Application of urine small extracellular vesicle lipid marker combined with clinical parameters in prostatic cancer risk grading diagnosis product

Abstract

The invention discloses an application of urine small extracellular vesicle lipid markers combined with clinical parameters in a prostatic cancer risk grading diagnosis product, which is characterized in that the lipid markers are phosphatidylcholine (15: 0 / 18: 3), lysophosphatidylcholine (16: 1) and phosphatidyl ethanolamine (15: 0 / 20: 5), and the clinical parameters are blood PSA concentration and age of a to-be-detected person. The invention further provides a construction method of the prostatic cancer risk grading prediction model, which comprises the following steps: obtaining predictive factors for distinguishing four risk levels of prostatic cancer based on lipidomics analysis and machine model analysis, obtaining weight coefficients of the predictive factors through ORM model analysis, and determining the risk levels of the prostatic cancer according to the weight coefficients of the predictive factors. The method has the advantages that four-level risk grading of no cancer, low and medium risk, high risk and extremely high risk is realized, and the accuracy is high.

Description

Technical Field

[0001] The present invention belongs to the field of prostate cancer risk assessment, and specifically relates to the application of a urine small extracellular vesicle lipid marker combined with clinical parameters in a prostate cancer risk grading diagnostic product. Background Art

[0002] Prostate cancer (PCa) is a malignant tumor that arises in the prostate epithelium. Currently, the main diagnostic methods for prostate cancer include tumor markers, imaging studies, and prostate biopsy. Blood prostate-specific antigen (PSA) testing is the leading diagnostic method, but its specificity is low and it cannot provide early diagnosis or monitor prostate cancer progression. Imaging studies, including transrectal ultrasound, CT, and magnetic resonance imaging (MRI), have a certain diagnostic value for prostate cancer, but their value for early diagnosis is insufficient and cannot accurately be used to stage tumor progression. However, the widely used screening biomarker, prostate-specific antigen (PSA), has a false-negative rate of up to 40% in the "gray zone" of 4-10 ng / mL. Needle biopsy is an invasive procedure with risks such as pain, bleeding, and urinary retention. Furthermore, due to sampling errors, it cannot fully capture the full picture of multifocal lesions, and the sampling process is invasive. Therefore, there is an urgent need to develop new liquid biopsy methods that can comprehensively characterize tumor molecular features.

[0003] Liquid biopsy is a non-invasive diagnostic method and is currently a research hotspot in the field of clinical non-invasive diagnostics. Small extracellular vesicles (sEVs) are lipid bilayer nanoparticles actively secreted by tumor cells. The molecular information they carry can dynamically reflect tumor-specific biological processes. Compared with fragmented circulating tumor DNA, the unique biogenesis mechanism of sEVs allows for the complete preservation of lipids, proteins, and nucleic acids, thereby providing information about the tumor microenvironment with spatiotemporal resolution. The lipidome of sEVs is remarkably stable across individuals, demonstrating promising research potential.

[0004] Prostate cancer patient survival is largely correlated with tumor grade. Risk stratification can be used to assess the overall risk of metastasis from in situ prostate cancer, determine optimal treatment options, and predict the likelihood of recurrence after treatment. Furthermore, patients with metastatic prostate cancer have the worst prognosis, essentially eliminating the opportunity for surgical cure and relying solely on endocrine therapy or chemotherapy, which severely impacts survival. Therefore, accurate grading and early diagnosis of prostate cancer are crucial. Summary of the Invention

[0005] The application aims to provide an application of urine small extracellular vesicle lipid markers combined with clinical parameters in a prostate cancer risk grading diagnostic product, which realizes four-level risk grading of non-cancer (NC), low-moderate risk (LR), high risk (HR) and very high risk (VHR) and has high accuracy.

[0006] The application adopts the technical scheme that the application of urine small extracellular vesicle lipid markers combined with clinical parameters in a prostate cancer risk grading diagnostic product, wherein the lipid markers are phosphatidylcholine PC (15:0 / 18:3), lysophosphatidylcholine LPC (16:1) and phosphatidylethanolamine PE (15:0 / 20:5), the clinical parameters are blood PSA concentration and age of the person to be tested, and the four-level risk grading is non-cancer (NC), low-moderate risk (LR), high risk (HR) and very high risk (VHR).

[0007] Further, the lipid markers are from urine small extracellular vesicles.

[0008] Further, the prostate cancer risk grading diagnostic product includes a prostate cancer risk grading nomogram product.

[0009] Further, the nomogram product includes a carrier and a nomogram arranged on the carrier, and the nomogram includes the sum of the probabilities of the person to be evaluated in the low-moderate risk, high risk and very high risk of prostate cancer, the sum of the probabilities of the person to be evaluated in the high risk and very high risk of prostate cancer, and the probability of the person to be evaluated in the very high risk of prostate cancer, which are obtained by a prostate cancer risk grading prediction model.

[0010] Further, the construction method of the prostate cancer risk grading prediction model includes the following steps: Step 1, enriching sample set urine sEV, after quantitatively detecting lipid molecules in urine sEV of each sample, combining LASSO-Boruta-double feature selection algorithm to optimize marker combination, obtaining prediction factors for distinguishing four risk levels of non-cancer, low-moderate risk, high risk and very high risk, and the prediction factors are three lipid markers PC (15:0 / 18:3), LPC (16:1) and PE (15:0 / 20:5) and two clinical parameters PSA concentration and age; Step 2, obtaining the weight coefficient of each prediction factor by ORM model analysis, summing the product of the concentration of the three lipid markers of the person to be evaluated, PSA concentration and age value and the corresponding weight coefficient, and calculating the sum of the probabilities of the person to be evaluated in the low-moderate risk, high risk and very high risk of prostate cancer, the sum of the probabilities of the person to be evaluated in the high risk and very high risk of prostate cancer, and the probability of the person to be evaluated in the very high risk of prostate cancer by the cumulative distribution function of the standard normal distribution.

[0011] Furthermore, step 2 is as follows: The weight coefficients of each prediction factor are obtained through ORM model analysis, which are PSA = 0.0644, age = 0.0522, lipid molecule PC (15:0 / 18:3) = -0.0643, LPC (16:1) = -0.04653, PE (15:0 / 20:5) = -1.0923; the prediction model is constructed as follows: ; ; ; ; Wherein, PSA represents the PSA concentration of the sample to be tested, in ng / mL; Age represents the age of the person being tested; PC represents the concentration of phosphatidylcholine (15:0 / 18:3) in the urine sample to be tested, in ng / mL; LPC is the concentration of lysophosphatidylcholine (16:1) in the urine sample to be tested, in ng / mL; PE represents the concentration of phosphatidylethanolamine (15:0 / 20:5) in the urine sample to be tested, in ng / mL; Φ represents the cumulative distribution function of the standard normal distribution; It represents the sum of the probabilities that the person being evaluated is at low-medium risk, high risk, and very high risk of prostate cancer; It represents the sum of the probabilities that the person being evaluated is at high risk or very high risk of prostate cancer; Indicates the probability that the person being evaluated is at very high risk of prostate cancer.

[0012] Compared with the prior art, the advantages of the present invention are: 1. Based on lipidomics analysis, machine model analysis, etc., the present invention screened out lipid markers phosphatidylcholine (15:0 / 18:3), hemolysophosphatidylcholine (16:1), phosphatidylethanolamine (15:0 / 20:5) and two clinical parameters that can be used for prostate cancer risk stratification. These lipid markers are combined with clinical parameters to distinguish four risk levels: no cancer, low-medium risk, high risk, and extremely high risk; these lipid markers are derived from urine and are sampled non-invasively, which is non-invasive to the evaluators, and the non-invasive continuous sampling can dynamically reflect the pathological changes of the urinary system.

[0013] 2. Regarding the method for prostate cancer risk stratification, this method combines the individual's PSA level, age, and the concentrations of the lipid markers phosphatidylcholine (15:0 / 18:3), lysophosphatidylcholine (16:1), and phosphatidylethanolamine (15:0 / 20:5) to assess the individual's prostate cancer risk. This method, which combines lipidomics data with clinical characteristics, has been proven to have high accuracy. Specifically, compared to conventional blood tests using PSA alone, the average accuracy of this method increased from 0.55 to 0.77. Furthermore, this method improved the area under the curve (AUC) score by approximately 40% compared to conventional blood tests. This method overcomes the limitation of the traditional single PSA indicator, which often misses cases with low PSA levels and high risk of malignancy.

[0014] 3. The nomogram product for prostate cancer risk stratification uses scales from the first to the tenth rows. First, it provides visualization, making it easier for clinicians to access the data intuitively. Clinicians can quickly determine the prostate cancer risk of the individual being assessed at the corresponding level, thereby improving the speed and reliability of clinical decision-making. Second, it enables prostate cancer risk stratification assessment and can evaluate risk probabilities at different levels (three levels of risk probability data can be directly read from the nomogram. For example, the first level is the sum of the probabilities of the individual being assessed at low-medium risk, high risk, and extremely high risk for prostate cancer; the second level is the sum of the probabilities of the individual being assessed at high risk and extremely high risk for prostate cancer; and the third level is the probability of the individual being assessed at extremely high risk for prostate cancer). A simple subtraction of the values ​​from the first to third levels yields the probabilities of the individual being cancer-free, low-medium risk, high risk, and extremely high risk, respectively. This allows for a more accurate and detailed prostate cancer risk assessment for the individual being assessed, facilitating subsequent development of appropriate treatment plans by clinicians.

[0015] In summary, the present invention combines urine small extracellular vesicle lipid markers with clinical parameters in a prostate cancer risk stratification diagnostic product. After enriching urine sEVs using EXODUS® ultrafast separation technology, high-throughput lipidomics screening is performed using a triple quadrupole mass spectrometry system. The marker combination is optimized using the Boruta-LASSO dual feature selection algorithm. Lipid candidate markers validated through targeted quantitative analysis are then input into a prostate cancer risk stratification prediction model based on small extracellular vesicle (sEV) lipid markers constructed using the ORMs method. This ultimately achieves a four-tiered risk stratification: no cancer (NC), low-intermediate risk (LR), high risk (HR), and very high risk (VHR). BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The results of sEV morphology identification using transmission electron microscopy, where A is urine sEV and B is plasma sEV; Figure 2The results of the sEV particle size distribution test using a nanoparticle tracking analyzer, where A is UsEV, representing urine sEV, and B is PsEV, representing plasma sEV; Figure 3 The results of Western blotting for the detection of protein markers of sEVs are shown in Table 1, where PsEV indicates plasma sEVs, UsEV indicates urine sEVs, WP indicates whole plasma lysate, and WU indicates whole urine lysate. Figure 4 To show the results of detecting sEV surface protein markers using nano-flow cytometry, Figure A shows, from left to right, the different fluorescent staining of urine-derived sEVs: first, the expression level of CD9; second, the expression level of CD63; and finally, the co-expression level of CD9 and CD63; Column B shows, from left to right, the different fluorescent staining of plasma-derived sEVs: first, the expression level of CD9; second, the expression level of CD63; and finally, the co-expression level of CD9 and CD63. FITC represents yellow-green fluorescence, PE represents red fluorescence, FITC-CD9 indicates that CD9 protein is labeled with yellow-green fluorescence, and PE-CD63 indicates that CD63 protein is labeled with red fluorescence. Figure 5 To analyze the lipid composition and differences of urine sEVs using non-targeted lipidomics technology, A is a circular diagram of urine sEV lipid composition, where each color represents a lipid subclass, and the area size indicates the proportion; B is the difference in urine sEV lipid molecules between different groups, where the color indicates the type of lipid molecule, FA represents fatty acyl (light blue), GP represents glycerophospholipids (red), SP represents sphingolipids (green), GL represents glycerolipids (dark blue), PR represents prenol lipids (pink), and ST represents sterol lipids (purple). The numbers indicate the number of differential lipid molecules; Figure 6 To detect the lipid composition and differences of plasma sEVs using non-targeted lipidomics, A is a circular diagram of plasma sEV lipid composition, where each color represents a lipid subclass and the area represents the proportion; B is the difference in plasma sEV lipid molecules between different groups, where the color represents the class of lipid molecules, FA represents fatty acyl (light blue), GP represents glycerophospholipids (red), SP represents sphingolipids (green), GL represents glycerolipids (dark blue), PR represents prenol lipids (pink), and ST represents sterol lipids (purple), and the numbers represent the number of differential lipid molecules; Figure 7To screen key lipid molecules associated with prostate cancer grade from quantitatively verified lipid molecules, Figure A shows 22 key features with non-zero coefficients identified using the Lasso method, where red indicates a positive correlation (positive) with prostate cancer, and blue indicates a negative correlation (negative). Figure B shows the importance of lipid molecules assessed using the Boruta method, where Confirmed indicates truly important features, shadowMax indicates the highest importance score among all shadow features, shadowMean indicates the average importance score of all shadow features, and shadowMin indicates the lowest importance score of all shadow features. Figure C presents lipid molecules identified as important by both the Lasso and Boruta methods. Figure 8 Correlation analysis between PSA (A), age (B), phosphatidylcholine PC (15:0 / 18:3) (C), lysophosphatidylcholine LPC (16:1) (D), phosphatidylethanolamine PE (15:0 / 20:5) (E) and prostate cancer risk classification; Figure 9 Performance comparison of four machine learning methods and analysis of optimal model characteristics. A is the accuracy evaluation of the four machine learning methods. ORM represents the ordered regression model, rpartScore represents the ordered decision tree, ordinalForest represents the ordered random forest, and vglmContRatio represents the generalized additive model. B is the contribution analysis of each predictor in the optimal ordered regression model (ORM). Figure 10 To evaluate the sensitivity and specificity of the constructed risk stratification model, Figure A is the ROC curve of the model in the training set; Figure B is the ROC curve of the model in the validation set; Figure 11 The optimal nomogram for ORM constructed based on the weights of multimodal predictors. DETAILED DESCRIPTION

[0017] The present invention will be described in further detail below with reference to the accompanying drawings and embodiments.

[0018] Specific Example 1: Screening of urine small extracellular vesicle lipid markers for prostate cancer risk stratification.

[0019] 1. sEV Preparation (1) Patients and clinical samples

[0020] This study was approved by the Ethics Committee of the First Affiliated Hospital of Ningbo University (approval number: 2021-R106-02), and all participants provided written informed consent. Morning fasting peripheral blood and urine samples were collected from men with histopathologically confirmed prostate cancer between 2021 and 2023 and from non-cancer controls (NC) aged 50 years or older. Enrollment was based on risk stratification according to the 2022 Guidelines for the Diagnosis and Treatment of Prostate Cancer, with patients divided into low-intermediate risk (LR), high-risk (HR), and very high-risk (VHR) groups. All biological samples were collected before treatment.

[0021] Exclusion criteria included: ① concurrent infectious diseases; ② recent (within 3 months) treatment with medications that affect PSA levels; and ③ a history of other malignancies or invasive treatments. The NC group consisted of age-matched healthy volunteers and men without prostate cancer (e.g., men with benign prostatic hyperplasia) confirmed by pathology.

[0022] (2) Urine sample processing and sEV isolation Urine samples were standardized within 2 hours of collection: first, impurities were removed by gradient centrifugation (4000 × g for 10 min to remove cells / bacteria; 2000 × g at 4°C for 10 min to remove cellular debris). The supernatant was then centrifuged at 12,000 × g at 4°C for 30 min to enrich the vesicle fraction. The supernatant was filtered through a 0.22 μm filter (Merck Millipore, Germany) and purified using the EXODUS Exosome Isolation System (H300, Huixin Biotechnology, China). The isolated sEV pellet was resuspended in 200 μL of sterile PBS buffer (HyClone, USA), aliquoted, and stored in a -80°C freezer (Thermo Scientific) to avoid repeated freeze-thaw cycles.

[0023] (3) Plasma sample collection, preparation, and sEV isolation

[0024] Whole blood samples were preprocessed within 2 hours of collection: cellular components were removed by three-stage centrifugation (3000 × g, room temperature, 15 min to remove red blood cells; 3000 × g, 4°C, 15 min to remove cellular debris; and 13,000 × g, 30 min to remove large vesicles). 0.5 mL of plasma supernatant was mixed with pre-chilled PBS buffer at a 1:10 volume ratio and filtered through a 0.22 μm filter. Separation of sEVs was performed using the EXODUS system (H300, Huixin Biotechnology, China). The resulting sEV pellet was resuspended in 200 μL of PBS buffer and immediately frozen at −80°C.

[0025] (4) sEV identification

[0026] Transmission electron microscopy morphological analysis: sEV ultrastructure characterization was performed using a H-7650 transmission electron microscope (Hitachi High-Tech, Japan). 10 μL sEV suspension was dropped onto a copper grid (Carbon film, Beijing Zhongjing Keyi) and adsorbed at room temperature for 10 min, then rinsed with deionized water, negatively stained with 2% uranyl acetate for 1 min, and the residual liquid was removed by the filter paper edge adsorption method. The grid was cured in a desiccator in the dark. The observation conditions of the electron microscope were set as follows: accelerating voltage 80 kV, equipped with Gatan Orius CCD image acquisition system. Representative images of urine and plasma sEVs obtained by transmission electron microscopy are shown in Figure 1 As shown in Figure 1 Fig. 1A and Figure 1 Fig. 1B, urine and blood sEVs both showed typical cup-shaped morphology and complete lipid bilayer structure (red arrow), indicating that the collected urine sEVs and plasma sEVs both met the morphological standard of exosomes.

[0027] Nanoparticle tracking analysis: sEV particle size distribution and concentration detection was performed using a NanoSight NS300 system (Malvern Panalytical, UK). The sample was diluted with sterile PBS buffer to the detection concentration (1×10 8 -1×10 9 particles / mL), and then injected into the nano-flow detection cell (NanoCell, part number: MK1). Under the parameter conditions of laser wavelength 405 nm, camera sensitivity 13, and detection threshold 5, 5 sets of 60 s dynamic video were continuously collected for each sample. Data analysis was performed using NTA 3.4 software, and the particle size distribution histogram results are shown in Figure 2 As shown in Figure 2 Fig. 2A and Figure 2 Fig. 2B, nanoparticle tracking analysis of urine and plasma sEVs revealed a bimodal distribution characteristic, with the main peak particle size of urine sEVs (abbreviated as UsEV) being 147.1±6.4 nm and that of plasma sEVs (abbreviated as PsEV) being 124.3±5.4 nm, which met the exosome particle size standard range.

[0028] Western blot (abbreviated as WB): total protein was obtained by oscillating lysis of the sample at 4°C for 30 min using 5×RIPA lysis buffer (Cell Biolabs, USA) containing 1×Halt™ protease / phosphatase inhibitor cocktail (Thermo Scientific, USA). After ultrasonic treatment (15 s / time, with an ice bath interval of 10 s, repeated for 4 cycles) of the lysis product in an ice water bath, the protein concentration was determined using a BCA protein quantification kit (Vazyme, E112). As shown in Figure 3As shown, Western blotting confirmed that UsEV and PsEV highly expressed specific markers HSP70, FLOT1, CD63, Syntenin-1, and CD9, while no exosome contamination markers APOA / THP were detected, indicating that both UsEV and PsEV met the marker criteria for exosomes.

[0029] Nanoflow cytometric analysis of sEV surface markers: sEV membrane surface marker expression was detected using a CytoFLEX Nano flow cytometer (Beckman Coulter, Pasadena, California). Samples were incubated with AF488-conjugated mouse anti-human CD9 antibody (NanoFCM, NHA009-A488-50T, 1:10 dilution) and PE-conjugated anti-human CD63 antibody (BioLegend, 353003, 1:20 dilution) at room temperature for 1 h in the dark. The samples were diluted 100-fold in PBS buffer before analysis. Unstained antibody samples served as isotype controls. Population fluorescence intensity was analyzed using FlowJo V10.8 software, using threshold parameters: side scatter (SSC) gain of 20% and forward scatter (FSC) threshold of 1 × 10 3 nm. Figure 4 The results of the detection of sEV surface protein markers using nano-flow cytometry are shown in Figure 2. Figure 4 Middle A shows, from left to right, the urine-derived sEVs stained with different fluorescent stains: first, the CD9 protein was stained with yellow-green fluorescence to detect the expression level of CD9 alone, with a CD9 positivity rate of 17.27%; second, the CD63 protein was stained with red fluorescence to detect the expression level of CD63 alone, with a CD63 positivity rate of 28.91%; finally, the two fluorescent proteins were incubated simultaneously to detect the co-expression of CD9 and CD63, with a co-expression positivity rate of 7.75%. Figure 4 Figure B shows, from left to right, plasma-derived sEVs stained with different fluorescent markers. Similarly, CD9 protein was first stained yellow-green, with a CD9 positivity rate of 20.32%. CD63 protein was then stained red, with a CD63 positivity rate of 1.19%. Finally, the two fluorescent markers were incubated simultaneously, with a co-expression rate of CD9 and CD63 of 9.29%. Comparison of the distribution of surface markers in urine and plasma sEVs reveals heterogeneity in the expression of CD9 and CD63 in sEVs from different sources.

[0030] In summary, the comprehensive morphology, particle size distribution, molecular characteristics and surface protein expression data confirmed that sEVs have high homogeneity and low contamination characteristics, meeting the strict technical requirements of lipidomic biomarker research.

[0031] 2. Lipid marker screening (1) Non-targeted lipidomics analysis

[0032] Lipids were extracted using a two-phase extraction method with methyl tert-butyl ether / methanol (MTBE:MeOH = 3:1, v / v). The organic phase was collected after vortexing and centrifugation (12,000 × g, 10 min), concentrated under nitrogen purge, and then reconstituted in isopropanol. Separation was performed using an ExionLC AD ultra-high performance liquid chromatography system (SCIEX, USA) with an ACQUITY UPLC® HSS T3 column (2.1 × 100 mm, 1.8 μm, Waters). The mobile phases A and B were water (0.1% formic acid) and acetonitrile / isopropanol (90:10, v / v, 0.1% formic acid), with a 30-min gradient elution. Mass spectrometry was acquired using a QTRAP 6500+ system (SCIEX) in full-scan mode (m / z 100–1500). Electrospray ionization parameters were: source voltage 5500 V, temperature 500°C.

[0033] The lipid composition characteristics of sEVs in urine and plasma of low-risk prostate cancer patients (n=3), high-risk prostate cancer patients (n=3), and controls (n=6) were analyzed by non-targeted lipidomics technology. Figure 5 and Figure 6 shown. Figure 5 Figure A shows the lipid composition of urinary sEVs, visualizing the relative abundance distribution of each lipid class. The results show that phosphatidylethanolamine (PE, 8.1%), phosphatidylcholine (PC, 7.83%), and ceramide (Cer-NS, 6.12%) dominate, with multiple specific molecules including PC (16:0 / 20:1), PC (18:0 / 22:6), and PE (P-16:0 / 20:5). Figure 6 Figure A shows the lipid composition characteristics of plasma sEVs, which is in sharp contrast to urine sEVs. Its metabolic pattern is mainly triglyceride (TG), with a relative abundance of TG as high as 27.4%, significantly higher than other lipid categories. At the same time, it can be seen that PC and sphingomyelin (SM) together constitute the characteristic lipid spectrum of plasma sEVs. Figure 5 Middle B and Figure 6 Figures B and C show that when comparing the control, low-risk, and high-risk groups, a high proportion of glycerophospholipid (GP) compounds showed significant differential expression among the groups. For example, in the comparison between NC (normal control group) and LR (low-medium risk group), the number 32 indicates that 32 differential lipid molecules belong to the GP class. This suggests that this lipid subclass may have specific response characteristics in the occurrence and progression of prostate cancer.

[0034] (2) Targeted lipid quantitative detection

[0035] A multiple reaction monitoring (MRM) method was established using a triple quadrupole mass spectrometer (AB Sciex 6500+) for quantitative detection of lipid molecules. Samples were pretreated with 1,2-diheptanoyl-sn-glycero-3-phosphocholine (DHPC) as an internal standard. Separation was performed on a Waters ACQUITY UPLC BEH C8 column (2.1×50 mm, 1.7 μm) at 45°C and a flow rate of 0.3 mL / min. Mass spectrometry parameters included collision energy of 15–40 eV, a dwell time of 20 ms, and acquisition in positive and negative ion switching modes. Quantitative analysis was performed using Analyst TF 1.8 software. A six-point calibration curve (R² > 0.99) was established based on standards, and relative lipid content was calculated normalized to the internal standard.

[0036] Through targeted validation with an expanded sample cohort (NC = 19, LR = 31, HR = 22, VHR = 26), a total of 607 lipid molecules were quantitatively detected. Sample details are shown in Table 1. This grouping strategy balanced statistical power with intergroup balance, ensuring the scientific nature of the study design.

[0037] Table 1 Baseline clinicopathological characteristics of prostate cancer patients

[0038] (3) Screening of key lipid molecules associated with prostate cancer risk stratification LASSO regression analysis and Boruta feature selection algorithm were used to screen variables. Correlation analysis was performed using the R language "glmnet" and "Boruta" packages. Specifically, lipid molecules with regression coefficients greater than 0 in the LASSO regression model were first extracted. A total of 22 non-zero coefficient features were screened, including two clinical parameters (PSA and age) and 20 lipid molecules, such as Figure 7 As shown in A, red indicates a positive correlation with prostate cancer, and blue indicates a negative correlation. The Boruta algorithm is then used to identify important lipid molecules, such as Figure 7 As shown in Figure B, features greater than shadowMax are determined to be confirmed, indicating truly important features. The importance of PSA and age is also confirmed, and core lipid markers such as PE (15:0 / 20:5), PC (15:0 / 18:3), and LPC (16:1) are also identified. Finally, the key feature indicators are determined to be PSA, age, PC (15:0 / 15:0), PC (15:0 / 18:3), PC (15:0 / 20:4), LPC (16:1), and PE (15:0 / 20:5) through the intersection strategy. Figure 7As shown in Figure C. To avoid multicollinearity, PSA, age, PC (15:0 / 18:3), LPC (16:1), and PE (15:0 / 20:5) were finally selected to construct the diagnostic model.

[0039] (4) Clinical parameters and expression characteristics of key lipid molecules in prostate cancer risk stratification The nonparametric Kruskal-Wallis test was used to analyze the differences in the expression of PSA, age, and lipid molecules (PC (15:0 / 18:3), LPC (16:1), PE (15:0 / 20:5)) in the control group (NC), low-risk group (LR), high-risk group (HR), and very high-risk group (VHR). Figure 8 shown.

[0040] Depend on Figure 8 From A, we can see that there are extremely significant differences in PSA among the four groups. From NC to VHR group, PSA concentration shows a clear upward trend. Figure 8 From Figure B, we can see that there are significant differences in age among the four groups. The age of VHR patients is significantly higher than that of the low-risk group and the control group. Figure 8 From C, we can see that PC (15:0 / 18:3) has extremely significant differences among the four groups, and it shows a downward trend from NC to VHR. Figure 8 D shows that LPC (16:1) has a very significant difference among the four groups, and it shows a downward trend from NC to VHR; Figure 8 Figure E shows that PE (15:0 / 20:5) was significantly different among the four groups, with a decreasing trend from NC to VHR. With increasing risk stratification, the average PSA level and age increased, while the concentrations of the three lipids were negatively correlated.

[0041] Specific embodiment 2: Construction and validation of a prostate cancer risk grading ordinal diagnostic model.

[0042] 1. Data partitioning and balance verification

[0043] Stratified random sampling was performed using the "tidymodels" package in R. A separate set of 98 prostate cancer samples was divided into a training set (76 patients) and a validation set (22 patients) in a 7:3 ratio. Cancer stage was used as the stratification variable to ensure that the proportions of each stage (NC / LR / HR / VHR) between the two groups were consistent with those in the original dataset. Statistical analysis of baseline clinicopathological characteristics (age, T stage, and Gleason score) verified the balanced distribution of the two groups. Table 2 shows that the training and validation sets were balanced in key parameters such as median age (interquartile range), T stage distribution, and Gleason score (p>0.05, no significant differences). This demonstrates that stratified sampling effectively avoids data bias and provides a reliable foundation for model training and validation.

[0044] Table 2 Comparison of baseline characteristics of the prostate cancer diagnostic model cohort

[0045] 2. Construction of ordinal diagnostic model Next, four ordinal regression models (ordered regression model ORM, ordered decision tree rpartScore, ordered random forest ordinalForest, and generalized additive model vglmContRatio) were used to construct a prostate cancer risk prediction model, and the model performance was evaluated based on the training set and validation set. The results are as follows Figure 9 Specific methods include: ordered regression model (ORM): using the "rms" package to model the linear relationship between stage probability and predictor variables (PSA, age, and lipid molecules) with the probit link function; ordered decision tree (rpartScore): using the "rpartScore" package to generate tree-like classification rules; ordered random forest (ordinalForest): using the "ordinalForest" package to integrate multiple decision trees; and generalized additive model (vglmContRatio): using the "VGAM" package to explore nonlinear effects.

[0046] like Figure 9 As shown in Figure A, model performance is measured using the Acc@1 (Accuracy@1) metric, which indicates the accuracy of the model's predictions on the validation set—that is, the proportion of the highest-probability predicted class that exactly matches the actual class. The results show that the ordinal regression model (ORM) performs best on the validation set (0.8), while the ordered decision tree (rpartScore) has a lower Acc@1, suggesting poor classification performance. Notably, the ordered random forest (ordinalForest) and generalized additive model (vglmContRatio) have high Acc@1 (close to 1.0) on the training set, but significantly decrease on the validation set, indicating a risk of overfitting.

[0047] Further analysis of the contribution of each predictor in the ORM model is as follows Figure 9Figure B shows that PSA (coefficient = 0.0644, p < 0.001) was significantly positively correlated with prostate cancer risk stratification, while lipid molecules PC (15:0 / 18:3) (coefficient = -0.0643, p < 0.05), LPC (16:1) (coefficient = -0.04653, p < 0.05), and PE (15:0 / 20:5) (coefficient = -1.0923, p < 0.05) were negatively correlated. These results not only validate the reliability of PSA as a traditional risk marker but also reveal the potential role of specific lipid molecules in risk stratification. Overall, the ORM model emerged as the optimal predictive tool in this study due to its stable predictive performance and interpretability. Acc@1, as an evaluation metric, effectively reflects the model's classification accuracy in the independent validation set.

[0048] Based on the comprehensive analysis results of the ORM model, the maximum likelihood estimation method was used to construct a probability equation for predicting prostate cancer risk grading, and its expression formula is: ; ; ; ; Wherein, PSA represents the PSA concentration of the sample to be tested, in ng / mL, and the value range of PSA is 0~160; Age represents the age of the person being tested, with a value range of 56 to 80; PC represents the concentration of phosphatidylcholine (15:0 / 18:3) in the urine sample to be tested, in ng / mL, with a value range of 15 to 5; LPC is the concentration of lysophosphatidylcholine (16:1) in the urine sample to be tested, in ng / mL, with a value range of 0~12; PE represents the concentration of phosphatidylethanolamine (15:0 / 20:5) in the urine sample to be tested, in ng / mL, with a value range of 0.8–2.2; Φ represents the cumulative distribution function of the standard normal distribution; It represents the sum of the probabilities that the person being evaluated is at low-medium risk, high risk, and very high risk of prostate cancer; It represents the sum of the probabilities that the person being evaluated is at high risk or very high risk of prostate cancer; Indicates the probability that the person being evaluated is at very high risk of prostate cancer.

[0049] 3. Model diagnostic effectiveness evaluation The diagnostic performance of the prostate cancer risk grading prediction model was systematically evaluated by multi-classification ROC curve analysis. Figure 10 As shown in , the model exhibits excellent discriminative performance: Figure 10 In the training set of medium A, the macro average AUC (Macro, equally weighted for each risk level) reached 0.91, and the micro average AUC (Micro, weighted based on sample size) was 0.89; Figure 10 In the independent validation set of ZhongB, both the macro- and micro-AUCs remained stable at a high level of 0.90. The curves show that specificity (true negative rate) on the horizontal axis reflects the model's ability to exclude non-cases, while sensitivity (true positive rate) on the vertical axis reflects the model's ability to identify cases. Even when specificity was set to the clinically high standard of 0.9, the model maintained high sensitivity. Notably, the macro-AUC focuses on assessing the model's ability to discriminate between different risk categories, while the micro-AUC more comprehensively reflects the model's overall performance in the overall sample. The high consistency between the training and validation set results (macro / micro AUC differences <0.02) further confirms the robustness and reliability of the model. These data strongly demonstrate that the prediction model not only accurately stratifies patients with different prostate cancer risks (macro-AUC >0.9) but also demonstrates excellent and stable diagnostic performance in the overall population (micro-AUC >0.89), providing a powerful decision-making tool for clinical practice.

[0050] Specific embodiment 3: Development of clinical application tool: nomogram product for prostate cancer risk stratification.

[0051] Based on the risk grading probability model constructed in the second embodiment, a clinical application tool is developed: a nomogram product for prostate cancer risk grading, the nomogram product includes a carrier and a nomogram set on the carrier, the carrier being a card and / or a computer. Figure 11 As shown in the figure, the standardized coefficients of each variable are converted into contribution scores of 0-200 points through a nomogram. Clinicians can quickly calculate the total risk score by superimposing the scores and map it to the risk probability. The nomogram includes the first to tenth rows; Among them, the first line is the score scale; The second to sixth rows are scales for the PSA concentration in the blood of the person to be evaluated, the age of the person to be evaluated, the concentration of phosphatidylcholine (15:0 / 18:3) in the urine of the person to be evaluated, the concentration of lysophosphatidylcholine (16:1), and the concentration of phosphatidylethanolamine (15:0 / 20:5), respectively, and correspond to the corresponding scores in the first row; The seventh row is the total score scale. Add the scores of the second to sixth rows corresponding to the first row to get the corresponding total score. The eighth line is a scale that represents the sum of the probabilities of low-risk, high-risk, and very high-risk prostate cancer; The ninth line is a scale for the sum of the probabilities of high-risk and very high-risk prostate cancer; The tenth row is a scale for the extremely high risk probability of prostate cancer; The total score obtained in the seventh row corresponds to the scales in the eighth, ninth, and tenth rows, respectively, representing the prostate cancer risk probability of the corresponding level of the person to be evaluated.

[0052] Specifically, the scale ranges of the first to seventh rows are 0~100, 0~160, 56~80, 55~15, 12~0, 2.2~0.8, 0~200, 0.01~0.99, 0.01~0.99, and 0.01~0.99 respectively; The scale values ​​of the fourth to sixth rows decrease equally from left to right, and the scale values ​​of the remaining rows increase equally from left to right.

[0053] The left end scales of the second to sixth rows are aligned with the left end scale of the first row, and the right end scales of the second to sixth rows are aligned with the scale values ​​100, 12.5, 25, 54.2, and 15 of the first row respectively; The left end scale of the eighth row is aligned with the scale value 49.6 of the first row, and the right end scale of the eighth row is aligned with the scale value 92.0 of the first row; The left end scale of the ninth row is aligned with the scale value 63.6 of the first row, and the right end scale of the ninth row is aligned with the scale value 106.0 of the first row; The left end scale of the tenth row is aligned with the scale value 76.0 of the first row, and the right end scale of the tenth row is aligned with the scale value 118.4 of the first row.

[0054] This tool breaks through the limitations of the traditional PSA single threshold and enables instant visual assessment of risk probability.

[0055] In summary, this application integrates urine sEV lipidomics with machine learning algorithms to identify five core features from 98 prostate cancer patients, combining clinical parameters (PSA, age) with lipid markers (PC (15:0 / 18:3), LPC (16:1), and PE (15:0 / 20:5)). This model then constructs an ordinal regression model for a four-tiered prostate cancer risk stratification (NC / LR / HR / VHR). This model demonstrates excellent diagnostic efficacy in both training and independent validation datasets, overcoming the limitations of traditional PSA testing in underdiagnosing low-PSA, high-risk cases. This model, for the first time, enables precise prostate cancer risk stratification based on noninvasive liquid biopsy. By leveraging the synergistic effects of lipid metabolism abnormalities and clinical parameters, this model provides an innovative solution for overcoming the bottlenecks of existing diagnostic systems in risk stratification.

[0056] The above description is not intended to limit the present invention, and the present invention is not limited to the above examples. Any changes, modifications, additions or substitutions made by persons of ordinary skill in the art within the spirit and scope of the present invention shall also fall within the scope of protection of the present invention.

Claims

1. Application of urine small extracellular vesicle lipid markers combined with clinical parameters in a prostate cancer risk stratification diagnostic product, characterized by: The lipid markers are phosphatidylcholine (15:0 / 18:3), hemolysophosphatidylcholine (16:1) and phosphatidylethanolamine (15:0 / 20:5), and the clinical parameters are the blood PSA concentration and age of the person to be tested, which are used for four-level risk classification: no cancer, low-medium risk, high risk and extremely high risk.

2. The use according to claim 1, characterized in that: The lipid markers are derived from urine small extracellular vesicles.

3. The use according to claim 1, characterized in that: The prostate cancer risk grading diagnostic product includes a prostate cancer risk grading nomogram product.

4. The use according to claim 3, characterized in that: The nomogram product includes a carrier and a nomogram disposed on the carrier, wherein the nomogram includes the sum of the probabilities of the person to be assessed being at low- to moderate-risk, high-risk, and very high-risk for prostate cancer, calculated using a prostate cancer risk grading prediction model; The sum of the probabilities that the person to be evaluated is at high risk and very high risk of prostate cancer, and the probability that the person to be evaluated is at very high risk of prostate cancer.

5. The use according to claim 4, characterized in that The steps of the method for constructing the prostate cancer risk grading prediction model are as follows: Step 1: Enrich urine sEVs from the sample set. After quantitatively detecting lipid molecules in each urine sEV, the LASSO-Boruta dual feature selection algorithm was used to optimize the marker combination to obtain predictive factors for distinguishing four risk strata: no cancer, low-intermediate risk, high risk, and very high risk. The predictive factors were three lipid markers (PC (15:0 / 18:3), LPC (16:1), and PE (15:0 / 20:5)) and two clinical parameters (PSA concentration and age). Step 2: Obtain the weight coefficient of each predictive factor through ORM model analysis. After summing the products of the concentrations of the three lipid markers, PSA concentration, and age value of the person to be evaluated and their corresponding weight coefficients, calculate the sum of the probabilities of the person to be evaluated being at low-medium risk, high risk, and extremely high risk of prostate cancer through the cumulative distribution function of the standard normal distribution; the sum of the probabilities of the person to be evaluated being at high risk and extremely high risk of prostate cancer; and the probability of the person to be evaluated being at extremely high risk of prostate cancer.

Images