptide; 2) No variable modification sites in the peptide, such as oxidation modification and alkylation modification; 3) Select peptides with high response values ​​that meet the above requirements;

[0068] Samples were analyzed using a reversed-phase octadecyl self-packed capillary column (75 μm × 100 mm) and separated using an EASY-nLC 1200 ultra-high performance liquid chromatography system. Mobile phase A was an aqueous solution containing 0.1% formic acid and 2% acetonitrile; mobile phase B was an aqueous solution containing 0.1% formic acid and 90% acetonitrile. The gradient settings were: 0–46 min, 6%–28% B phase; 46–53 min, 28%–45% B phase; 53–60 min, 100% B phase. The flow rate was maintained at 350 nL / min, and the column temperature was 50 °C. After separation by an ultra-high performance liquid chromatography system, the peptides were injected into a sodium spray ion source (voltage 2.1 kV) and detected by a ThermoScientific™ Q ExactiveHF-X mass spectrometer. The mass spectrometry data were acquired using a high-sensitivity parallel reaction monitoring mode with the following parameters: fragmentation mode was high-energy collisional dissociation, collision energy was 32%, first-order resolution was 60,000, second-order resolution was 15,000, scan range was 100–1800 m / z, scan time was 100 ms, +2 to +4 charge states including precursors were screened, dynamic exclusion time was 30 s, isolation window was 4 Da, and time window was 7 min.

[0069] Parallel reaction monitoring data processing was performed using Skyline 3.6 software. All results were imported into Skyline, the correct peaks were selected, and results for all peptides were obtained from all samples. The parallel reaction monitoring results included protein names and peak areas, which were further output for analysis, and differential proteins in different groups were screened and compared.

[0070] Based on the relative quantitative results obtained from the Skyline peak area, the R language was used to perform correlation analysis and visualization of the differences, and the R package was used to plot the receiver operating characteristic (ROC) curve and calculate the area under the curve.

[0071] Preferably, in the step of monitoring biomarkers of the anti-programmed death protein-1 monoclonal antibody treatment response in the non-small cell lung cancer patient samples to obtain the anti-programmed death protein-1 monoclonal antibody treatment response prediction model, the non-small cell lung cancer patient samples are classified into three subtypes: prediction group, drug resistance group, and benefit group according to the expression trend of modular proteins and modifications.

[0072] The method for constructing a predictive model for anti-programmed death protein-1 (AMP) therapy response provided by this invention uses pre-treatment and post-treatment subsamples as a pair of paired samples. Then, for peripheral blood mononuclear cells of patient samples before and after AMP therapy, protein extraction, reductive alkylation, and enzymatic digestion are performed to obtain sample peptides. After further processing, biomarkers for the efficacy of AMP therapy response are obtained. Monitoring these biomarkers can then yield a predictive model for AMP therapy response, which can predict the efficacy of AMP therapy in non-small cell lung cancer patients. Attached Figure Description

[0073] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0074] Figure 1 A flowchart illustrating the steps of constructing a predictive model for anti-programmed death protein-1 monoclonal antibody therapy provided in this embodiment of the invention;

[0075] Figure 2 A graph showing the molecular subtype characteristics of protein molecule 1 after anti-programmed death protein-1 monoclonal antibody treatment in non-small cell lung cancer patients (using K-means clustering algorithm).

[0076] Figure 3 A is a variable importance graph generated by a random forest algorithm for screening potential response biomarkers for anti-programmed death protein-1 monoclonal antibody therapy, measuring the average reduced Gini coefficient of accuracy for each variable between pre-treatment responders and non-responders;

[0077] Figure 3 B is the area under the curve (top 20) for screening potential response biomarkers for anti-programmed death protein-1 monoclonal antibody therapy, used to assess the ability of a single protein to distinguish between responding and non-responding patients before treatment;

[0078] Figure 3 C represents the Venn diagram of 154 intersecting proteins used for screening potential biomarkers of response to anti-programmed death protein-1 monoclonal antibody therapy.

[0079] Figure 3 Figure D shows 20 key feature proteins (yellow indicates acceptable) obtained from random forest combined with Boruta feature screening for potential response biomarkers of anti-programmed death protein-1 monoclonal antibody therapy.

[0080] Figure 4This is a candidate protein interaction network that can be used to predict the effectiveness of anti-programmed death protein-1 (APP-1) monoclonal antibody therapy; the colors in the figure represent the fold change in protein expression (wherein, before receiving APP-1 monoclonal antibody therapy, the protein expression level in peripheral blood mononuclear cells of patients in the response group was lower than that in the non-responder group); the thickness of the lines indicates whether the protein expression levels are correlated (Spearman correlation analysis), the thicker the line, the stronger the correlation; the size of the circle represents the area under the curve for each protein;

[0081] Figure 5 A represents the receiver operating characteristic curves (ROCs) of five protein clusters as single and combined indicators, respectively.

[0082] Figure 5 B is a receiver operating characteristic curve (ROC) curve predicted by five protein clusters as single and combined indicators, and is a ROC curve obtained by progressively combining multiple indicators.

[0083] Figure 6 For the remaining 16, the subject operating characteristic curves were used as individual indicators;

[0084] Figure 7 A represents the ubiquitin modification profile of peripheral blood mononuclear cells from non-small cell lung cancer patients who received anti-programmed death protein-1 monoclonal antibody therapy (distribution of peptide matching profiles for known modifications (chemical and biological modifications) and unknown modifications);

[0085] Figure 7 B represents an overview of ubiquitin modifications in peripheral blood mononuclear cells from non-small cell lung cancer patients who received anti-programmed death protein-1 monoclonal antibody therapy (number of peptide matching maps for each subclass of biological modifications).

[0086] Figure 7 C represents the ubiquitin modification profile of peripheral blood mononuclear cells from non-small cell lung cancer patients who received anti-programmed death protein-1 monoclonal antibody therapy (an overview of peptide matching profiles for each class of oxidative stress modifications);

[0087] Figure 7 D represents the ubiquitous protein modification profile of peripheral blood mononuclear cells from non-small cell lung cancer patients who received anti-programmed death protein-1 monoclonal antibody therapy (a summary of peptide matching profiles for each class of metabolite modifications).

[0088] Figure 8 A graph showing the molecular subtype characteristics of modified molecules after anti-programmed death protein-1 monoclonal antibody treatment in non-small cell lung cancer patients (using K-means clustering algorithm).

[0089] Figure 9 The receiver operating characteristic curve is used as a single indicator for prediction.

[0090] Figure 10A is used to discover five protein clusters selected in a centralized screening as single and combined indicators for prediction, and the receiver operating characteristic (ROC) curves of the test results are monitored in a validation cohort using parallel reactions (ROC curves for single indicators).

[0091] Figure 10 B is used to identify five protein clusters selected through centralized screening as single and combined indicators for prediction. The receiver operating characteristic (ROC) curves of the test results are monitored using parallel reactions in the validation cohort (ROC curves of multiple indicators stepwise combined). Detailed Implementation

[0092] In view of this, the present invention provides a method for constructing a predictive model for anti-programmed death protein-1 monoclonal antibody therapy response, which can predict the anti-programmed death protein-1 monoclonal antibody therapy response in non-small cell lung cancer patients, thereby predicting the efficacy of anti-programmed death protein-1 monoclonal antibody therapy in non-small cell lung cancer patients, and thus is more suitable for practical use.

[0093] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a method for constructing a predictive model for anti-programmed death protein-1 monoclonal antibody therapy based on the present invention. In the following description, different "embodiments" or "embodiments" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0094] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships, such as A and / or B. Specifically, it can mean that A and B can be included at the same time, A can exist alone, or B can exist alone, and any of the above three situations can be met.

[0095] Peripheral blood mononuclear cell samples were collected from non-small cell lung cancer patients before and after two cycles (6 weeks) of anti-programmed death protein-1 monoclonal antibody (sintilimab, camrelizumab, toripalimab) treatment (Table 1). Patients who achieved partial remission (PR, 9 cases) or stable disease (SD, 28 cases) on their first computed tomography scan at 6 weeks post-treatment were classified as responders (Rs) of anti-programmed death protein-1 monoclonal antibody treatment; patients with progressive tumors on their first computed tomography scan and patients with progressive disease (PD, 16 pairs) who still had disease progression within 6 weeks of treatment initiation were classified as non-responders (NRs) of anti-programmed death protein-1 monoclonal antibody treatment.

[0096] The inclusion and exclusion criteria for samples are as follows:

[0097] (1) Must be 18 years of age or older;

[0098] (2) The pathological diagnosis was non-small cell lung cancer;

[0099] (3) Those who are receiving anti-programmed death protein-1 monoclonal antibody therapy for the first time are enrolled, and those who die or change medication midway are discharged.

[0100] (4) It has complete clinical information.

[0101] Table 1 Clinical baseline data of patients in the discovery set cohort

[0102]

[0103] 1 :n(%); median.

[0104] 2 Pearson chi-square test; Kruskal-Wallis test; Fisher exact probability test.

[0105] Protein extraction and enzymatic hydrolysis

[0106] 1. Remove peripheral blood mononuclear cell samples (approximately 2 x 10⁻⁶ cells) from the liquid nitrogen tank. 6 After thawing in a 37°C water bath, centrifuge at 1200 rpm for 3 minutes and discard the frozen solution.

[0107] 2. Prepare 1x medium-efficiency radioimmunoprecipitation lysis buffer (radioimmunoprecipitation lysis buffer: protease inhibitor = 99:1);

[0108] 3. Add radioimmunoprecipitation lysis buffer (1 x 10⁻⁶) according to the amount of peripheral blood mononuclear cells. 6 Each cell was added to approximately 200 μL of radioimmunoprecipitation lysis buffer.

[0109] 4. Add two homogenization beads for lysis, homogenize once (65Hz, total number of grinding times: 2, 45s on, 10s off), incubate on ice for 10min, repeat twice;

[0110] 5. Centrifuge at 12000 rpm for 10 min at 4℃, and remove the supernatant;

[0111] 6. Determination of protein concentration using the Bicinchoninic acid (BCA) method.

[0112] 7. Accurately pipette 200 μg of protein solution, add 3 times the volume of ice-cold acetone, and freeze at -80°C for 1 hour;

[0113] 8. After 1 hour, remove the protein and centrifuge at 12000 rpm for 10 minutes at 4℃. Discard the upper acetone layer and add 500 μL of 80% ice acetone to the lower protein layer. Vortex to mix and centrifuge again under the same conditions. Discard the supernatant.

[0114] 9. Add 200 μL of 50 mM ammonium bicarbonate solution, sonicate for 10 min (ice bath) to reconstitute the protein;

[0115] 10. Add 2 μL of 5 mM dithiothreitol (final concentration 0.05 mM), vortex to mix, then centrifuge briefly at low speed (1000 rpm, 1 min), and incubate in a 37°C metal bath for 30 min.

[0116] 11. Add 6 μL of 5 mM iodoacetamide (final concentration 0.15 mM), vortex to mix, then centrifuge briefly at low speed (1000 rpm, 1 min), and incubate in a 55℃ metal bath for 30 min.

[0117] 12. Trypsin hydrolysis: Add 10 μL of sequencing-grade trypsin (0.5 μg / μL, protein:trypsin = 40:1).

[0118] Vortex to mix, then centrifuge briefly at low speed (1000 rpm, 1 min), digest in a constant temperature shaking metal bath (37℃, 900 rpm) for 12 h, add 5 μL of trypsin the next day, and digest again under the same conditions for 2 h.

[0119] 13. After 2 hours, remove the sample, add 2 μL of 0.1% formic acid, vortex for 3 seconds to terminate the reaction, and store at -80°C.

[0120] Peptide desalting

[0121] 1. Install the octadecyl solid-phase extraction column on the solid-phase extraction device, and place a centrifuge tube below the column to collect the waste liquid.

[0122] 2. Activation: Add 500 μL of 100% acetonitrile to activate the desalting column, repeat twice.

[0123] 3. Equilibration: Add 500 μL of 0.1% formic acid to wash away residual acetonitrile, repeat twice.

[0124] 4. Add sample: Add the peptide to the octadecyl solid phase extraction column, collect the eluent, repeat twice and then discard the eluent.

[0125] 5. Desalting: Wash the column with 500 μL of 0.1% formic acid, repeating 2-3 times to remove as much salt as possible.

[0126] 6. Elution of peptides: Add 200 μL of 70% acetonitrile to elute the peptides, repeat twice, and collect the eluent.

[0127] Peptide quantification

[0128] After desalting, the peptides were quantified using a peptide concentration assay kit.

[0129] Vacuum concentration and centrifugal drying

[0130] Turn on the low-temperature freeze dryer, set the pressure to 150 kPa, and the cold trap temperature to -56℃ for pre-cooling. After the temperature drops to the predetermined value, place the sample, turn on the centrifuge and vacuum valve in sequence, and close the atmospheric valve. Vacuum concentrate and dry the 400 μL eluent collected from desalting for 3 hours.

[0131] Peptide rehydration

[0132] After complete lyophilization, 50 μL of 0.1% formic acid prepared with mass spectrometry water was added to dissolve the lyophilized peptides. After thorough vortexing, the mixture was centrifuged at 12,000 rpm for 10 min at 4 °C. 45 μL of the supernatant (approximately 200 μg of peptides) was injected into high-performance liquid chromatography (HPLC) for fractionation.

[0133] High performance liquid chromatography-preparative chromatographic fractionation

[0134] Because peripheral blood mononuclear cells contain a wide variety of cell types, the extracted protein components are also quite complex. Therefore, fractionating the obtained peptide samples using high-performance liquid chromatography (HPLC) followed by liquid chromatography-mass spectrometry (LC-MS) analysis can yield more in-depth protein and modification chromatographic information.

[0135] Preparative high-performance liquid chromatography (HPLC) fractionation involves loading peptides onto a specific chromatographic column using a mobile phase, followed by high-alkalinity reverse-phase elution with different proportions of buffer A and buffer B to achieve peptide separation. The elution gradient is as follows:

[0136]

[0137] The specific steps for preparative high-performance liquid chromatography (HPLC) fractionation are as follows:

[0138] Preparation of liquid phase:

[0139] 100x stock solution: Dilute mass spectrometry grade concentrated ammonia to a concentration of 2M. Take 6.7mL of 2M ammonia and add it to 43.3mL of mass spectrometry water. Adjust the pH to 10.5 and prepare 50mL of 100x stock solution. Store at 4℃ for later use.

[0140] Buffer A (Aqueous Phase): Take 485 mL of mass spectrometry water, add 10 mL of chromatographic grade acetonitrile, and then add 5 mL of prepared 100x stock solution to prepare a mixed solution of 500 mL of 97% mass spectrometry water, 2% chromatographic grade acetonitrile, and 1% 100x stock solution.

[0141] Buffer B (organic phase): Add 2 mL of mass spectrometry water to 196 mL of chromatographic grade acetonitrile, then add 2 mL of prepared 100x stock solution to prepare a mixed solution of 200 mL of 1% mass spectrometry water, 98% chromatographic grade acetonitrile, and 1% 100x stock solution.

[0142] 10% Methanol: Take 10 mL of chromatographic grade methanol and add it to 90 mL of mass spectrometer-grade water to prepare 10% methanol for later use.

[0143] After preparing the required solution, tighten the cap and sonicate for 15 minutes to remove air bubbles from the liquid.

[0144] Setting parameters:

[0145] Column temperature 40℃, sample tray temperature 7℃, pressure 4000psi.

[0146] Hierarchical operation:

[0147] Turn on the power, column oven, fraction collector, computer, and UV monitor in sequence. Replace with freshly prepared mobile phase, and perform wet perfusion, rinse the injector, clean the injection needle, clean the pump head, and perform UV monitoring in sequence, collecting absorption peaks at wavelengths of 214 nm and 280 nm.

[0148] After injecting 45 μL of sample into the chromatographic column using an autosampler, gradient elution was started and the fractions were collected simultaneously. The fraction collector was programmed to collect one tube per minute for a total of 56 minutes, collecting 56 tubes. The first tube was discarded, leaving 55 tubes. The tubes were labeled with serial numbers, and the eluent was combined in a certain order to form the final 20 fractions, which were then centrifuged, concentrated, and lyophilized.

[0149] Liquid chromatography-mass spectrometry (LC-MS) detection

[0150] This experiment used an Orbitrap Fusion Lumos (Thermo Fisher Scientific) ultra-high resolution protein spectrometer in conjunction with an Easy-nLC1200 nano-level liquid chromatography system to acquire signals from fractionated peptides.

[0151] The chromatographic column used octadecyl group as packing material, with a length of 15 cm, an inner diameter of 75 μm, and a packing particle size of 1.9 μm. The mobile phase was prepared as follows: Phase A: 0.1% formic acid; Phase B: 80% acetonitrile, 0.1% formic acid. Each sample was eluted by the liquid phase for 60 min, with the following elution gradient:

[0152]

[0153] High-resolution mass spectrometry (HMS) detection employed high-energy collisional dissociation (HCD) fragmentation at an energy of 28%. The mass analyzer was an orbital ion trap, and data-dependent acquisition (DDA) was used. The first-level scan (MS1) resolution was set to 60,000, and the automatic gain control target was 3e⁻¹. 6 The maximum ion implantation time was 50 ms, and the scan range was 100–1800 mass-to-charge ratio (m / z). The second-stage scan (MS2) selected the top 20 precursor ions by ion intensity from MS1 for further fragmentation, with a resolution of 15000 and an automatic gain control target of 5e. 4 The maximum ion implantation time is 45ms, and the dynamic exclusion time is 50s.

[0154] Candidate biomarker screening

[0155] Proteomic data analysis

[0156] Protein Discover 2.4 software was used to search the mass spectrometry raw data. The 20 fractions of each sample were combined and matched with the human protein library downloaded from the protein database (https: / / www.uniprot.org / ). The parameters were set as follows: the number of amino acids in the constituent peptides is greater than or equal to 6, a maximum of 2 missed cleavage sites are allowed, the mass tolerance of the precursor ion is set to -10ppm-10ppm, and the mass tolerance of the fragment ion is set to -20ppm-20ppm.

[0157] Protein data standardization

[0158] After the results identified by Protein Discover 2.4 are exported, the data are merged to generate a matrix. The protein quantification values ​​are standardized using quantile, i.e., protein abundance / (sample average abundance / average abundance of all samples) to eliminate inter-sample protein quantification errors caused by sample volume, instrument operation, etc.

[0159] Proteome data screening

[0160] Based on the results identified by Protein Discover 2.4, after removing contaminating proteins, peptide, protein, and precursor ion mass tolerance distribution analyses were performed to assess the quality of the mass spectrometry data. Basic characteristic analysis, annotation from functional databases including Gene Ontology and the Kyoto Encyclopedia of Genes and Genomes, and quantitative analysis were also performed on the identified proteins, including overall differential analysis of identified proteins, screening of differentially expressed proteins, and expression pattern clustering analysis.

[0161] Modification group data analysis

[0162] Twenty raw files obtained from mass spectrometry of each sample were searched and matched against a human protein database downloaded from the protein database (https: / / www.uniprot.org / ) using OpenDelta. Known modifications were systematically annotated using an existing modification database (https: / / www.unimod.org / ), and unknown modifications were searched and annotated using MSFragger software. Acetylation at the N-terminus of the protein and oxidation on methionine were set as variable modifications, allowing for open searches of both known and unknown modifications. The molecular weight deviation range for precursor ions was set to -10ppm to 10ppm, and the molecular weight deviation range for fragment ions was set to -20ppm to 20ppm.

[0163] Data preprocessing and standardization

[0164] The analysis results of the open search engine (OpenDelta is used in this patent) are calculated as (number of peptide matching maps of a certain modification at a certain protein site) / (total number of peptide matching maps of the unmodified, modified and other arbitrary modifications of the protein site), in order to remove the influence of protein abundance.

[0165] Modification group data filtering

[0166] Based on the identification results of OpenDelta, potential biomarkers were screened after removing in vitro modifications, including alkylation modifications introduced by pretreatment and oxidation modifications of methionine.

[0167] Biomarker screening

[0168] Statistical analysis

[0169] Differential expression analysis used the ratio of the means of all biological measurements for each protein (modification) in two compared sample groups as the fold change (FC). To determine the statistical significance of the difference, a t-test was performed on the quantitative values ​​of each protein in the two compared sample groups, and the corresponding p-value was calculated as the significance index. A fold change > 2 and a p-value < 0.05 were used as the criteria for screening differentially expressed proteins.

[0170] The quantitative values ​​of each modification in two compared sample groups were subjected to a rank-sum test, and the corresponding P-values ​​were calculated as significance indicators. The P-values ​​were corrected using the false discovery rate method, and differential proteins were screened with a fold change > 2 and a P-value < 0.05.

[0171] Random forest was used for simulation screening to further explore candidate biomarkers that could be used to predict the benefit of anti-programmed death protein-1 monoclonal antibody therapy. In the random forest analysis, 50,000 trees were constructed using the R package Random Forest (version 4.5.14), with a maximum of 5 terminal nodes per tree. To evaluate the model's effectiveness, receiver operator characteristic (ROC) curves were plotted, and the area under the curve (AUC) of its 95% confidence interval was calculated.

[0172] Protein expression pattern clustering analysis

[0173] Two methods were used to perform cluster analysis on differentially expressed proteins: heatmaps and pseudo-time sequence analysis based on fuzzy mean clustering (MFUZZ).

[0174] Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis

[0175] The ClueGO plugin within Cytoscape 3.8.1 was used to perform gene ontology and Kyoto Encyclopedia of Genes and Genomes functional and enrichment analyses on differentially expressed proteins; these analyses mainly included biological process and pathway enrichment analyses.

[0176] Protein-protein interaction network analysis

[0177] Import the target protein into the STRING website (https: / / cn.string-db.org / ) to obtain protein interaction relationships, and import the results into Cytoscape 3.8.1 software to draw a protein interaction network diagram. Use Cytoscape's internal plugins Mcode and cytoHubba to obtain the corresponding core modules and core nodes, respectively.

[0178] Validation queue

[0179] Research subjects

[0180] This study used peripheral blood mononuclear cell (PBMC) samples from 100 non-small cell lung cancer (NSCLC) patients who received anti-programmed death protein-1 (AMP) monoclonal antibodies (sintilimab, camrelizumab, and toripalimab) before treatment (Table 2). This cohort only included PBMC samples from NSCLC patients who received AMP treatment before treatment.

[0181] Table 2 Clinical baseline data of patients in the validation set cohort

[0182]

[0183] 1 :n(%); median.

[0184] 2 Pearson chi-square test; Kruskal-Wallis test; Fisher exact probability test.

[0185] Parallel reaction monitoring sample pretreatment

[0186] The protein pretreatment for the validation cohort was the same as in section 1.1, except that high-performance liquid chromatography (HPLC) preparative chromatography fractionation was not performed.

[0187] Establishment of a parallel reaction monitoring method

[0188] Proteomics: Parallel reaction monitoring methods are established by selecting peptides of target proteins according to the following criteria: 1) Only protein-specific peptides are selected; 2) There are no miscleavage sites in the peptides; 3) There are no variable modification sites in the peptides, such as oxidation and alkylation modifications; 4) Peptides that meet the above requirements are selected based on their high response values; 5) At least one and a maximum of three protein-specific peptides are selected for each protein.

[0189] Modification group: Establish parallel reaction monitoring method for target modified peptides: 1) No miscleavage sites in the peptide; 2) No variable modification sites in the peptide, such as oxidation modification and alkylation modification; 3) Select peptides with high response values ​​that meet the above requirements.

[0190] Parallel reaction monitoring methods

[0191] To ensure data quality, the analysis of pooled samples served as quality control samples throughout the analysis process to observe the stability of the instrument signal. To avoid systematic errors, samples from different groups were analyzed by mass spectrometry in a random order. Samples were analyzed using a reversed-phase octadecyl self-packed capillary column (75 μm × 100 mm) and separated using an EASY-nLC 1200 ultra-high performance liquid chromatography system (mobile phase A was an aqueous solution containing 0.1% formic acid and 2% acetonitrile; mobile phase B was an aqueous solution containing 0.1% formic acid and 90% acetonitrile; gradient settings: 0–46 min, 6%–28% B phase; 46–53 min, 28%–45% B phase; 53–60 min, 100% B phase, flow rate maintained at 350 nL / min, column temperature 50 °C). After separation by an ultra-high performance liquid chromatography (UHPLC) system, the peptides were injected into a sodium spray ion source (2.1 kV) and detected using a ThermoScientific™ Q Exactive HF-X mass spectrometer. Mass spectrometry data were acquired using a high-sensitivity parallel reaction monitoring (PRC) mode with the following parameters: high-energy collisional dissociation (HACD), collision energy of 32%, first-order resolution of 60,000 m / s, second-order resolution of 15,000 m / s, scan range of 100–1800 m / s, scan time of 100 ms, screening for +2 to +4 charge states including precursors, dynamic exclusion time of 30 s, isolation window of 4 Da, and time window of 7 min.

[0192] Parallel reaction monitoring data analysis

[0193] Parallel reaction monitoring (PRM) data processing was performed using Skyline 3.6 software. All results were imported into the software, and the correct peaks were manually selected to obtain results for all peptides across all samples. PRM results, including protein names and peak areas, were further analyzed to screen and compare differentially expressed proteins across different groups. Existing studies did not use internal control proteins, resulting in poor normalization; therefore, unnormalized peptide abundance was used. If time permits, efforts could be made to normalize the abundance of each peptide in each sample and correct the sample loading amount and mass spectrometry signal intensity using the total ion chromatogram of each sample.

[0194] Based on the relative quantitative results obtained from the Skyline peak area, the R language was used to perform correlation analysis and visualization of the differences, and the R package was used to plot the receiver operating characteristic curve and calculate the area under the curve (AUC).

[0195] result

[0196] Screening of candidate biomarkers for anti-programmed death protein-1 monoclonal antibody therapy in non-small cell lung cancer based on proteomics.

[0197] In this study, 11,531 high-confidence proteins were identified using PD software, of which 77 were contaminants. After removing the contaminants, 11,454 proteins remained. Analysis and screening of these proteins yielded potential biomarkers for predicting the efficacy of anti-programmed death protein-1 (APMP) monoclonal antibody therapy. This study focuses on proteomics-based anti-APMP monoclonal antibody therapy for molecular subtypes of non-small cell lung cancer.

[0198] After summing all differentially expressed proteins (1152 in total), the K-means clustering algorithm was used to obtain 6 feature modules. Figure 1 Based on clinical data, these modular proteins and their modification expression trends can be categorized into three subtypes: the "predictive group," the "drug-resistant group," and the "benefit group."

[0199] Before receiving anti-programmed death protein-1 (AMP) monoclonal antibody (AMP) therapy, the protein expression level of module A was significantly higher in peripheral blood mononuclear cells (PBMCs) of responders than in non-responders, while PD-1 AMP therapy had no significant regulatory effect on the protein expression level of module A. The protein expression trend of module D was similar to that of module A, and PD-1 AMP therapy had no significant regulatory effect on the protein expression level of this module. However, before receiving AMP therapy, the protein expression levels of both modules in PBMCs of responders were significantly lower than in non-responders. In summary, the proteins of modules A and D can be used to predict whether patients will benefit from AMP therapy, and therefore they are classified as the "prediction group" subtype.

[0200] Using random forest combined with Boruta features to screen potential biomarkers for predicting responses

[0201] To screen for potential biomarkers related to response, proteins and modifications that showed significant differences (p < 0.05, abs(log difference fold) > = 1) between the "predicted group" and the non-responder group before treatment were first identified. Then, machine learning (random forest) was used to obtain the Gini value for each protein. Figure 2 A); simultaneously plot the receiver operating characteristic curve for each protein and obtain the corresponding area under the curve (A). Figure 2 B); Combining these three indicators, a preliminary screening yielded 154 proteins that could potentially predict whether a patient would benefit from anti-programmed death protein-1 monoclonal antibody therapy. Figure 2 C). Further analysis using random forest combined with Boruta features identified 20 key signature proteins in both the pre-treatment response and non-response groups. Figure 2 D).

[0202] In summary, potential biomarkers for predicting benefit from anti-programmed death protein-1 monoclonal antibody therapy can be obtained. Five highly interconnected core protein clusters can be identified by constructing an interaction network of these response proteins. Figure 3 It also contains 10 core proteins. The core proteins are mainly distributed in core protein cluster 1. The first protein cluster is enriched with CDC42, ROCK1, RHOG, ARF6, RHOA, CYBA, CYBB, MMP9, S100A8, and S100A9; the second protein cluster is enriched with LSM7, LSM3, LRMDA, and SNRPD3; the third protein cluster is enriched with PKM, PGAM1, HK2, and FBP1, which are related to glucose metabolism; the fourth protein cluster includes RETN, MMP8, and DEFA1B; and the fifth protein cluster includes GM2A, GBA, and ASAH1.

[0203] Anti-programmed death protein-1 (APP-1) monoclonal antibody therapy may exert its therapeutic effect by regulating the energy metabolism processes of immune cells. The screening results also indicate that potential biomarkers for predicting whether patients will benefit from APP-1 monoclonal antibody therapy are mainly concentrated on mitochondrial energy metabolism and glucose metabolism-related proteins. Among these, proteins affecting the functional state of immune cells are mainly distributed in protein cluster 1. RHOG downregulation promotes T cell receptor signaling and proliferation, affecting late T cell activation; RHOA plays an important role in the formation of immune synapses between dendritic cells and antigen-specific T cells; in tumor cells, aberrant activation of ARF6 is a potential therapeutic target. When cellular energy is sufficient, ARF6 can regulate the distribution of mitochondria within the cell, protecting mitochondria from oxidative damage; neutrophils, as the most abundant innate immune cells in circulation, are among the first immune cells recruited, and S100A8 / A9 can induce neutrophil activation, thereby enhancing the immune response. PGAM1 in protein cluster 3 is a glucose metabolism-related protein; inhibiting PGAM1 can promote CD8+ T cell infiltration, synergistically exerting an anti-tumor effect with APP-1 immunotherapy. RETN in protein cluster 4 can improve the chemotaxis of immune cells, which is beneficial for tumor immune escape; DEFA1B belongs to the α-defensin subset and activates CD8+. + The chemotactic properties of T cells influence the tumor microenvironment, triggering neutrophil degranulation and thus exerting a strong anti-tumor immune response. Therefore, these proteins can all affect mitochondrial energy metabolism and the functional state of immune cells, thereby influencing the body's response to anti-programmed death protein-1 (APP-1) monoclonal antibody therapy. This suggests that these proteins have the potential to predict whether non-small cell lung cancer patients will benefit from APP-1 monoclonal antibody therapy, but their predictive efficacy requires further investigation in new clinical cohorts.

[0204] Proteome biomarker validation

[0205] Further, by comparing the receiver operating characteristic (ROC) curves of individual proteins and the combined ROC curves within these protein clusters... Figure 4 A combined biomarker model was obtained to predict the effectiveness of anti-programmed death protein-1 monoclonal antibody therapy. In the first protein cluster, RHOG had the highest area under the curve (AUC) (0.82), followed by CYBB (0.81), and CYBA had the lowest (AUC of only 0.65). However, combining RHOG and CYBB increased the AUC to 0.83. With progressively increasing protein concentration, the AUC gradually increased, culminating in a predictive model with an AUC of 1 using RHOG, CYBB, CDC42, RHOA, ROCK1, ARF6, S100A8, and S100A9 as combined indicators. In protein cluster 2, LSM3, LSM7, LRMDA, and SNRPD3 had AUCs ranging from 0.66 to 0.82, but with progressively increasing protein concentration, the AUC increased from 0.82 to 0.88. Protein clusters 3, 4, and 5 exhibited the same trend. Preliminary findings suggest that these proteins, when used as combined diagnostic markers, offer higher diagnostic sensitivity.

[0206] Therefore, a preliminary combination of biomarkers that can be used to predict the effectiveness of anti-programmed death protein-1 monoclonal antibody therapy was obtained. This combination uses protein cluster 1 (RHOG, CYBB, CDC42, RHOA, ROCK1, ARF6, S100A8, and S100A9), protein cluster 2 (LSM3, SNRPD3), protein cluster 3 (FBP1, PGAM1, PKM), protein cluster 4 (DEFA1, MMP8, RETN), and protein cluster 5 (ASAH1, GM2A) as combined markers; and ARF6, EEF1D, EIF2S3, FARSB, GCA, GNG2, H4C9, LAMP1, LMAN1, MNDA, NAMPT, NDUFS3, PSMC4, PSMD6, RAB21, RAB35, and RNASE2 as single markers in an anti-programmed death protein-1 monoclonal antibody therapy response prediction model. Figure 5 ).

[0207] Screening of candidate biomarkers for anti-programmed death protein-1 monoclonal antibody therapy in non-small cell lung cancer based on ubiquitin modification proteomics.

[0208] This study used an open search engine to map ubiquitin modifications in peripheral blood mononuclear cells (PBMCs) from non-small cell lung cancer (NSCLC) patients treated with anti-programmed death protein-1 (APC-1) monoclonal antibodies. A total of 804 modifications were identified at 92,937 sites on 9,013 proteins in the PBMC data from NSCLC patients, resulting in a total of 2,952,321 peptide-matched molecular atlases (PSMs).

[0209] Based on the mass shift of each modification, its corresponding amino acid (AA), and the amino acid's position within the peptide, systematic annotation was performed using existing modification databases (https: / / www.unimod.org / ). According to the annotation results, 721 modifications from 2,923,027 (99%) peptide matching maps were known modifications; the remaining 29,294 peptide matching maps contained 85 peptides with unknown mass shifts, potentially representing peptides with novel modifications, accounting for 1.0% of the total identified peptide matching maps.

[0210] Based on the known sources of the modifications, the 721 known modifications derived from 2,923,027 (99%) peptide-matching maps were further categorized into chemical modifications (2,128,220 PSMs, 72.00%). Figure 6 Two main subclasses were identified: A) and biological modifications (512, 794,807 PSMs, 27%). A total of 211 known chemical modifications were identified in this dataset, including derivatization caused by sample pretreatment, such as propionylation (+40.0313 Da, 540,908 PSMs) caused by acetone precipitation, and aminomethylation (+57.0215 Da, 9,353 PSMs) resulting from alkylation; and some metal ion additions introduced by mass spectrometry analysis, such as sodium ions (+21.9819 Da, 44,271 PSMs). Although this experiment minimized contact with oxygen during protein pretreatment by using a nitrogen chamber and other systems, and avoided heating operations as much as possible, a significant amount of oxidative modification occurred on methionine (+15.9949 Da, 895,302 PSMs). Figure 6 Furthermore, propionylation (+40.0313 Da) was also observed, primarily occurring at the N-terminus of glycine, consistent with the mass shift from glycine to proline substitution; this modification may have been due to acetone treatment during pretreatment. Additionally, significant amounts of deamination oxidation (+0.9840 Da) modification were observed. Moreover, metal ion addition modifications were predominantly associated with sodium (+21.9819 Da), ferric (+53.9193 Da), ferrous (+52.9115 Da), and calcium (+37.9469 Da), mainly occurring on the acidic amino acids glutamic acid and aspartic acid, as well as at the C-terminus of peptides. These sites can generate negatively charged carboxyl groups, which can then add to positively charged metal ions. Dehydration-induced dehydration (-18.0105 Da) modification primarily occurred on hydroxyl-containing tyrosine, threonine, and serine. These results demonstrate the reliability of the modification sites identified using OpenDelta in this study.

[0211] Proteomics-based anti-programmed death protein-1 monoclonal antibody therapy for molecular subtypes of non-small cell lung cancer

[0212] First, rank-sum analysis (P<0.05) and fold-over (fold-over >2 or fold-over <0.5) were used to identify 3549 differentially expressed site modifications (866 proteins) screened from peripheral blood mononuclear cells of non-small cell lung cancer patients in both the response and non-responder groups before anti-programmed death protein-1 (APMP-1) treatment. Of these, 102 modification sites (67 proteins) were upregulated, and 3447 modification sites (847 proteins) were downregulated. After AMP-1 treatment, 2896 differentially expressed site modifications (796 proteins) screened from peripheral blood mononuclear cells of non-small cell lung cancer patients in both the response and non-responder groups were identified. Of these, 435 modification sites (235 proteins) were upregulated, and 2461 modification sites (682 proteins) were downregulated. Compared to pre-treatment levels, 478 differentially expressed modifications (300 proteins) were identified in peripheral blood mononuclear cells (PBMCs) of non-small cell lung cancer (NSCLC) patients before and after treatment with anti-programmed death protein-1 (APP-1) antibody. Of these, 206 modifications (148 proteins) were upregulated, and 272 modifications (182 proteins) were downregulated. Following APP-1 antibody treatment, 510 biological modifications were identified in PBMCs of unresponsive NSCLC patients, distributed across 33,894 sites on 4,196 proteins. Compared to pre-treatment levels, 78 differentially expressed modifications (63 proteins) were identified in PBMCs of unresponsive NSCLC patients, of which 37 modifications (44 proteins) were upregulated, and 34 modifications (30 proteins) were downregulated.

[0213] Proteomics-based anti-programmed death protein-1 monoclonal antibody therapy for molecular subtypes of non-small cell lung cancer

[0214] After summarizing the differences obtained from the above analysis, the K-means clustering algorithm was used to obtain four feature modules. Figure 7 Based on clinical data, the modification expression trends of the four modules can be categorized into three subtypes: "prediction group," "drug resistance group," and "benefit group." In summary, based on the expression characteristics of protein modifications before and after anti-programmed death protein-1 monoclonal antibody treatment in responding and non-responding patients, they can be divided into the "prediction group" (modules B and C), the "benefit group" (modules C and D), and the "drug resistance group" (module A).

[0215] Module B contains 345 biological modifications distributed across 1796 sites on 552 proteins. The expression characteristics of these modifications in peripheral blood mononuclear cells (PBMCs) of non-small cell lung cancer (NSCLC) before and after anti-programmed death protein-1 (AMP) therapy were as follows: Before treatment, the expression level of Module B protein modifications in PBMCs of responding patients was significantly lower than that of non-responders. AMP therapy had no significant regulatory effect on the protein modifications of this module. Therefore, the protein modifications of Module B can be used to predict whether patients will benefit from AMP therapy. Module C contains 445 biological modifications distributed across 2126 sites on 673 proteins. The expression characteristics of these modifications in peripheral blood mononuclear cells of NSCLC before and after anti-AMP therapy were as follows: Similar to Module B, before AMP therapy, the expression level of Module B protein modifications in PBMCs of responding patients was significantly lower than that of non-responders. However, anti-programmed death protein-1 (APP-1) monoclonal antibody treatment downregulated the protein modification expression of module C, approaching that of the response group. Therefore, this may be related to the benefit of APP-1 treatment.

[0216] Screening and validation of biomarkers for modified groups

[0217] Significantly different modifications were selected from the "prediction group," and receiver operating characteristic (ROC) curves were plotted. Based on the area under the curve (AUC > 0.6), 93 candidate modified peptides were identified. These were then validated using a new cohort and parallel response monitoring method, revealing 62 modified peptides that could serve as potential biomarkers for predicting the efficacy of anti-programmed death protein-1 monoclonal antibody therapy. Figure 8 ).

[0218] Mass spectrometry-based proteomics has been widely used in research to discover biomarkers for disease prognosis and diagnosis. Parallel reaction monitoring (PRM), a quantitative proteomics method, focuses on small target sets and can be used to measure most proteins or proteins with low detection counts due to dynamic range limitations. It is considered one of the most sensitive and specific mass spectrometry-based quantitative methods. Therefore, this chapter uses a novel clinical cohort (100 cases) to validate potential biomarkers identified through proteomics screening that can predict the efficacy of anti-programmed death protein-1 monoclonal antibody therapy using parallel reaction monitoring. A joint prediction model should be established based on the validation results.

[0219] This study employed fuzzy mean clustering-based pseudo-temporal sequence analysis (MFUZZ) and K-means clustering algorithms to obtain protein subtypes, categorizing them into "predicted group," "benefit group," and "drug-resistant group." By constructing the protein-protein interaction network of histones in the "predicted group," and combining it with MCODE to obtain core modules and proteins, we further screened potential protein biomarkers that could predict the efficacy of anti-programmed death protein-1 monoclonal antibody therapy using random forest, Boruta feature analysis, and area under the curve (AUC). These biomarkers were then validated using a novel clinical cohort combined with parallel response monitoring.

[0220] The embodiments of the present invention ultimately obtained a combination of biomarkers that can be used to predict the effectiveness of anti-programmed death protein-1 monoclonal antibody therapy in non-small cell lung cancer patients, including 37 proteins (Table 1) and 62 protein-modifying peptides (Table 2).

[0221] Table 1. Potential biomarkers in the proteome that can be used to predict the efficacy of anti-programmed death protein-1 monoclonal antibody therapy in non-small cell lung cancer.

[0222] Joint Indicators RHOG, CYBB, CDC42, RHOA, ROCK1, ARF6, S100A8 and S100A9 Joint Indicators LSM3,SNRPD3 Joint Indicators FBP1,PGAM1,PKM Joint Indicators DEFA1,MMP8,RETN Joint Indicators ASAH1,GM2A Single indicator ARF6 Single indicator EEF1D Single indicator EIF2S3 Single indicator FARSB Single indicator GCA Single indicator GNG2 Single indicator H4C9 Single indicator LAMP1 Single indicator LMAN1 Single indicator MNDA Single indicator NAMPT Single indicator NDUFS3 Single indicator PSMC4 Single indicator PSMD6 Single indicator RAB21 Single indicator RAB35 Single indicator RNASE2

[0223] The method for constructing a predictive model for anti-programmed death protein-1 (AMP) therapy response provided by this invention uses pre-treatment and post-treatment subsamples as a pair of paired samples. Then, for peripheral blood mononuclear cells of patient samples before and after AMP therapy, protein extraction, reductive alkylation, and enzymatic digestion are performed to obtain sample peptides. After further processing, biomarkers for the efficacy of AMP therapy response are obtained. Monitoring these biomarkers can then yield a predictive model for AMP therapy response, which can predict the efficacy of AMP therapy in non-small cell lung cancer patients.

[0224] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0225] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for constructing a predictive model for treatment response to anti-programmed death protein-1 monoclonal antibody, characterized in that, Includes the following steps: Obtain samples from non-small cell lung cancer patients, wherein the patients are those who have received anti-programmed death protein-1 monoclonal antibody treatment; Peripheral blood mononuclear cells were extracted from the patient samples before and after anti-programmed death protein-1 monoclonal antibody treatment. For each patient sample, there was a pair of samples: a pre-treatment subsample and a post-treatment subsample. Peripheral blood mononuclear cells from the patient sample before and after anti-programmed death protein-1 monoclonal antibody treatment were subjected to steps including protein extraction, reductive alkylation, and enzymatic digestion to obtain sample peptides. The sample peptides are processed to obtain processed sample peptides. Based on the processed sample peptide data, basic feature analysis and functional annotation were performed to obtain the protein profile and modification profile of peripheral blood mononuclear cells from patients treated with anti-programmed death protein-1 monoclonal antibody. Based on the protein and modification profiles of peripheral blood mononuclear cells from patients treated with the anti-programmed death protein-1 monoclonal antibody, machine learning was used to screen for biomarkers of response to anti-programmed death protein-1 monoclonal antibody treatment. Biomarkers for the response to anti-programmed death protein-1 monoclonal antibody therapy in the non-small cell lung cancer patient samples were monitored to obtain the anti-programmed death protein-1 monoclonal antibody therapy response prediction model. The non-small cell lung cancer patient samples were classified into three subtypes, namely the prediction group, the drug resistance group, and the benefit group, based on the expression trends of the modular proteins and modifications. The inclusion criteria for the non-small cell lung cancer patient sample included: patients aged 18 years or older, patients clinically diagnosed with non-small cell lung cancer, patients with complete clinical information, and patients receiving anti-programmed death protein-1 monoclonal antibody therapy for the first time. The anti-programmed death protein-1 monoclonal antibody treatment response prediction model can predict the efficacy of anti-programmed death protein-1 monoclonal antibody treatment in patients with non-small cell lung cancer. Based on the protein and modification profiles of peripheral blood mononuclear cells from patients treated with the aforementioned anti-programmed death protein-1 monoclonal antibody, the step of screening for biomarkers of response to anti-programmed death protein-1 monoclonal antibody treatment using machine learning includes the following screening criteria: The quantitative values ​​of each protein in two compared sample groups were subjected to a T-test, and the corresponding P-values ​​were calculated as significance indicators. The criteria for screening differentially expressed proteins were a fold change > 2 and a P-value < 0.

05. The quantitative values ​​of each modification in two compared sample groups were subjected to rank-sum tests, and the corresponding P-values ​​were calculated as significance indicators. The false discovery rate method was used to correct the P-values, and the difference fold > 2 and P-value < 0.05 were used as the criteria for screening differential modifications. Random forest was used for simulation screening to further explore candidate biomarkers that can be used to predict the benefit of anti-programmed death protein-1 monoclonal antibody therapy. In the random forest analysis, 50,000 trees were constructed, and the maximum number of terminal node trees in the forest was set to 5. To evaluate the effectiveness of the model, subject survival characteristic curves were plotted, and the area under the curve (AUC) of the 95% confidence interval was calculated.

2. The method for constructing a predictive model for anti-programmed death protein-1 monoclonal antibody therapy response according to claim 1, characterized in that, In the step of obtaining samples from non-small cell lung cancer patients, wherein the patients are those treated with anti-programmed death protein-1 monoclonal antibody, The exclusion criteria for the non-small cell lung cancer patient samples included: the patient died during treatment with anti-programmed death protein-1 monoclonal antibody, or the patient was undergoing treatment with anti-programmed death protein-1 monoclonal antibody.

3. The method for constructing a predictive model for anti-programmed death protein-1 monoclonal antibody therapy response according to claim 1, characterized in that, The patient samples were peripheral blood mononuclear cells before and after treatment with anti-programmed death protein-1 monoclonal antibody. The process involved protein extraction, reductive alkylation, and enzymatic digestion to obtain sample peptides. The protein extraction included the following steps: Peripheral blood mononuclear cell samples were taken from the liquid nitrogen container, with a cell count of 2 × 10⁻⁶. 6 After thawing in a 37°C water bath, centrifuge at 1200 rpm for 3 min and discard the frozen solution. Prepare 1× medium-efficiency radioimmunoprecipitation lysis buffer (RIPA), wherein the volume ratio of RIPA to protease inhibitor is 99:

1. Add intermediate-efficiency radioimmunoprecipitation lysis buffer according to the amount of peripheral blood mononuclear cells, wherein each 1×10 6 Add 200 μL of medium-efficiency radioimmunoprecipitation lysis buffer to each cell. Two homogenization beads were added for lysis. Homogenization was performed once, with a frequency of 65 Hz and a total of 2 grinding cycles. The cycle was 45 s on, 10 s off, and incubated on ice for 10 min. This process was repeated twice. Centrifuge at 4 ℃ and 12000 rpm for 10 min, then collect the supernatant; Accurately pipette 200 μg of protein solution, add 3 times the volume of ice-cold acetone, and freeze at -80 ℃ for 1 h. 2h, After 1 hour, remove the protein and centrifuge at 12,000 rpm for 10 minutes at 4 °C. Discard the upper acetone layer and add 500 μL of 80% ice acetone to the lower protein layer. Vortex to mix and centrifuge again under the same conditions. Discard the supernatant. Add 200 μL of 50 mM ammonium bicarbonate solution, sonicate for 10 min, and incubate on ice to reconstitute the protein; Add 2 μL of 5 mM dithiothreitol to bring the final concentration to 0.05 mM, vortex to mix, then centrifuge at 1000 rpm for 1 min and in a 37 ℃ metal bath for 30 min. Add 6 μL of 5 mM iodoacetamide to bring the final concentration to 0.15 mM, vortex to mix, then centrifuge at 1000 rpm for 1 min, and incubate in a 55 ℃ metal bath for 30 min. Trypsin hydrolysis: Add 10 μL of 0.5 μg / μL sequencing-grade trypsin to make the protein to trypsin mass ratio 40:

1. Vortex to mix, then centrifuge at 1000 rpm for 1 min, digest in a metal bath at 37 ℃ with shaking at 900 rpm for 12 h, and add 5 μL of sequencing grade trypsin the next day, digest again under the same conditions for 2 h. After 2 hours, the sample was removed, and 2 μL of 0.1% formic acid was added. The reaction was terminated by vortexing for 3 seconds and then frozen at -80 °C to obtain the sample peptide.

4. The method for constructing the anti-programmed death protein-1 monoclonal antibody treatment response prediction model according to claim 3, characterized in that, The specific steps involved in processing the sample peptides to obtain the processed sample peptides are as follows: An octadecyl solid-phase extraction column is installed on a solid-phase extraction device, and a centrifuge tube is placed below the column to collect waste liquid. Activation: Add 500 μL of 100% acetonitrile to activate the octadecyl solid-phase extraction column, repeat twice; Equilibration: Add 500 μL of 0.1% formic acid to wash away residual acetonitrile, repeat twice; Add sample: Add the peptide to an octadecyl solid-phase extraction column, collect the eluent, repeat twice, and then discard the eluent; Desalting: Wash the column with 500 μL of 0.1% formic acid, repeating 2-3 times to remove as much salt as possible; Elution of peptides: Add 200 μL of 70% acetonitrile to elute the peptides, repeat twice, and collect the eluent; After desalting, peptides were quantified using a peptide concentration assay kit. Turn on the low-temperature freeze dryer, set the pressure to 150 kPa, and the cold trap temperature to -56 ℃ for pre-cooling. After the temperature drops to the predetermined value, place the sample, turn on the centrifuge and vacuum valve in sequence, and close the atmospheric valve. Vacuum concentrate and dry the 400 μL eluent collected by desalting for 3 h. After complete lyophilization, 50 μL of 0.1% formic acid prepared with mass spectrometry water was added to dissolve the lyophilized peptides. After thorough vortexing, the mixture was centrifuged at 12000 rpm for 10 min at 4 ℃. 45 μL of the supernatant was injected into high-performance liquid chromatography for fractionation to obtain the processed sample peptides.

5. The method for constructing the anti-programmed death protein-1 monoclonal antibody treatment response prediction model according to claim 4, characterized in that, The fractionation by the high-performance liquid chromatography preparative chromatography specifically includes the following steps: 100× Stock Solution: Dilute mass spectrometry grade concentrated ammonia to a concentration of 2 M. Take 6.7 mL of 2 M ammonia solution and add it to 43.3 mL of mass spectrometry water. Adjust the pH to 10.5 to prepare 50 mL of 100× stock solution and store at 4 °C for later use. Aqueous buffer solution: Add 10 mL of chromatographic grade acetonitrile to 485 mL of mass spectrometry-grade water, then add 5 mL of prepared 100× stock solution to prepare a 500 mL mixture of 97% mass spectrometry-grade water, 2% chromatographic grade acetonitrile, and 1% 100× stock solution. Organic phase buffer: Add 2 mL of mass spectrometry-grade water to 196 mL of chromatographic grade acetonitrile, then add 2 mL of prepared 100x stock solution to prepare a 200 mL mixture of 1% mass spectrometry-grade water, 98% chromatographic grade acetonitrile, and 1% 100x stock solution. 10% methanol: Take 10 mL of chromatographic grade methanol and add it to 90 mL of mass spectrometer-grade water, mix well, and you will get the product. After the required solution is prepared, tighten the cap and sonicate for 15 minutes to remove air bubbles from the liquid. Under the conditions of column temperature 40 ℃, sample tray temperature 7 ℃, and pressure 4000 psi, the power supply, column oven, fraction collector, computer, and UV monitor were turned on sequentially. A freshly prepared mobile phase was then used. Wet perfusion, syringe flushing, syringe needle cleaning, pump head cleaning, and UV monitoring were performed sequentially, collecting absorption peaks at 214 nm and 280 nm wavelengths. After injecting 45 μL of sample into the chromatographic column via an autosampler, gradient elution was started and the fractions were collected simultaneously. The fraction collector was programmed to collect one tube per minute for a total of 56 minutes, collecting 56 tubes. The first tube was discarded, leaving 55 tubes. These tubes were labeled and then combined in a specific order to form the final 20 fractions. The fractions were then centrifuged, concentrated, and lyophilized to obtain the processed sample peptides.

6. The method for constructing a predictive model for anti-programmed death protein-1 monoclonal antibody therapy response according to claim 1, characterized in that, For the processed sample peptides, basic feature analysis and functional annotation are performed to obtain the protein profile and modification profile of peripheral blood mononuclear cells from patients treated with anti-programmed death protein-1 monoclonal antibody. Specifically, this includes the following steps: The raw mass spectrometry data were searched using the proteomics discovery software Protein Discover 2.

4. Twenty fractions from each sample were combined and matched against a human protein database downloaded from a protein database to obtain protein identification and quantification results. The database search parameters were set as follows: the number of amino acids constituting the peptide was set to at least six, a maximum of two missed cleavage sites were allowed, and the mass tolerance of the precursor ion was set to [value missing]. Mass tolerance of precursor ion fragments in the range of 10 ppm – 10 ppm Within the range of 20 ppm; Protein quantification values ​​are standardized using the formula: protein abundance / (sample average protein / average protein abundance of all samples) to eliminate inter-sample errors in protein quantification values ​​caused by sample volume and instrument operation. Based on the results identified by Protein Discover 2.4, after removing contaminating proteins, peptide, protein, and precursor ion mass tolerance distribution analysis was performed to assess the quality of the mass spectrometry detection data. Basic characteristic analysis, annotation of functional databases including Gene Ontology and Kyoto Encyclopedia of Genes and Genomes database, and quantitative analysis were performed on the identified proteins, including overall difference analysis of identified proteins, screening of differentially expressed proteins, and expression pattern cluster analysis. Twenty raw files obtained from mass spectrometry of each sample were matched with a 2022 human protein library downloaded from the uniprot database using the open search engine OpenDelta. Systematic annotation of known modifications was performed using existing modification databases, with acetylation at the protein N-terminus and oxidation on methionine residues set as variable modifications. An open search was conducted for both known and unknown modifications, and the molecular weight deviation range of precursor ions was set. The molecular weight deviation range of fragment ions is set at 10 ppm - 10 ppm. 20 ppm - 20 ppm; The analysis results of the open search engine OpenDelta are calculated as (number of peptide matching maps of a certain modification at a certain protein site) / (total number of peptide matching maps of the unmodified protein site, the modified site, and other arbitrary modifications), in order to remove the influence of protein abundance. Based on the identification results of an open search engine, potential biomarkers were screened after removing in vitro modifications, including alkylation modifications introduced during pretreatment and oxidative modifications of methionine.

7. The method for constructing a predictive model for anti-programmed death protein-1 monoclonal antibody therapy response according to claim 1, characterized in that, Based on the protein and modification profiles of peripheral blood mononuclear cells from patients treated with the aforementioned anti-programmed death protein-1 monoclonal antibody, the step of screening for biomarkers of response to anti-programmed death protein-1 monoclonal antibody treatment using machine learning includes the following analytical methods: Protein expression pattern clustering analysis: Two methods were used to perform clustering analysis on differentially expressed proteins: heatmaps and fuzzy mean-based quasi-temporal sequence analysis (MFUZZ). Gene ontology and Kyoto Encyclopedia of Genes and Genomes Functional Enrichment Analysis: The ClueGO plugin of Cytoscape 3.8.1 was used to perform gene ontology and Kyoto Encyclopedia of Genes and Genomes Functional and Enrichment Analysis on differentially expressed proteins, mainly including biological pathway and pathway enrichment analysis. Protein interaction network analysis: The target protein was imported into the STRING database to obtain the protein interaction network, and the results were imported into Cytoscape 3.8.1 software to draw the protein interaction network diagram. The corresponding core modules and core nodes were obtained by using the Cytoscape internal plugins Mcode and cytoHubba, respectively.

8. The method for constructing a predictive model for anti-programmed death protein-1 monoclonal antibody therapy response according to claim 1, characterized in that, Based on the protein and modification profiles of peripheral blood mononuclear cells from patients treated with the aforementioned anti-programmed death protein-1 monoclonal antibody, the verification method in the step of screening for biomarkers of response to anti-programmed death protein-1 monoclonal antibody treatment using machine learning includes: The protein pretreatment for the validation cohort includes protein extraction, reduction, alkylation, enzymatic digestion, and desalting; Proteomics: The following criteria were used to select peptides of target proteins to establish a parallel reaction monitoring (PRM) method: 1) Only protein-specific peptides were selected; 2) The peptides contained no miscleavage sites; 3) The peptides contained no variable modification sites, oxidation modifications, or alkylation modifications; 4) Peptides that met the above requirements were selected based on their high response values; 5) At least one and a maximum of three protein-specific peptides were selected for each protein. Modification group: Establishing a parallel reaction monitoring method for target modified peptides: 1) No miscleavage sites in the peptide; 2) No variable modification sites, oxidation modification, or alkylation modification in the peptide; 3) Select peptides with high response values ​​that meet the above requirements; Samples were analyzed using a reversed-phase octadecyl self-packed capillary column (75 μm × 100 mm) and separated using an EASY-nLC1200 ultra-high performance liquid chromatography (UHPLC) system. Mobile phase A was an aqueous solution containing 0.1% formic acid and 2% acetonitrile; mobile phase B was an aqueous solution containing 0.1% formic acid and 90% acetonitrile. The gradient settings were: 0–46 min, 6%–28% B phase; 46–53 min, 28%–45% B phase; 53–60 min, 100% B phase. The flow rate was maintained at 350 nL / min, and the column temperature at 50 °C. After separation by the UHPLC system, peptides were injected into a sodium spray ion source at 2.1 kV and analyzed using Thermo Scientific™ Q Exactive ionization. The data were obtained using an HF-X Thermo Fisher Scientific mass spectrometer. The mass spectrometry data were acquired using a high-sensitivity parallel reaction monitoring mode with the following parameters: high-energy collisional dissociation as the fragmentation method, collision energy of 32%, first-order resolution of 60,000 m / z, second-order resolution of 15,000 m / z, scan range of 100 ~ 1800 m / z, scan time of 100 ms, screening of charge states from +2 to +4 including precursors, dynamic exclusion time of 30 s, isolation window of 4 Da, and time window of 7 min. Parallel reaction monitoring data processing was performed using Skyline 3.6 software. All results were imported into the software, the correct peaks were selected, and results for all peptides were obtained from all samples. The parallel reaction monitoring results included protein names and peak areas, which were further output for analysis, and differential proteins in different groups were screened and compared. Based on the relative quantitative results obtained from the Skyline peak area, we used R language to perform correlation analysis and visualization of the differences, and used R packages to plot the receiver operating characteristic curve and calculate the area under the curve.

Images