Development of parallel reaction monitoring (PRM) assays for the detection of aggressive prostate cancer using urinary glycoproteins

Abstract

Recently, we have found two urinary glycoproteins, Prostatic Acid Phosphatase (ACPP) and Clusterin (CLU) combined with serum prostate-specific antigen (PSA) can serve as a three-signature panel for detecting aggressive prostate cancer (PCa) based on a quantitative glycoproteomic study. To facilitate the translation of candidates into clinically applicable tests, robust and accurate targeted parallel reaction monitoring (PRM) assays that can be widely adopted in multi-labs were developed in this study. The developed PRM assays for the urinary glycopeptides, FLN*ESYK from ACPP and EDALN*ETR from CLU, demonstrated good repeatability and sufficient working range covering three to four orders of magnitude, and their performances in differentiating the aggressive prostate cancer were assessed by the quantitative analysis of urine specimens collected from 69 non-aggressive (Gleason score = 6) and 73 aggressive (Gleason ≥ 8) PCa patients. When ACPP combined with CLU, the discrimination power was improved from an AUC of 0.66 to 0.78. By combining ACPP, CLU and serum PSA to form a three-signature panel, the AUC was further improved to 0.83 (sensitivity: 84.9%, specificity: 68.1%). Since the serum PSA test alone had an AUC of 0.68, our results demonstrated that the new urinary glycopeptide PRM assays can serve as an adjunct to the serum PSA test to achieve better predictive power toward aggressive PCa. In summary, our developed PRM assays for urinary glycopeptides were successfully applied to clinical PCa urine samples with a promising performance in aggressive PCa detection.

Graphical Abstract

Introduction

Prostate cancer (PCa) is the most common diagnosed male malignancy and the second-leading cause of death for men in developed countries¹. Most PCa is indolent at the time of diagnosis. The indolent PCa is slow growing and poses limited threat to patients’ life even without therapy intervention applied. Once the PCa is developed into a more advanced stage, it will progress rapidly increasing mortality rate of patients; thus, systematic and intensified therapy interventions will be required^2,3. Currently, the clinical characterization of the aggressiveness of the PCa is mainly assessed by serum prostate-specific antigen (PSA) testing, digital rectal examination (DRE), repeated prostate tissue needle biopsies to derive the Gleason score and TNM staging system⁴. Due to the low specificity of serum PSA testing aggressiveness of PCa as well as the pain and complications caused by the invasive tissue biopsy, it is crucial to identify non-invasive biomarkers associated with aggressive PCa and implement the biomarkers into robust clinical testing methods to guide PCa risk stratification.

Urine is an appealing substrate for the discovery of non-invasive biomarkers associated with PCa, since urinary system is proximal to the prostate gland and may contain molecular signatures shed or secreted from diseased prostate tissue, such as tumor cells, DNA/RNA and proteins⁵. Indeed, isolation of circulating tumor cells from urine of PCa patients was achieved by the microfluid chip strategy and the amount of tumor cells displayed moderate correlation to the Gleason score⁶. Great efforts have also been directed toward the investigation of urine-derived genetic biomarkers including long non-coding RNA, micro-RNA, DNA, and gene fusion. For example, prostate cancer antigen-3 (PCA3) is the first urinary biomarker approved by US Food and Drug Administration (FDA) for PCa; assessments of its individual performance as well as in combination with urinary TMPRSS2:ERG gene fusions as diagnostic or prognostic biomarkers for PCa were conducted in multiple studies^7-12. Additionally, a urine exosome derived gene expression assay was used to predict high-grade PCa at the initial biopsy, which displayed a good discrimination power in differentiating Gleason 6 group from Gleason ≥ 7 group¹³. A previous study reported a three-gene panel (HOXC6, TDRD1, and DLX1) selected from urinary sediments for the early detection of PCa with biopsy Gleason ≥7¹⁴. Proteomic signatures obtained from urine specimens were also investigated for their ability in distinguishing aggressive PCa^15,16. Although many urinary candidate biomarkers were proposed for aggressive PCa detection, only a few of them entered clinical trials; many candidates still require to further validate their clinical utility^5,17.

In our previous study, an automated high-throughput urine sample preparation platform was coupled with data-independent acquisition (DIA) of mass spectrometry (MS) to discover urinary glycoproteins associated with aggressive PCa, of which glycopeptides from the urinary glycoproteins, Prostatic Acid Phosphatase (ACPP) and Clusterin (CLU), demonstrated promising results¹⁸. Serum ACPP, instead of urinary ACPP, was once served as the world’s first clinically useful biomarker for PCa until it was replaced by serum PSA test as the latter demonstrated improved predictive power for early stage PCa^19,20. Nonetheless, our recent studies have suggested the emerging roles of ACPP as a potential prognostic biomarker in detecting aggressive PCa¹⁸. ACPP is a prostate specific protein and its abundance is fifty times higher in prostate tissue than tissues from other organs in human²¹; hence, its abundance variation in different urine specimens may reflect the secretory function of the prostate gland or the progression of PCa. Therefore, FLN*ESYK (an N-linked glycopeptide, * indicates the glycosite) from ACPP identified via our DIA-based approach is worth further evaluation by targeted proteomics. Besides ACPP, glycopeptide EDALN*ETR from CLU showed association with aggressive PCa via DIA. CLU is a glycoprotein involved in diverse biological events including cell proliferation and cell death; it is also associated with tumor progression and neurodegenerative disorders²². CLU represents as one of the multi-functional proteins whose expression is altered in both inflammation and cancer²². Although the investigation of the specific role of CLU in PCa is still ongoing^23,24, the up-regulated glycopeptide from CLU discovered in our previous study still worth further explored using a targeted approach.

To develop accurate and easily extendible quantitative strategy which can be applied to large-scale clinical cohorts as well as in different laboratories ^25,26, in this study, targeted parallel reaction monitoring (PRM) assays were established for the two glycopeptides from ACPP and CLU, which were discovered in our previous work using DIA-MS approach¹⁸. The developed PRM assays were subsequently applied to the first cohort consisted of 142 PCa urine samples to examine the performance of candidate glycopeptide assays in distinguishing aggressiveness of PCa by receiver operating characteristic (ROC) analysis in repeated 10-fold cross validation with logistic regression. Finally, the PRM assays were further evaluated for their capability in supplementing each other and serum PSA test to achieve a better discrimination power towards aggressive PCa detection. The results indicated that accurate and robust PRM assays were successfully developed for urinary glycopeptides and they can provide additional predictive value for the diagnosis of aggressive PCa relative to serum PSA test.

Materials and Methods

Detailed information on chemicals and reagents, automated sample preparation and enrichment of urinary glycopeptide, LC-MS/MS analysis is provided in Supplemental Methods.

Urine samples

Post-DRE raw urine specimens from PCa patients of first cohort (69 of Gleason score = 6, 73 of Gleason score ≥ 8) and second cohort (13 of Gleason score = 6, 40 of Gleason score = 8, used for validation) were collected and processed by the Department of Urology at Johns Hopkins University School of Medicine following the EDRN’s PCA3 urine processing standard operating procedure (SOP). The collected urine samples were labeled with specimen identification number and stored at −80 °C freezer. The information of the urine samples and the associated prostate cancer patients were obtained with approval from the Institutional Review Board of Johns Hopkins University and informed consent. The first cohort was the same set of samples used in our previous DIA discovery study¹⁸ to demonstrate the possibility of translating DIA MS-discovered candidates into PRM assays. However, the experimental design, sample preparation, data acquisition, and data analysis were independently from the previous published work. Detailed information on the urine specimens is listed in Supplemental Table S1.

Database search and PRM assay evaluation

PRM raw files were analyzed by Skyline (version 20.1.0.76) for peak integration and quantification of the targeted glycopeptides. All peaks were inspected manually to ensure the correct detection of precursor and fragment ions. The heavy (H) and light (L) peptide transition, retention time and peak boundary were also used to confirm peptide identity. A minimum of four transitions was required for the correct detection of the target peptides. The MS responses of endogenous glycopeptides (i.e., light peptides) identified in each urine sample were normalized to their corresponding heavy isotope-labeled internal standards for relative quantification. The ratio of L/H were exported for further statistical analysis.

The calibration standards were prepared using glycopeptides isolated from pooled PCa urine samples as background to mimic the condition of real clinical samples as well as to control matrix effects. To determine the linear dynamic range of MS response, a 12-point dilution series of heavy isotope-labeled standard peptides (5, 2.5, 1, 0.5, 0.25, 0.1, 0.05, 0.025, 0.01, 0.005, 0.001, and 0.0005 pmol per injection) were spiked into the PCa urine derived glycopeptides background (from 50 ul of urine sample per injection) and were measured by PRM in triplicate for their MS response. Due to the wide dynamic range of the abundance of our candidate glycopeptides, a 1/x² weighting was performed when generating the linear regression of the calibration curves.

The limit of detection (LOD) and limit of quantitation (LOQ) of the targeted glycopeptides were determined based on the standard deviation of the linear regression curves using the following formula²⁷, LOD = 3.3 * Sa / b, LOQ = 10 * Sa / b, where Sa is the standard deviation of the y-intercepts and b represents the slope of the calibration curves.

Statistical Analysis

All the analyses were carried out in R (version 3.5). The expression fold change of the targeted glycopeptides between aggressive and non-aggressive PCa groups were computed and the p-values were calculated using Mann-Whitney U test. For each glycopeptide, its discriminatory power as an individual marker or in combination with others through logistic regression was evaluated by ROC analysis in two repeated 10-fold cross validations. The mean ROC curves from the repeated 10-fold cross validations were depicted and area under a curve (AUC) was computed for the mean ROC curves. The predictive models with cross validation were built using caret²⁸ (version 6.0-85). ROC curves were generated using pROC²⁹ (version 1.13) whereas AUC along with 95% confidence interval (95% CI) as well as sensitivity and specificity at the best cutoff point along with 95% CI were obtained via MLeval (https://CRAN.R-project.org/package=MLeval). The best cutoff point on ROC curve generated the maximal summed sensitivity and specificity. The generated predictive models were further investigated using the second cohort.

Results

Overview of the experimental workflow

In our recent study, two glycopeptides from urinary protein ACPP and CLU were discovered with promising results in detecting aggressive PCa based on quantitative DIA analysis¹⁸. To evaluate the clinical utility of the candidate glycopeptides and facilitate the translation of the MS-based candidate biomarkers to routine clinical implementation in future, we developed easily extendable PRM quantitative assays for the aforementioned urinary glycopeptides. We also applied the assays to clinical urine specimens, which were quantitatively analyzed in our previous discovery study using DIA to assess the performance of the candidates in separating aggressive PCa from non-aggressive PCa. The schematic workflow of this study is illustrated in Fig. 1. In brief, urine specimens were processed using an automated high-throughput platform to isolate glycopeptides^30,31. The PRM assays were developed for the two candidate glycopeptides. The analytical performances of the PRM assays were assessed using heavy-isotope labeled synthetic peptides spiked in the pooled urinary glycopeptides. The enriched urinary glycopeptides from each urine sample were subjected to targeted quantitative LC-MS/MS analysis using the PRM assays. Finally, the relative quantification results of targeted glycopeptides were statistically analyzed to evaluate the clinical performance of PRM assays in combination with serum PSA for discrimination power toward aggressive PCa.

Fig. 1. Schematic overview of the experimental workflow.

Urinary glycopeptides were first isolated from the clinical samples using an automated high-throughput method. The PRM assays were developed using heavy isotope-labeled peptides and the analytical performances were determined. The isolated glycopeptides along with the spike-in heavy isotope-labeled peptides were quantified using the newly established targeted PRM assays. The quantification results were statistically analyzed and the clinical performance of the candidate peptides in detecting aggressive PCa were investigated and further evaluated using a second cohort.

Establishment and characterization of the PRM assays

To develop a targeted PRM assay with high accuracy and high repeatability, it is essential to establish a linear relationship between the measured MS signal response (i.e., intensity) of a targeted peptide and its quantity in the relevant biological matrix background³² (e.g., serum or urine). The measured intensity can reflect true abundance of the targeted peptide if only the MS signal response falls into the established linear detection range. Therefore, we constructed the reversed calibration curves for our candidate glycopeptides by spiking in a serial dilution of heavy isotope-labeled standard peptides to generate 12-point linear curves, spanning from 0.5 fmol/injection to 5,000 fmol/injection, into urinary glycopeptides as the sample matrix background to assess the linear relationship between the MS signals of targeted glycopeptides and their quantities under the real sample matrix condition. Each dilution point was analyzed in triplicate using PRM and the linearity of the generated curves was evaluated by regression analysis. The LOD and LOQ were determined based on the derived linear curves (Materials and Methods). The top four transitions from each peptide were used to calculate the response ratios with regards to reduce the influence of ion selection and improve the assay robustness.

The generated 12-point linear curves for the heavy isotope-labeled glycopeptide of ACPP (FLN*ESYK[+8]) indicated that the working range of the PRM assay covered about four orders of magnitude from 1.3 fmol to 5,000 fmol, with the coefficient of determination value (R²) of 0.996 (Fig. 2A). The LOD and LOQ of the PRM assay for glycopeptide FLN*ESYK from ACPP were determined to be 0.4 fmol and 1.3 fmol, respectively. In addition, the extracted ion chromatography for PRM transitions of this glycopeptide and its heavy isotope-labeled counterpart demonstrated a consistency among the transitions since the transitions had similar peak shapes and retention time indicating no interference with contaminants and thus reliable quantification was achieved (Figs. 2B-C). For the heavy isotope-labeled glycopeptide EDALN*ETR[+10] from CLU, as the peptide concentration increased, the linear correlation between the concentration and MS signal response became weaker; thus, data points that were outside the linear range were therefore removed. Consequently, a 9-point linear calibration curve was constructed with R² of 0.997; the effective working range ranged from 1.7 fmol to 1000 fmol, which covered three orders of magnitude (Fig. 2D). Reliable quantification of the glycopeptides was observed with consistency among different transitions (Figs. 2E-F). Overall, LOD and LOQ of glycopeptide EDALN*ETR from CLU were 0.5 and 1.7 fmol, respectively.

Fig. 2. The analytical performance of the established PRM assays for glycopeptides from urinary CLU and ACPP.

A. Reversed calibration curves for FLDN*ESYK from ACPP. B. The extracted ion chromatography for the PRM transitions of heavy isotope-labeled peptide FLDN*ESYK[+8] from ACPP. C. The extracted ion chromatography of PRM transitions of endogenous peptide FLDN*ESYK from ACPP. D. Reversed calibration curves for glycopeptide EDALN*ETR from CLU. E. The extracted ion chromatography of the PRM transitions of heavy-isotope labeled peptide EDALN*ETR[+10] from CLU. F. The extracted ion chromatography of the PRM transitions of endogenous peptide EDALN*ETR from CLU. # Calibration curves of the triplicate along with their average are plotted. The linear regression model and R² values were calculated based on the average calibration curve.

Repeatability of PRM quantification is crucial in candidate biomarker assessment, especially when a large cohort of samples is analyzed since it will take a much longer period of time for data collection. To examine the intra-day (with-in day) variability of the PRM assays, four repeated PRM analyses were conducted using the same test sample on the same day (the test sample were prepared by the addition of 0.1 pmol heavy isotope-labeled peptides per injection into the urine-derived glycopeptide background). The obtained peak area for each transition as well as for each peptide across the four repeated analyses was compared (Figs. 3A-B). The coefficient of variations (CVs) for the peptides and transitions were less than 6% (data shown in Supplemental Table S2), indicated the consistency of PRM measurement (Figs. 3A-B). To assess the inter-day variance of the PRM assays, we further utilized the data collected for the response curves to conduct a validation since (1) the replicates of each concentration were collected across 5 days, which was suitable for inter-day variation evaluation; (2) the amount of the heavy isotope-labeled peptides spanning from 0.5 fmol to 5,000 fmol, covered the entire working range of the established PRM assays. As shown in Fig. 3C, the CVs for the peptide response are less than 10% (on average 4.4%) with at least 5 fmol peptides. On the other hand, for the two data points with the lowest concentration (1 fmol or 0.5 fmol spike-in heavy peptides), relatively larger CVs were obtained. Notably, when a medium abundance (around 100 fmol) of heavy peptides were analyzed in the urinary glycopeptide sample, the PRM assays for both glycopeptides (FLN*ESYK and EDALN*ETR) yielded the smallest CVs, which were less than 2% (Fig. 3C). Therefore, 150 fmol heavy isotope-labeled peptides were spiked into the individual urine-derived glycopeptide sample in our following experiment. In summary, PRM assays of both urinary glycopeptides from ACPP and CLU were successfully established with good repeatability and sufficient working range covering three to four orders of magnitude.

Fig. 3. Reproducibility of the established PRM assays.

A. PRM analysis of the targeted glycopeptides from ACPP was performed in four replicates. The intensities (peak area) of the top four PRM transitions for heavy isotope-labeled peptide FLDN*ESYK[+8] and endogenous peptide FLDN*ESYK from ACPP across the four replicates are displayed in (i) and (ii), respectively. B. The intensities of the top four PRM transitions for heavy-isotope labeled peptide EDALN*ETR[+10] and endogenous peptide EDALN*ETR from CLU across the four replicates are displayed in (i) and (ii), respectively. C. The data points from the calibration curves of the triplicate (data was collected across 5 days) were used to assess the repeatability of the PRM assays. The CVs of the measured heavy and light ratios at different heavy isotope-labeled peptide spike in levels are plotted. #H/L ration is the ratio between heavy isotope-labeled peptide and endogenous peptide.

Implementation of the developed PRM assays to individual urine specimens

To facilitate the translation of candidate markers that we previously discovered for aggressive PCa using a DIA approach into a clinical setting in the future, we applied the PRM assays to the quantitative analysis of glycopeptides enriched from PCa urine specimens in order to evaluate their discrimination power in differentiating aggressive and non-aggressive PCa via such targeted glycoproteomic approach using PRM. First, glycopeptides isolated from individual urine samples were analyzed by PRM together with heavy isotope-labeled internal standard peptides. Upon examining the extracted ion chromatography of the transitions from the endogenous glycopeptides to ensure accuracy in peak picking, we found that the two targeted glycopeptides were quantified in all of the 142 urine samples with diverse intensities (Supplemental Figs. S1A and S1C). For example, the intensity of FLN*ESYK from sample S124 was 1440 times higher than sample S114 (Supplemental Fig. S1A). The diversity among the clinical samples were plausible since each sample had distinct abundances of candidate proteins. On the other hand, the MS intensities of heavy isotope-labeled peptides were less diverse and more evenly distributed among the samples since the same level of peptides were spiked into each sample (Supplemental Figs. S2A and S2C). The variation of heavy isotope-labeled peptides was expected since each sample had distinct abundances of proteins and glycopeptides, which contributed to the differences in signal suppression effect of our heavy isotope-labeled peptides. Therefore, the addition of heavy isotope-labeled internal standard peptides is essential to achieve accurate quantification result by utilizing the ratio between endogenous glycopeptides and their heavy isotope-labeled counterparts for relative quantification of our endogenous glycopeptides. Although, the MS intensities of the PRM transitions seemed distinct among samples, however, the proportion of intensity (i.e., the proportion of each transition to the sum of the four transitions in each sample) of each transition was consistent across the 142 samples (Supplemental Figs. S1B, S1D, S2B and S2D) implying the stable fragment pattern and the correct determination of the targeted glycopeptides. Hence, the responses of the endogenous glycopeptides relative to heavy were exported from Skyline and used for further statistical analysis.

Evaluation of the performance of the PRM assays in detecting Aggressive PCa

To examine the performance of the glycopeptides from ACPP and CLU in distinguishing groups of samples with different aggressiveness of PCa, PRM quantification results (Supplemental Table S3) of the targeted glycopeptides were statistically analyzed. The expression profile of the glycopeptide from ACPP was significantly decreased in aggressive PCa relative to non-aggressive PCa, with 2.8-fold differences (p-value <0.01; Fig. 4A and Supplemental Table S4). By contrast, the glycopeptide from CLU showed higher expression level in the aggressive PCa group with a fold change of 1.54 (p-value<0.05; Fig. 4B and Supplemental Table S4). Using serum PSA as a reference, we observed that the median of total serum PSA levels in aggressive PCa group was 1.73 times higher than the non-aggressive group (Fig. 4C and Supplemental Table S4).

Fig. 4. Expression of ACPP and CLU based on the ratio of endogenous glycopeptides to heave-isotope labeled peptides and their predictive power towards aggressive PCa via the established PRM assays along with serum PSA.

A. Expression profiles of the glycopeptide from ACPP between aggressive (Gleason 8, n=73) and non-aggressive (Gleason 6, n=69) PCa groups. B. Expression profiles of glycopeptide from CLU between aggressive (Gleason 8, n=73) and non-aggressive (Gleason 6, n=69) PCa groups. C. Expression profiles of serum PSA between aggressive (Gleason 8, n=73) and non-aggressive (Gleason 6, n=69) PCa groups. D. The performance of urinary ACPP, urinary CLU and serum PSA in detecting aggressive PCa individually. E. The performance of urinary ACPP, urinary CLU, or serum PSA test combined into as two- or three-signature panels.

We further evaluated the discriminatory power of the PRM assays of glycopeptides from ACPP and CLU as well as serum PSA test as an individual marker signature in separating the two PCa subgroups by ROC analysis. As shown in Fig. 4D, ACPP has an AUC of 0.66 (95 % CI: 0.57 to 0.75), while CLU has an AUC of 0.60 (95 % CI: 0.50 to 0.69) (Supplemental Table S5) suggesting the glycopeptide from ACCP has better separation capability. However, when the two glycopeptides were combined together to form a two-signature panel, the predictive power towards aggressive PCa increased with an AUC of 0.78 (95 % CI: 0.70 to 0.86; specificity of 58% at 87.7% sensitivity) (Fig. 4E and Supplemental Table S5). Serum PSA had a moderate performance as an individual signature (AUC of 0.68, 95% CI: 0.59 to 0.77, Fig. 4D); as we combined ACPP, CLU and serum PSA to create a three-signature panel, both urinary ACPP and CLU could supplement serum PSA test to improve the overall predictive power (Fig. 4E and Supplemental Table S5) achieving an AUC of 0.83 (95% CI: 0.77 to 0.90; specificity of 68.1% at 84.9% sensitivity). By further evaluating the PRM assays using urine specimens from an independent cohort (13 non-aggressive and 40 aggressive PCa patients), we observed similar outcome for the glycopeptide from ACCP (AUC=0.71, 95% CI: 0.54 to 0.87) and the two-signature panel composed of ACPP and serum PSA (AUC=0.81, 95% CI: 0.68 to 0.93). Detailed PRM quantification and results of the second cohort are in Tables S6-7. Compared with our previous study using DIA, we observed similar outcome using the current targeted PRM methods. The two glycopeptides from urinary ACPP and CLU demonstrated consistent performance and capability in differentiating different PCa subgroups. Therefore, the results of this study elucidated that the PRM assays were successfully developed for urinary glycopeptides and were applicable to the quantitative analysis of targeted peptides from real clinical specimens as well as supplementing serum PSA test to gain improved discrimination power.

Discussion

To avoid the possibility of overtreating indolent PCa as well as providing appropriate therapy to high-risk PCa patients, it is crucial to develop easy-adapted techniques with high precision to make candidate biomarkers clinically available for detecting aggressive PCa to aid patient risk stratification and therapy selection. We previously discovered two urinary glycopeptides, FLN*ESYK from ACPP and EDALN*ETR from CLU, as candidate markers for detecting aggressive PCa via DIA-approach¹⁸. To facilitate the translation of MS-discovered candidates to clinical lab as well as to further clarify their clinical utility, robust and accurate PRM assays that can be easily adopted in multi-labs were developed in the current study (Fig. 1).

Non-invasive liquid biopsy based on urine is a tool for detecting cancer and monitoring cancer progression, which has received great attention³³. However, high composition complexity of urine and the possibility of contamination can cause ionic suppression of the targeted peptides when analyzed under complex sample background. Therefore, it poses a great challenge for LC-MS/MS based quantification and hinder the quantitative accuracy³⁴. To circumvent these issues, reversed calibration curves were generated using a serial dilution of heavy isotope-labeled standard peptides (0.5 fmol to 5,000 fmol) spiking into the sample matrix background to establish the linear relationship between MS signals of targeted glycopeptides and their quantities under the real sample matrix condition; thus, ensuring the detected MS signals falling into the linear detection range and reflect true abundance of the targeted peptides. Each dilution point was analyzed in triplicate using PRM for the glycopeptides of ACPP and CLU (Fig. 2). The working range of the PRM assay for urinary ACPP covered from 1.3 fmol (LOQ) to 5000 fmol with LOD of 0.4 fmol, whereas the working range was 1.7 fmol (LOQ) to 1000 fmol for urinary CLU with 0.5 fmol. By assessing intra-day and inter-day variabilities of the established PRM assays, CVs <6% and <10% were achieved, respectively, indicating the PRM assays had good repeatability (Fig. 3).

The developed PRM assays were applied to 142 PCa urine specimens for evaluating their performances in detecting aggressive PCa (Fig. 4). Compared to the performance of individual candidates, a two-signature panel consisted of urinary CLU and urinary ACPP as well as the three-signature panel with both glycoproteins and serum PSA achieved better performance in differentiating aggressive PCa and non-aggressive PCa with AUCs of 0.78 and 0.83, respectively (Fig. 4E). Since serum PSA test along can only provide an AUC of 0.68 indicating the established PRM assays for urinary glycoproteins can be effective supplements to serum PSA test in PCa stratification. Similar outcome was achieved using the current targeted PRM methods compared to our previous study using DIA approach suggesting consistent performance of urinary ACPP and CLU in detecting aggressive PCa. In summary, we developed PRM assays for urinary glycopeptides from CLU and ACPP showing high precision and providing reliable results in detecting aggressive PCa as well as in supplementing serum PSA test to further improve discrimination power towards the diagnosis of aggressive PCa. The developed PRM assays can be applied to large sample cohort or multi-center study for further validation.

The expression profile of glycopeptides may be attributable to either the expression of their corresponding proteins or the glycan occupancy of the glycosylation sites. Measurement on the total protein level can be an alternative strategy if the expression profiles of the protein and the glycopeptide shows similar trend. However, it will not work if the change in glycopeptide abundance is due to the increase in glycosylation occupancy. However, PRM assays specifically developed for glycopeptides can work in both situations.

In this study, we intended to measure the abundance of targeted glycopeptides enriched from the same volume of urine; thus, the abundance of targeted glycopeptides from the same volume of urine (50 μL, one-tenth of 500 μL) were compared across the whole sample cohort without further normalization. The rationale of using the same volume of urine for prostate protein analysis assumes that the abundance of proteins secreted from prostate to urine is independent of rest of protein contents from urine due to fact that urine collects prostatic components in urethra, while total urine protein content can come from kidney and bladder.

Supplementary Material

Supporting Information

Figure S1, Intensity and proportion of intensity of the PRM transitions from endogenous glycopeptides across the discovery cohort (142 clinical urine specimens)

Figure S2, Intensities and proportion of intensity of the PRM transitions from heavy isotope-labeled peptides across the 142 clinical urine specimens (PDF).

Supporting Tables

Table S1, information of the first cohort (142 clinical urine specimens) and second cohort (for validation, 53 clinical urine specimens)

Table S2, reproducibility of the established PRM assays

Table S3, PRM quantification result of glycopeptides from ACPP and CLU in the 142 urine specimens

Table S4, statistical analysis of the expression levels of glycopeptides from urinary ACPP and CLU between AG PCa and NAG PCa from 142 urine samples

Table S5, ROC analysis results of urinary ACPP, CLU, and serum PSA of first cohort

Table S6, PRM quantification result of glycopeptides from ACPP and CLU of second cohort

Table S7, validation of the predictive models using second cohort (XLSX).

Abbreviations:

ACPP

Prostatic Acid Phosphatase

CLU

Clusterin

PSA

prostate specific antigen

PCa

prostate cancer

PRM

parallel reaction monitoring

DRE

digital rectal examination

PCA3

prostate cancer antigen-3

FDA

Food and Drug Administration

DIA

data-independent acquisition

MS

mass spectrometry

LOD

limit of detection

LOQ

limit of quantitation

95% CI

95% confidence interval

CVs

coefficient of variations

Data availability:

The LC-MS/MS data have been deposited to the PeptideAtlas repository³⁵ with the dataset identifier: PASS01676.