Proteomic discovery of non-invasive biomarkers of localized prostate cancer using mass spectrometry

Abstract

Prostate cancer is the second most frequently diagnosed non-skin cancer in men worldwide. Patient outcomes are remarkably heterogeneous and the best existing clinical prognostic tools such as International Society of Urological Pathology Grade Group, pretreatment serum PSA concentration and T-category, do not accurately predict disease outcome for individual patients. Thus, patients newly diagnosed with prostate cancer are often overtreated or undertreated, reducing quality of life and increasing disease-specific mortality. Biomarkers that can improve the risk stratification of these patients are, therefore, urgently needed. The ideal biomarker in this setting will be non-invasive and affordable, enabling longitudinal evaluation of disease status. Prostatic secretions, urine and blood can be sources of biomarker discovery, validation and clinical implementation, and mass spectrometry can be used to detect and quantify proteins in these fluids. Protein biomarkers currently in use for diagnosis, prognosis and relapse-monitoring of localized prostate cancer in fluids remain centred around PSA and its variants, and opportunities exist for clinically validating novel and complimentary candidate protein biomarkers and deploying them into the clinic.

Prostate cancer is the second most diagnosed non-skin cancer in men worldwide¹, and will account for ~26% of non-skin cancer diagnoses and ~11% of cancer-related deaths in men in the United States in 2021 (REF²). Many prostate cancers are indolent in nature, but others are aggressive and are either diagnosed when already metastasized or rapidly spread beyond the prostate, leading to morbidity and potential prostate cancer-specific mortality³. Distinguishing indolent from aggressive disease to enable personalized treatment regimens is, therefore, a key challenge in prostate cancer research⁴.

Screening, diagnosis and prognosis of localized prostate cancer remain heavily reliant on clinical features such as serum PSA level, clinical T category and clinical International Society of Urological Pathology Grade Group (ISUP GG) that do not have optimal validity (sensitivity and specificity) for non-indolent disease^5–7. For example, both upgrading (~20–30%) and downgrading (~10%) could occur between needle biopsy and radical prostatectomy around the commonly used boundary between clinically insignificant (ISUP GG = 1) and clinically significant disease (ISUP GG ≥2)^8,9. Thus, the best existing risk-stratification schemes, such as the National Comprehensive Cancer Network and European Association of Urology risk groups systems, still lead to overtreatment and under-treatment for individual patients¹⁰. Men with low-risk disease managed non-curatively have a 8.9% cumulative prostate cancer-specific mortality at 15 years (compared with a 49.5% risk of dying from other causes)¹¹, whereas patients with actively managed radiorecurrent, high-risk prostate cancer experience rapid metastases and have a 30% chance of dying from their prostate cancer¹². Current management pathways are in need of complementary biomarkers that accurately reflect prostate cancer aggressiveness at screening and diagnosis, and improve risk stratification^13,14. Similarly, biomarkers that indicate clinically meaningful progression and can be used for frequent longitudinal monitoring are needed to support follow-up monitoring during active surveillance (AS) or after local definitive treatment (FIG. 1).

Fig. 1 |. The role of biomarkers in prostate cancer management.

Common tests and treatment modalities in the management of patients with prostate cancer and the applicability of molecular markers at each stage. At clinical diagnosis and staging, molecular biomarkers can help to inform whether initial or repeat biopsy is required and to improve knowledge in instances in which PSA is elevated but multiparametric magnetic resonance imaging (mpMRI) and transrectal ultrasonography (TRUS)-guided biopsy are negative. During active surveillance, fluid-based biomarkers can be used more frequently than TRUS-guided biopsy and mpMRI, therefore, enabling clinicians to rapidly identify patients whose disease has progressed. Biomarkers could be used in conjunction with PSA for monitoring patients following definitive local treatment, and could direct patients with probable clinically significant biochemical recurrence (BCR) for further imaging. At the advanced-disease stage, fluid-based biomarkers can be used for predicting a patient’s response to therapy, for which no metric exists. In addition, biomarkers can be used to detect low-volume metastases and early signs of androgen deprivation therapy failure. DRE, digital rectal examination; PET, positron emission tomography.

Identifying molecular biomarkers that can complement existing clinical tools for diagnosis, prognosis and disease monitoring has been a major component of prostate cancer research since the early 2000s. Fluids that are in contact with and/or are in close proximity to the prostate — prostate-associated fluids — have always held promise as sources of molecular biomarkers. Protein abundance can be measured accurately and precisely in a targeted manner in fluids, making proteins a promising molecule for biomarker discovery in prostate-associated fluids^15–16. With increasing interest in personalized medicine and efforts to profile tumour proteomes through The Human Protein Atlas, The Cancer Genome Atlas and the Clinical Proteomic Tumour Analysis Consortium, great opportunities exist for prostate cancer protein-based biomarker discovery and validation^17,18.

Protein abundance.

A measure of the number of proteins in a sample, estimated from direct measurements of peptide abundance.

Mass spectrometry.

(MS). An analytical technique that separates ions by mass:charge ratio in a mass analyser.

Proteomics.

The large-scale study of proteomes, which are sets of proteins produced in a biological context such as by a cell, tissue or organism.

Mass spectrometry-based proteomics can be used to detect and quantify thousands of proteins as well as their post-translational modifications¹⁹. Mass spectrometry is a biophysical technique that enables the structural analysis of various biomolecules in the form of gas phase ions, resulting in their detection and quantification. In this Review, we outline mass spectrometry (MS) technologies that enable the systematic discovery and targeted validation of protein-based biomarkers in prostate-associated fluids. We discuss clinically relevant biospecimens for protein-based biomarker discovery and highlight candidate protein-based biomarkers applicable at each step of the clinical decision-making pathway. We also discuss the challenges of clinical translation and application of these biomarkers and present potential solutions.

Need for biomarkers in disease management

Diagnostic, prognostic and predictive biomarkers enable patient-specific management of prostate cancer. Serum PSA quantification, digital rectal examination (DRE) and systematic transrectal ultrasonography (TRUS)-guided biopsies form the cornerstone of prostate cancer management^13,14, but they have been increasingly supplemented by molecular assays, germline testing and multiparametric magnetic resonance imaging (mpMRI), which provide complementary information^20–22. However, novel biomarkers are still needed at every step of prostate cancer management to improve personalized patient care.

Some of the limitations in the clinical management pathways for prostate cancer can be attributed to the weaknesses in the key clinical assays used. For example, no single value of PSA is able to achieve simultaneously high sensitivity and specificity, as elevated PSA levels can indicate non-cancer states such as BPH or prostatitis²³. Similarly, DRE has both low sensitivity and specificity (<60% for both) in the primary-care setting⁵. TRUS-guided biopsies are prone to undersampling, especially when tumour volume is low and prostate volume is high (with a false-negative rate of up to 30%⁷), and miss prostate cancer in the anterior zone, which can account for up to 20% of prostate cancers²⁴. In addition to pain and discomfort, biopsies are also associated with a small but substantial risk of potentially life-threatening urosepsis²⁵. Switching to a transperineal approach could reduce the risk of infection, although it is still associated with pain and discomfort²⁶.

Many fluid-based molecular biomarkers, such as the Prostate Health Index, 4Kscore, Progensa, MIPs, SelectMDx, ExoDX Prostate, Apifiny and Proclarix, have been developed to support prediction of aggressive disease upon biopsy or repeat biopsy²¹. These tests are typically approved for limited use in the diagnostic setting, but no consensus exists on which biomarker should be used in which scenario²⁷. Approved tissue-based molecular biomarkers, including ProMark, OncotypeDX, Decipher and Prolaris, have received positive recommendations for use in patients with very-low-risk or low-risk treatment-naive disease for deciding between AS or definitive therapy, but no randomized control trials have studied their utility, comparative effectiveness or economic efficiency, limiting their uptake¹⁴. Additionally, needle biopsies are needed to obtain tissue-based biomarkers, which are associated with morbidity and are limited by the relatively low volume and spatially heterogeneous disease in the prostate, resulting in potential sampling bias²⁸.

mpMRI has emerged as a promising tool to improve detection of clinically significant tumours (negative predictive value 63–98%)^29,30. When combined with systematic biopsy, mpMRI-targeted biopsy also enables diagnosis of more prostate cancer and more clinically significant disease than either method alone^29,30. However, up to 35% of clinically significant tumours in any lesion remain invisible to mpMRI, including high-grade (Gleason pattern 4 and 5) tumours^31,32. Sequential mpMRI in AS can be used to indicate the need for repeat biopsy or potentially replace biopsy completely, but no conclusive evidence has shown an association between mpMRI progression and pathological upgrading^33,34. Furthermore, interobserver variability, reader experience, limited availability of 3-T instruments in non-academic settings, and the cost associated with mpMRI can hinder the timely and frequent adoption of mpMRI in prostate cancer management^35,36.

The current goal of prostate cancer biomarker discovery studies is to develop assays that provide complementary information to existing modalities at each step of disease management. Prostate cancer is highly heterogenous between and within patients; thus, combining biomarkers with existing modalities has the most potential for achieving personalized management in diagnosis, prognosis and treatment prediction. To support prostate cancer screening and diagnosis, a biomarker needs to be inexpensive, reproducible and non-invasive to enable its longitudinal application in large populations. Importantly, a screening and diagnostic biomarker should be developed to detect only clinically significant disease when used alone or in combination with mpMRI after suspicious PSA and/or DRE analysis. The biomarker should have a high negative predictive value, so that negative results in patients with benign or clinically insignificant disease will enable them to confidently avoid biopsies. Ideally, such a diagnostic pathway would remove the need for biopsy as a diagnostic tool and reserve it for the use of tissue-based risk stratification for patients requiring active management. At the least, a diagnostic biomarker should provide information to distinguish between BPH and prostate cancer when combined with PSA, meaning that it should be most informative in the PSA range of 4–10 ng/ml, in which the outcome is most uncertain³⁷

A diagnostic biomarker that is able to predict aggressive disease can also be used for risk stratification in men diagnosed with prostate cancer. The key biomarker of interest in this setting is one that has high sensitivity for the prediction of disease progression in order to encourage men with very-low to favourable-intermediate-risk disease and negative results to select and remain on AS to avoid overtreatment. As prostate cancer is spatially heterogeneous and influenced by microenvironmental factors such as hypoxia^38,39, a biomarker that reflects the presence of aggressive disease anywhere in the prostate and indicates unfavourable tumour microenvironmental conditions (such as a protein biomarker that is elevated in patients with hypoxic tumours) could provide holistic information for prognosis. In order for the same prognostic biomarker to be used for monitoring during AS, it should be non-invasive for ease of frequent sampling. A highly standardized assay with results that can be transferred across health-care providers is also desirable for the long-term surveillance of these patients. The ideal biomarker should also indicate meaningful disease progression so that re-biopsy is only performed to confirm upgrading in these patients.

Serum PSA testing for monitoring of biochemical recurrence (BCR) remains the basis of follow-up monitoring after local treatment^13,14. The ideal biomarker for monitoring disease progression after definitive local therapy shares many of the same characteristics desirable in a biomarker for monitoring patients on AS. Its primary goal would be to complement PSA in the determination of clinically relevant BCR, as the definition of BCR is based on PSA but not all BCR results in morbidity or increased mortality⁴⁰. Ideally, the biomarker would also provide additional prognostic and predictive information upon disease progression. Biomarkers for monitoring after local treatment could provide a low-cost tool that is complementary to imaging-based assessments, such as mpMRI, bone scans, abdominopelvic CT and PET, and help the early detection of recurrent disease. Studies to discover such a biomarker will require multidisciplinary collaboration and the collection of a sizable matching cohort of biospecimens and imaging results from patients with extensive follow-up data.

Easily accessible fluids that can be collected using routine, standardized protocols can provide the basis of biomarker assays that need to be performed non-invasively and serially. Serum-based and urine-based biomarkers are especially attractive in the setting of prostate cancer, as they could easily be collected and assayed alongside PSA. Fluids might also be more likely to fully capture diverse information that reflects intratumoural, microenvironmental and systemic conditions than tissue biopsies, which are limited to small regions of prostate tissue. Urine is particularly appealing for its potential to be collected outside of medical facilities during times of interrupted health-care availability. Fluid-based molecular biomarkers can also be measured with highly standardizable and reproducible targeted MS assays that can be transferred between clinical laboratories⁴¹.

Fluid-based protein biomarkers are particularly attractive for several reasons. First, proteins are more stable in fluids than RNA or DNA^15,16. Second, processes associated with cancer such as angiogenesis, cell adhesion and migration are mediated by proteins at the plasma membrane, which are shed into prostate-associated fluids and can be detected at high concentrations⁴². Third, secreted and cell-surface proteins are involved in crosstalk with the tumour microenvironment and are differentially expressed in tumour cells^43,44. Fourth, fluid-based protein biomarkers can also capture the spatial heterogeneity of the prostate compared with proteins obtained from tissue biopsies, which are limited by spatial sampling. Finally, prostate cancer typically has a ‘quiet’ genome, with low rates of most types of mutations⁴⁵, but a high level of subclonal diversity^38,46, which considerably limits the utility of single-gene mutational biomarkers and means that biomarker panels are needed to fully and accurately capture this heterogeneity⁴⁷. Secreted proteins are the cumulative result of all levels of information and regulation through the central dogma and, therefore, have the potential to capture dysregulation beyond somatic mutations. Multiple protein biomarkers can also be developed into multiplexed protein panels or combined with other molecular biomarkers from the same fluid to fully capture disease heterogeneity.

Despite the promises offered by fluid-based protein biomarkers, few studies have succeeded in identifying candidates with biomarker potential. A big challenge impeding prostate cancer candidate biomarker discovery studies from yielding meaningful outcomes is the limited information provided by the small discovery cohorts, which hinders the assessment of measurement variability and limits potential candidates to those with large effect sizes. In addition, few studies share the same biomarker candidates owing to a lack of standardized methods for sample collection, sample preparation, data acquisition and bioinformatics analysis. Many studies identify statistically significant biomarker candidates, typically by testing whether an area under the receiver operating characteristic curve (AUC) is significantly different from 0.5. However, these study premises inherently lead to the discovery of molecules that could never be used as biomarkers as they are either clinically irrelevant or aimed at being proxies for existing modalities. For example, urine biomarkers that can distinguish men with prostate cancer from those without are not clinically needed, as all men suspected to have prostate cancer would either be symptomatic or have abnormal PSA and/or DRE. Similarly, a biomarker that performs marginally better than PSA (albeit significant) for the detection of clinically significant disease has no clinical utility, as clinical paradigms could not be changed overnight and whether the biomarker could reliably compliment PSA in distinguishing clinically insignificant and significant disease is a more important question. In addition, studies that do not develop a standardizable validation assay for candidates are difficult to interpret, as how the initial discoveries are dependent on the protocol used is unclear. Finally, the lack of independent validation cohorts reduces confidence in any biomarker candidate, and validation cohorts that are underpowered and do not have relevant clinical annotations available also mean that, ultimately, assessing whether candidates add value to existing modalities is impossible. Furthermore, several key questions remain unanswered in biomarker development and validation for prostate cancer. To what degree current molecular and biomarker studies underestimate the true molecular heterogeneity of prostate as a result of insufficient or biased spatial sampling is unclear^38,48,49. Additionally, how the short-term and long-term temporal dynamics of biomarkers affect their validity and clinical utility is uncertain^22,50. The influence of germline polymorphisms and race on biomarkers also remains underinvestigated^51,52. Easily accessible fluids in combination with standardized targeted proteomics assays have great promise for analysing the large cohorts required to answer these questions and identify biomarkers that overcome these challenges.

Proteomic technologies

MS can be used at various stages of the biomarker discovery process to identify and validate protein biomarker candidates (FIG. 2). Commonly used MS-based proteomics technologies include shotgun proteomics, native peptidomics, imaging mass cytometry (IMC) for biomarker discovery and targeted MS for biomarker validation.

Shotgun proteomics.

An untargeted workflow for identifying and quantifying proteins by mass spectrometry via proteolytic digestion of proteins into peptides.

Dynamic range.

In mass spectrometry, this term is the range of protein abundances in a sample.

Precursor ion.

Intact ions that later dissociate into smaller fragment ions.

Fragment ion.

An ion that is the product of fragmentation. For peptides, fragment ions are produced from fragmentation at the peptide backbone.

Data-dependent acquisition.

(DDA). A mass-spectrometry acquisition method in which the top N most intense peptides are selected for fragmentation.

Data-independent acquisition.

(DIA). A mass-spectrometry acquisition method in which all peptides within a given mass window (such as 15–50 m/z) are selected for fragmentation. Peptides in a selected mass range are fragmented using sequential windows.

Fig. 2 |. Biomarker discovery to translation into the clinic.

Mass spectrometry-based assays can be used at several points in biomarker discovery and translation. Once a clinical need has been determined and clinical samples are collected or preclinical models are established, samples can be characterized by shotgun proteomics or imaging mass cytometry (IMC) at the biomarker discovery stage to identify protein biomarkers of interest. Following proteolytic digestion of proteins from the sample of interest, peptides are separated by liquid chromatography, then ionized through electrospray ionization (ESI), and analysed using mass spectrometry⁵³. Data-dependent acquisition (DDA) or data-independent acquisition (DIA) scan modes can be used for analysis. In DDA-MS, the top N most abundant peptide precursor ions in each scan are sequentially selected and isolated within a narrow mass window (typically 1–2 m/z) for fragmentation. The resulting MS2 spectrum is matched to in silico spectra via database searching for peptide identification. In DIA-MS, all peptide precursor ions in a predefined mass window (typically 10–50 m/z) are isolated for fragmentation, yielding complex MS2 spectra. MS2 spectra are matched to peptides via deconvolution using a spectral library as a reference. IMC can also be used in the discovery stage to characterize tumour spatial heterogeneity at subcellular resolution to distinguish between cell populations of interest. Cells are labelled with metal-conjugated antibodies and laser ablation coupled to inductively coupled plasma (ICP)-MS-based detection is used to quantify reporter ion intensity in each region of the tissue. Statistical analysis is used to select candidates for validation by targeted MS, and for building multipeptide biomarker panels from targeted MS results. A robust statistical analysis workflow requires feature selection, model selection, as well as model hyper-parameter tuning and training to be performed solely on the discovery cohort, after which the model is frozen (that is, the parameters and hyperparameters can no longer be adjusted). The frozen model can then be evaluated in unseen validation cohorts to gauge its performance and generalizability. Targeted MS can be used to validate biomarker candidates in a larger cohort. Sample preparation for targeted proteomics is similar to shotgun proteomics, with the addition of stable isotope-labelled ‘heavy’ peptides synthesized for peptides that best represent the protein candidate of interest. These peptides are placed on an inclusion list, and only peptides on this list are fragmented and analysed, resulting in a lower limit of detection and higher reproducibility in quantification than shotgun proteomics. Parallel reaction monitoring (PRM) can be used to quantify all the fragment ions generated from each peptide of interest, whereas selected reaction monitoring or multiple reaction monitoring (SRM/MRM) can be used to quantify select fragment ions (typically 3) from each peptide. Once these assays are developed and standardized, they can be used as-is in future clinical trials and in the clinic.

Shotgun proteomics.

In the biomarker discovery stage, shotgun proteomics⁵³, also called bottom-up proteomics, is often used to detect and quantify proteins from biological samples. In shotgun proteomics, a mixture of proteins are digested using proteases to obtain peptides before analysis. Trypsin and Lys-C are most commonly used for digestion because they yield peptides with positively charged C-termini that are amenable to ionization and are, therefore, detectable by MS⁵⁴. Optionally, sample complexity can be further reduced by fractionating samples using off-line orthogonal methods, such as high pH-reversed phase or strong cation exchange chromatography^55,56. Off-line fractionation increases proteome coverage (number of detected proteins)^56,57, but it requires increased instrument time and input material (>100 μg of protein), reducing feasibility with many clinical samples. Thus, this technique is not often performed in biomarker discovery studies using clinical samples. Then, in order to increase the dynamic range of observed peptide concentrations and increase detection of low-abundance ions, peptides are separated using nanolitre flow-rate liquid chromatography (LC) columns coupled directly (termed ‘in-line’) to the mass spectrometer (called LC–MS). C18 resins that separate peptides based on hydrophobicity are most commonly used for separation, although other modes of separation, such as capillary electrophoresis (CE), which separate peptides based on their electrophoretic mobility, can also be used⁵⁸. As peptides elute from the chromatography column, they are ionized by electrospray ionization, through protonation of tryptic peptides by mild acids in the mobile phases. In addition to the separation methods described above, ions can be further separated by ion mobility before mass acquisition^59,60, in which peptide ions are separated by their mass-to-charge ratio (m/z) in mass analysers⁵³.

The mass spectrometer records two independent but related spectra: the MS1 spectrum, which records the intensity and m/z of intact peptide ions (termed precursor ion), and the fragment ion mass spectrum (termed MS2 or MS/MS spectrum), which is generated through fragmentation of peptides present in the MS1 spectrum. These peptides are fragmented along the peptide backbone through collision with inert gas molecules such as helium, nitrogen or argon⁶¹, producing smaller fragment ions. The MS2 spectrum records the intensity and m/z of fragmented peptide ions. To determine the identity of a peptide, both the MS1 and matching MS2 spectra are required. The most common strategy is database searching, in which experimental fragment ion spectra are matched to theoretical spectra generated by in silico digestion of a reference proteome⁵³. To control for the error rates associated with spectral matching in complex proteomes, experimental fragment ion spectra are also matched against ‘dummy sequences’ generated through inversion or scrambling of known protein sequences (such as the target-decoy approach)^62,63. Matched spectra are carried forward for reporting and protein inference (that is, the process of matching detected peptide sequences to available protein accession, which is similar to aligning sequence reads to genomic coordinates). Relative peptide abundance can be quantified from the MS1 spectrum, usually by integrating the area under the curve to the ion abundance over elution (the retention time) of the peptide. This approach is commonly referred to as label-free quantification (LFQ), as no stable isotope standards or chemical labelling strategies are employed⁶⁴. Peptide quantity can also be derived by label-based quantification using isobaric labels, such as tandem mass tags (TMT) and others (for example iTRAQ or EASI-tag). Isobaric labelling enables the multiplexed analyses of up to 16 samples with potentially increased quantitative precision compared with LFQ^65,66, but it has decreased quantitative accuracy and peptide identification rates^67,68.

Two independent MS scan modes are commonly used in shotgun proteomics — data-dependent acquisition (DDA) and data-independent acquisition (DIA)^69,70. In DDA, the top N most abundant ions from each MS1 scan are selected for fragmentation, in which N is selected based on the instrument’s scan speed and sample complexity. Each cycle consists of one MS1 scan, followed by N fragment ion (MS2) scans. The main drawback of DDA is that selecting peptides for fragmentation is semi-stochastic; thus, lowabundance peptides might not be reproducibly detected and quantified in biological replicates. Nevertheless, protein quantification by DDA is robust⁴⁷. In DIA, all peptides in successive 10–50-m/z windows are selected for fragmentation¹⁹. Owing to the higher rates of co-fragmentation than DDA, which results in mixed spectra, spectral matching to theoretical spectra is not commonly used for peptide inference. Instead, experimental data from DIA are matched to spectral libraries that consist of previously recorded MS2 spectra, usually generated from pooled or highly fractionated samples previously acquired in DDA mode. The spectral library is used to deconvolve the mixed fragment ion spectra from multiple peptides⁷¹. Although peptide detection and quantification in each DIA run is limited to peptides present in the spectral library, data acquisition is continuous and, therefore, low-abundance peptides can be quantified more reproducibly than in DDA⁷¹.

Performing shotgun proteomics on specific subsets of the proteome, such as post-translationally modified (PTM) proteins, or extracellular vesicle (EV) proteomes, can complement biomarker discovery by highlighting proteins involved in specific tumour pathways. For example, enriching the sample for EVs by ultracentrifugation or filtration can increase detection of EV proteins, which include low-abundance proteins such as signalling molecules^72,73. Enriching for PTM proteins can also improve the detection of specific subsets of the proteome. The two most common PTM-enrichment workflows are phosphoproteomics and N-glycoproteomics. Enriching for phosphoproteins from cells and tissues using immobilized metal affinity chromatography or titanium dioxide magnetic beads can help to identify biomarkers associated with aberrant signalling pathways⁷⁴. For example, phosphoproteomics on prostate cancer tissues from patients with metastatic castration-resistant prostate cancer (mCRPC) revealed actionable therapeutic pathways such as cell-cycle regulation, nuclear receptor, PI3K–AKT–mTOR and sternness pathways, as well as associated kinases that are targetable with currently available clinical inhibitors⁷⁵. N-glycoproteomics can enrich for cell surface proteins, as the majority of proteins at the cell membrane are N-glycosylated⁷⁶. Applying these subset-specific enrichment workflows to biomarker discovery is currently limited by the large amount of input material required for enrichment, as PTM-modified peptides are relatively low in abundance compared with their unmodified forms^77,78. Historically, phosphoproteomics and N-glycoproteomics workflows required a minimum of 500–1,000 μg of protein as input material^79,80, which might be challenging to obtain from clinical samples. Developments in sample preparation and improvements in MS technology for low-input proteomics have reduced the required input material by up to 20-fold while maintaining a similar number of detected peptides, reproducibility and signal^80,81. These methods use magnetic bead capture, robotic automation or spin tips to capture peptides within a small volume, therefore, reducing non-specific sample losses while enabling the use of harsh detergents for cell lysis^80–83. These low-input methods are promising, but are not yet widely used in clinical proteomics owing to their novelty, and have not been applied to prostate cancer-associated fluids.

Native peptidomics.

The native peptidome, consisting of low-molecular-weight proteins and cleaved peptides, can also be analysed using LC-MS/MS or capillary electrophoresis MS (CE-MS), an analogous technique to LC-MS⁸⁴. In native peptidomics, peptides are isolated from fluids by filtration through a low-molecular-weight cut-off filter, then separated by charge and size before MS.

Selected reaction monitoring/multiple reaction monitoring.

(SRM/MRM). A targeted mass spectrometry method that sequentially isolates and records pre-selected fragment ion masses from a peptide.

Triple-quadrupole mass spectrometer.

A tandem mass spectrometer consisting of two quadrupole mass analysers arranged sequentially for mass isolation with an additional quadrupole in the middle that is used for collision-induced dissociation.

Parallel reaction monitoring.

(PRM). A targeted mass spectrometry method that isolates and records all fragment ion masses from a peptide.

Orbitrap mass analyser.

An ion-trap mass analyser that detects m/z signals by oscillating ions around a cylindrical electrode with tapered ends.

Imaging mass cytometry.

An emerging technology for mapping the spatial distribution of proteins on a tissue section at single-cell resolution is IMC⁸⁵. In this technique, sectioned tissues are stained with antibodies complexed to heavy metal reporter ions, and laser ablation coupled with MS-based detection is used to quantify reporter ion intensity in each region of the tissue⁸⁵. In addition, images generated from IMC can be co-registered onto gross pathology, such as haematoxylin and eosin-stained tissue or other small molecule stains (for example, pimonidazole staining for hypoxia). IMC has not yet been applied to prostate tissues, as the technology is very novel, having only been applied to breast cancer tissue in 2020 (REF⁸⁵). In breast cancer, this technique has been shown to be able to distinguish single tumour cells and stromal cells, and characterize tissue organization and heterogeneity⁸⁵. IMC has the potential to enable the assessment of proteins specific to adverse prostate cancer histologies (such as intraductal carcinoma and Gleason pattern 4 and 5) while retaining spatial information and provide potential biomarker candidates that could be quantified in prostate-associated fluids.

Targeted proteomics.

Protein biomarker candidates in fluid can be validated using western blot and enzyme-linked immunosorbent assays (ELISAs)⁸⁶. Western blot has low throughput, as each experiment is limited by the number of wells in a gel; thus, ELISA is most commonly used for protein biomarker validation owing to its higher throughput than western blot. Nevertheless, several caveats to using antibody-based assays for validation of protein biomarkers exist. First, they are limited to proteins for which high-quality antibodies are available^87,88. In addition, ELISA development is expensive (costing $100,000–$1,000,000 per assay) and inefficient (1–2 years for development)⁸⁹. Thus, only a small number (in the order of tens) of high-priority candidates can move on to the validation stage, limiting the number of potential candidate biomarkers that can be validated.

To overcome the limitations of using antibody-based validation of candidate biomarkers, targeted MS can be used in a high-throughput and cost-effective manner. Hundreds of candidates can be multiplexed and evaluated in a single analysis⁹⁰ at a cost of <$5,000 and ~6 months for assay development^89,91–93. Targeted MS assays, such as selected reaction monitoring/multiple reaction monitoring (SRM/MRM) for triple-quadrupole mass spectrometers, and parallel reaction monitoring (PRM) for orbitrap mass analysers, quantify fragment ions of selected proteotypic peptides from each candidate protein. A proteotypic peptide is a peptide that is unique to a protein or protein isoform and produces a good MS response⁹⁴. Proteotypic peptides that best represent proteins of interest from the discovery stage are selected for measurement. To ensure that quantification is reproducible across different laboratories, stable isotope-labelled peptides at known concentrations are synthesized for every candidate peptide and used as spike-in standards. When an assay has been validated on an instrument, it can be transferred across sites and clinical laboratories⁹⁴.

In summary, shotgun proteomics using LC-MS/MS is most suited to protein biomarker discovery in medium-complexity to high-complexity samples such as tissues, cell lysates and patient fluids. It provides a broad overview of protein abundance in a biological sample for comparisons between different clinical groups. As no standard of known quantity is spiked into each sample, quantification is relative⁵³. Of the acquisition methods used for shotgun proteomics, DDA is the best established⁵³, and DDA acquisition can be multiplexed by labelling samples with isobaric labels^65,66. DIA is gaining in popularity as it enables continuous data acquisition for improved scalability and robust quantification¹⁹. However, like DDA, DIA is most suited to the biomarker discovery stage, as quantification is relative¹⁹. Similar to LC-MS/MS, CE-MS is also most suited to the biomarker discovery stage, such as for native peptidomics, especially for studying PTM-modified peptides like phosphopeptide positional isomers that cannot be separated by LC-MS⁹⁵, as well as for profiling samples with low protein input⁹⁶. However, as CE-MS instruments are not as readily available as LC-MS instruments, its potential for identifying peptides in fluids has not been fully explored.

When candidate peptides have been selected from the discovery stage, targeted MS proteomics assays (such as SRM or PRM) are most suited to validating targets, as they have lower limits of detection and better quantitative accuracy than shotgun proteomics methods when used in combination with stable isotope-labelled synthetic peptides at known concentrations⁹⁷. For clinical implementation, once a clinical laboratory has a LC-MS/MS system with SRM or PRM functionality, targeted proteomics assays can be easily implemented and transferred.

For both shotgun and targeted proteomics, spatial molecular information is lost. IMC can complement shotgun proteomics analyses and provide information about the spatial organization of proteins and protein post-translational modifications within the tissue, as well as reveal protein heterogeneity within tumour regions. This tool is best used to study the tumour microenvironment and the immune landscape of the tumour⁸⁵. Although these complementary MS technologies provide a basis for improved protein biomarker discovery and validation, careful sample selection is still required to improve protein detection.

Sample selection for MS

Technological advances in MS instrumentation and sample preparation mean that 2,100 proteins can now be detected in 100 μl of urine⁹⁸ and 450 proteins in 10 μl of platelet-free plasma in a single MS run⁹⁹. This ability presents an opportunity to discover and validate protein biomarkers in various sample types (FIG. 3) and patient cohorts, enabling improved analysis of intertumoural and intratumoural heterogeneity⁴⁷. One challenge of shotgun proteomics is that the dynamic range of peptide detection in a single scan is limited to approximately four orders of magnitude¹⁰⁰. For example, for matrices that contain proteins at vastly different concentrations, a deep proteome might be difficult to obtain, as high-abundance proteins are sampled more frequently than low-abundance proteins¹⁰¹. High-abundance ions that co-elute with low-abundance ions can also reduce the ionization efficiency (that is, reduce the signal:noise ratio) of low-abundance ions, termed ion suppression¹⁰². Because proteins that are leaked or secreted by tissue and/or are involved in signalling are present at low concentrations in fluids in general¹⁰¹, careful sample selection is required to help to increase the detection of prostate cancer-derived proteins (TABLE 1).

Fig. 3 |. Clinically relevant fluids in prostate cancer.

a | Prostatic fluid can be obtained through vigorous prostate massage, termed expressed prostatic secretions (EPS). EPS can be collected after digital rectal examination (DRE) in voided first-catch urine, termed post-DRE urine. These prostatic fluids are the most interesting for biomarker discovery as they contain cells from the prostate as well as secreted proteins and extracellular vesicle (EV) proteins released from both nonmalignant and tumour cells. As the prostate is well vascularized, secreted proteins can also be found in blood. b | EVs are formed either by direct budding from the cell membrane or by fusion of intraluminal vesicles containing exosomes with the cell membrane. EVs also have a lipid bilayer that protects proteins from degradation by extracellular proteases. Secreted proteins can also be detected in fluids.

Table 1 |.

Clinically relevant fluid biospecimens for biomarker discovery

Sample	Advantages	Disadvantages	Ideal use
Blood	Non-invasive collection; contains proteins secreted and shed from prostate epithelial cells, and lysed cells	High dynamic range, detection of low-abundance proteins is challenging; prostate-derived proteins are diluted	Detection; longitudinal monitoring (AS, treatment response)
EPS	Highly enriched in prostate-derived proteins	Collection is challenging; not routinely collected in the clinic	Biomarker discovery
Post-DRE urine	Enriched in prostate-derived molecules; non-invasive collection; increased peptide stability compared with blood	Contains non-prostate-derived proteins (from the kidneys and the bladder); protein concentration varies between individuals	Detection; longitudinal monitoring (AS, treatment response)
Extracellular vesicles	Found in serum, post-DRE urine, prostatic secretions; deep coverage of the prostate proteome; enables multi-omics analysis	Isolation is challenging, no gold-standard method exists, tedious, and requires large volumes of fluid	Biomarker discovery

AS, active surveillance; DRE, digital rectal examination; EPS, expressed prostatic secretions.

Blood.

Owing to its ease of collection, blood is applicable to every stage of prostate cancer clinical management and is ideal for longitudinal monitoring. Analysis of blood is able to provide information about systemic disease; thus, this fluid is especially suitable for relapse monitoring after definitive treatment. Blood is best suited to validating potential biomarkers using targeted MS assays that have a low limit of detection, but is less suited to MS-based biomarker discovery as the dynamic range of blood proteins spans 10 orders of magnitude (from 5 pg/ml to 50 mg/ml)¹⁰¹. In addition, tumour-derived proteins are present at lower concentrations in the circulation than in fluids proximal to the prostate¹⁰³ and are, therefore, difficult to detect using shotgun proteomics. Immunodepletion can be used to remove high-abundance proteins before MS analysis to reduce the range of protein concentrations¹⁰⁴. However, some caveats of immunodepletion include potential non-specific binding of non-target and carrier proteins, varying efficiencies of depletion for high-abundance proteins (ranging from 30% to 99%) and reduced throughput of sample preparation¹⁰⁴.

Prostatic fluid.

Prostatic fluid is an attractive source for biomarker discovery as most of its proteome is prostate-derived and largely devoid of proteins from the seminal vesicles and testes¹⁰⁵. Prostatic fluid is protein rich, containing proteins that are secreted or shed by epithelial cells and released into the lumen of the prostatic ducts, which empty into the urethra¹⁰⁵. The abundance of prostate-enriched proteins in prostatic fluid differs between disease states, for example, PSA, PAP and TGM4 are differentially expressed in the prostatic fluid of patients with organ-confined disease compared with those with extracapsular disease¹⁰⁶. Prostatic fluid can be collected by ejaculation as a component of seminal plasma¹⁰⁷ or expressed directly from the prostate following prostate massage and collected through the urethra, termed expressed prostatic secretions (EPS)¹⁰⁸. EPS is not routinely collected in the clinic owing to patient discomfort¹⁰⁹, but can be collected before radical prostatectomy while the patient is under anaesthesia¹⁰⁸. This setting for sample collection is ideal as it causes no additional discomfort to the patient, and presurgical and postsurgical clinical information (such as ISUP GG) are available to guide biomarker discovery for future validation in fluids that are more accessible than EPS.

Urine.

Urine can be collected routinely in the clinic and used in both diagnostic and prognostic settings. This fluid is ideal for longitudinal monitoring during AS, and can be applied to longitudinal post-treatment monitoring of patients who are undergoing radiotherapy. Urine-derived peptides are more stable than those from blood, as most protein degradation by proteases would have been completed in the bladder before collection¹⁶. The primary limitation of using urine for biomarker discovery as well as clinical assays is the variability of the urine proteome, which can be affected by hydration status¹¹⁰. Various different methods for normalizing protein abundance in urine have been suggested to account for hydration status, time of collection, and other conditions that can affect protein abundance, such as proteinuria. These methods include normalizing by fluid volume, total protein concentration, peptide concentration, urine osmolarity and urinary creatinine levels¹¹¹. However, no consensus exists on the best normalization method. By contrast, the proteome of post-DRE urine, defined as first-catch urine following prostatic massage via DRE, is more stable than neat urine (urine collected without a DRE) as it contains prostate-derived proteins from EPS along with the kidney-derived and bladder-derived proteins generally found in urine^112,113. This increased stability is evident in the increased abundance of prostate-derived proteins and RNAs (such as TGM4, KLK3 and HOXB13) in post-DRE urine compared with neat urine^112,113 and, therefore, post-DRE urine can be considered a more accessible form of EPS than neat EPS. As post-DRE urine contains many prostate-derived proteins and can be easily collected in the clinic, it is an ideal fluid for biomarker discovery and for the validation of candidates previously discovered in EPS or tissue¹¹⁴. Post-DRE urine is already in clinical use for the detection of the FDA-approved biomarker PCA3, a long non-coding RNA¹¹⁵.

Extracellular vesicles.

EVs are membrane-bound particles containing proteins, nucleic acids and lipids, which are shed from most cell types via various mechanisms¹¹⁶, and can be readily found in bodily fluids such as blood and urine¹¹⁶. EVs are released into the extracellular space either by direct budding from the cell membrane or by fusion of intraluminal vesicles containing exosomes with the cell membrane¹¹⁶. EVs also have a lipid bilayer that protects proteins from degradation by extracellular proteases¹¹⁶. In cancer, host cells communicate with neighbouring cells by releasing EVs to promote inflammation, cell migration and proliferation¹¹⁷. EVs are well-suited to biomarker discovery as they can be readily obtained from fluids; thus, enriching for EVs by ultracentrifugation, filtration or size-exclusion chromatography in prostatic secretions and urine might improve the detection of secreted prostate-derived proteins by reducing the amounts of high abundance kidney-derived or blood-derived soluble proteins present in the fluid^72,113. The most abundant proteins in the urine-derived EV proteome of patients with prostate cancer and benign prostatic conditions consist of prostate-enriched proteins (for example, PAP, PSA, PSMA and TGM4) that are derived from the prostate and EV-enriched proteins (for example, CD9, CD63 and CD81) that are common in EVs in general^72,118. Kidney-derived proteins (such as uromodulin) and blood-derived proteins (such as albumin and apolipoproteins) remain highly abundant in urine EVs, but their concentration is reduced compared with urine itself^72,118. A considerable challenge in using EVs in the clinic is the lack of a standardized protocol for EV isolation. Ultracentrifugation is the most commonly used method for isolating EVs, but the use of different centrifugation speeds, rotors, centrifugation time and chemical treatment can make comparison of results challenging¹¹⁹. The feasibility of isolating EVs via ultracentrifugation from large clinical cohorts is an additional challenge of EV-based biomarker discovery.

The unique properties and caveats of each sample type should be carefully considered for each stage of biomarker discovery and validation. Blood and urine are most suited for longitudinal monitoring owing to their ease of collection, but are not the most ideal for biomarker discovery because of their high dynamic range that reduces the depth of protein coverage¹⁰¹. Conversely, prostatic fluid and EVs are highly enriched in prostate-derived proteins and are, therefore, most suited to biomarker discovery in which a deep proteome is needed, but less suited to validation owing to the challenges in sample collection and processing¹⁰⁹.

Non-invasive protein biomarker candidates

Novel biomarker candidates are emerging from studies that have employed proteomics techniques on prostate cancer tissue and relevant fluids. Over the past 10 years, studies that have overcome challenges in non-invasive protein biomarker discovery have been able to identify and validate biomarker candidates (TABLE 2). These biomarkers readily lend themselves to targeted assay validation and have the potential to be translated into the clinic to improve prostate cancer diagnosis, prognosis and prediction algorithms.

Table 2 |.

Promising prostate cancer biomarkers identified using mass spectrometry

Biospecimen	Biomarker	Assay	Validation	Status	Ref.
Diagnosis and decision to biopsy
Post-DRE urine EVs	TGM4	TMA, IHC	Prostate cancer (n = 136) versus tumour-adjacent nonmalignant tissue (n = 98): AUC = 0.81, 95% CI = 0.74–0.88, P < 0.001	Needs independent validation in post-DRE urine	122
Urine	β2M, PGA3, MUC3 protein abundance, tPSA	Western blot	Prostate cancer (n = 90) versus BPH (n = 83): AUC = 0.812, 95% CI = 0.740–0.885, P < 0.001	Needs further validation	123
Serum	Proclarix (Proteomedix): CTSD, THSB, %fPSA, tPSA, age	ELISA	Patients within the PSA ‘grey zone’ (tPSA 2–10 ng/ml, normal DRE, prostate volume ≥35 ml by TRUS); ISUP GG 1 versus ISUP GG ≥2: sensitivity 90%, specificity 43% (95% CI = 39–46%, P < 0.001, nno PCa = 546, nISUP GG 1 = 239, nISUP GG ≥2 = 170)	CE-marked in vitro diagnostic (Europe)	127,128,181–183
Post-DRE urine	Peptides from the proteins IDHC, SERA, IGJ, EF2 and KCRB	SRM-MS	BPH (n = 48) versus prostate cancer (n = 90): AUC = 0.77, 95% CI = 0.68–0.87, P < 0.05	Needs further validation in larger cohort	114
Prognosis and active surveillance monitoring
Post-DRE urine EVs	FABP5	SRM-MS	ISUP GG 1 (n = 5) versus ISUP GG ≥2 (n = 13): AUC = 0.86, 95% CI = 0.71–1.00, P = 0.002	Outperforms tPSA but needs validation in larger post-DRE urine cohort	130
Post-DRE urine	Peptides from the proteins 6PGL, SERA, GELS, PEDF, PARK7, 1433 S and RINI	SRM-MS	Extraprostatic (n = 29) versus organ-confined disease (n = 61): AUC = 0.74, 95% CI = 0.62–0.85, P < 0.05	Needs further validation in larger post-DRE cohort with clinical follow-up data	114
Post-DRE urine EVs	TGM4	TMA, IHC	ISUP GG 1–2 (n = 50) versus ISUP GG ≥ 3 (n = 86) disease: AUC = 0.82, 95% CI = 0.69–0.91, P < 0.001 BCR (n = 64) versus no BCR (n = 72): AUC = 0.80, 95% CI = 0.69–0.91, P < 0.001	Needs further validation in post-DRE urine with clinical follow-up data	122

%fPSA, percentage free-to-total serum PSA as measured by ELISA; AUC, area under the ROC (receiver operating characteristic) curve; BCR, biochemical recurrence, BPH, benign prostatic hyperplasia; CI, confidence interval; DRE, digital rectal examination; ELISA, enzyme-linked immunosorbent assay; IHC, immunohistochemistry; ISUP GG, International Society of Urological Pathology Grade Group; MS, mass spectrometry; OR, odds ratio; SRM-MS, selected reaction monitoring mass spectrometry; TMA, tissue microarray; tPSA: total serum PSA as measured by ELISA; TRUS, transrectal ultrasonongraphy.

Diagnosis and decision to biopsy.

Protein-based diagnostic biomarkers for prostate cancer currently only include the classic prostate cancer biomarkers: immunoassay-based quantification of serum level total PSA, free PSA and [−2]pro-PSA¹²⁰. Many studies on prostate cancer biomarkers focus on identifying proteins and protein panels that can distinguish prostate cancer from benign conditions (such as BPH) in liquid biopsy assays, with the aim of sparing patients with BPH from biopsy sampling and overdiagnosis.

A targeted approach is sometimes employed in diagnostic biomarker discovery studies when analysing fluid samples to focus only on proteins that are known to have elevated abundance in prostate tissues. TGM4 is a protein that is highly enriched in prostate tissues compared with non-prostate tissues, not predicted to be actively secreted to the blood¹²¹, but can be detected in neat urine, post-DRE urine and EVs isolated from urine^112,122. In one study investigating post-DRE urine EVs using SRM-MS, TGM4 was found to be elevated in benign disease (BPH and high-grade prostatic intraepithelial neoplasia (HGPIN)) compared with prostate cancer (prostate cancer:benign ratio = 0.60, benign n = 54, prostate cancer n = 53)¹²². In the same study, the diagnostic value of TGM4 was validated in an independent tissue microarray cohort including 98 tumour-adjacent nonmalignant and 136 prostate cancer samples (from a total of 165 patients) using IHC. Consistent with what was observed in urine EVs, TGM4 was shown to have elevated abundance in adjacent nonmalignant tissues compared with prostate cancer tissues¹²². The study results further demonstrated that TGM4 abundance in prostate cancer tissue was higher in ISUP GG 1–2 (n = 50) prostate cancer than in ISUP GG 3–5 prostate cancer (n = 86) and the IHC score achieved an AUC of 0.82 (95% CI = 0.71–0.92, P < 0.001) for distinguishing between these two groups. TGM4 was also prognostic for BCR in this tissue microarray cohort as its abundance in prostate cancer tissue was higher in patients without BCR compared with those with BCR (n_BCR= 64, n_{no BCR} = 72; AUC = 0.80, 95% CI = 0.69–0.91, P < 0.001). The validation of TGM4 in tissue cannot demonstrate its potential as a fluid biomarker and the ideal validation of TGM4 could have been carried out in post-DRE urine using targeted MS, but this study design should be a model for future projects in which maximization of the applicability of potential biomarkers in the clinic is desired. By using clinical cohorts that enable the characterization of both diagnostic and prognostic abilities of biomarker candidates, studies could be steps ahead in narrowing biomarker candidates for validation, as only diagnostic biomarkers that distinguish clinically significant disease are of interest in the clinic.

The number of proteins that are highly enriched in prostate tissues compared with non-prostate tissues is limited; thus, studies commonly characterize prostate-associated fluid proteomes using shotgun proteomics. This study design enables identification of proteins that can be detected in fluids and are differentially expressed in patients with benign prostatic conditions compared with those with prostate cancer. In one study, DDA-MS was used to characterize neat urine, and three proteins (p2M, PGA3 and MUC3) were identified as potentially able to distinguish prostate cancer from BPH when used individually or in combination¹²³. The discovery cohort with proteomics characterization is of very limited size (n_BpH = 4 n_{prostate cancer} = 4) but the three proteins were validated in a larger, non-independent cohort of 173 neat urine samples (83 patients with BPH and 90 patients with prostate cancer) by western blot. The combined three-protein biomarker achieved AUC of 0.71 (95% CI = 0.63–0.79, P < 0.001), which did not outperform PSA in the same cohort (AUC = 0.73, 95% CI = 0.65–0.81, P < 0.001). However, when the three-protein biomarker was applied in combination with PSA categories (0–4, 4.1–10 and >10 ng/ml), the multimodal biomarker performed significantly better than any alone (AUC = 0.81, 95% CI = 0.74–0.89, P < 0.001). Ideally, the proteins would have been validated using a more quantitative and reproducible assay than western blot, such as ELISA or targeted MS, but the results of this study demonstrate the possibility of finding biomarkers that are complementary to PSA.

The potential of a multimodal combination biomarker discovery approach has also been demonstrated in a study comparing serum and post-DRE urine from patients with BPH, HGPIN, localized prostate cancer or metastatic prostate cancer using iTRAQ-labelled DDA-MS¹²⁴. Of the 193 patients included in the study as part of a consecutive series, 166 had blood and post-DRE urine collected, of which 10 patients from each group were randomly selected as the discovery cohort. After detection and quantification, protein biomarker candidates that were significantly differentially expressed between the prostate cancer groups and the benign disease groups were further selected based on criteria including potential oncogenic activity, poor characterization in the literature, whether the protein is secreted and antibody availability. Two serum proteins (PF4V1 and TAGLN2) and one post-DRE urine protein (CRISP3) were selected for validation using ELISA for serum samples and western blot for post-DRE urine samples from 126 patients from the rest of the cohort (38 BPH, 22 HGPIN and 66 prostate cancer). TAGLN2 did not show differences in abundance between the patient groups in the validation cohort, but serum PF4V1 (AUC = 0.80, 95% CI = 0.71–0.88, P < 0.05) and post-DRE urine CRISP3 (AUC = 0.87, 95% CI = 0.80–0.94, P < 0.05) were both validated as independent diagnostic biomarkers distinguishing BPH from prostate cancer (local and metastatic combined), as well as potential biomarkers distinguishing HGPIN from prostate cancer (AUC_PF4V1 = 0.756, ACU_CRISP3 = 0.912, confidence interval and P value not provided). Serum PF4V1 abundance was higher in patients with BPH or HGPIN than in those with prostate cancer and was not significantly different within the benign and prostate cancer groups. Urine CRISP3 showed the opposite trend, in which its abundance was higher in the prostate cancer groups than in the benign groups (with no difference within groups). The multimodal biomarker combining serum PF4V1, urine CRISP3 and PSA was significantly superior to any one alone for distinguishing BPH from prostate cancer (AUC = 0.94, 95% CI = 0.90–0.98, P< 0.05), although results on the combination of only PF4V1 and CRISP3 were not reported. CRISP3 is an interesting protein in the context of prostate cancer because high CRISP3 protein abundance in tissue is associated with advanced tumour stage, high ISUP GG, positive surgical margins, and early BCR¹²⁵. In addition, CRISP3 expression was increased in CRPC compared with benign prostate tissue, and high levels of CRISP3 is more common in prostate tumours that are ERG-positive and have PTEN loss compared with those that do not¹²⁶. However, PF4V1 has only been reported as a tumour suppressor and was not previously associated with prostate cancer¹²⁴.

Glycoproteomics.

The large-scale study of the glycoproteome, the set of glycosylated proteins, by selective enrichment of N-glycosylated or O-glycosylated peptides.

The results of these studies have shown biomarker candidates in various fluids that can distinguish prostate cancer from benign prostatic disease, profiled using targeted or shotgun proteomics and validated using western blot or ELISA. However, they fail to address how the biomarker performs in asymptomatic men with elevated PSA (4–10 ng/ml) and negative DRE, for whom the dilemma of whether to biopsy is the greatest. Quantification of glycoproteins THBS1 and CTSD using ELISA in serum combined with %fPSA has been shown in a cohort of men with PSA levels of 2–10 ng/ml to discriminate between patients with a positive biopsy and those with a negative biopsy (AUC = 0.86 95% CI = 0.82–0.91, P < 0.001)¹²⁷. The glycoproteins THBS1 and CTSD were previously identified as potential biomarker candidates for prostate cancer diagnosis using DDA-MS glycoproteomics characterization of PTEN-null mouse model tumour and serum¹²⁸. The three biomarkers, THBS1, CTSD and %fPSA, were combined into a single multimodal biomarker using multivariable logistic regression, and trained on a training set of 237 patients, 130 of whom were diagnosed with prostate cancer on biopsy. The multimodal biomarker outperformed %fPSA alone (AUC = 0.64, 95% CI = 0.57–0.71, P < 0.001) in the validation set of 237 men, 108 of whom had a biopsy that was positive for prostate cancer, and had 62% specificity for prostate cancer at 90% sensitivity compared with a specificity of 23% at 90% sensitivity for %fPSA alone. Furthermore, the three biomarkers were also used to develop multimodal biomarkers to distinguish between insignificant (ISUP GG 1) from clinically significant (ISUP GG ≥2) prostate cancer (clinical ISUP GG: AUC = 0.73, 95% CI 0.65–0.81, n_{ISUP GG 1} = 54, n_{ISUP GG ≥ 2} = 54; pathological ISUP GG: AUC = 0.83, 95% CI 0.77–0.88, n_{ISUP GG 1} = 22, n_{ISUP GG ≥ 2} = 86). The strength of this study was that the patient cohort was restricted to men with total serum PSA levels between 2 and 10 ng/ml, negative DRE, and an enlarged prostate (determined using TRUS). These results have led to the successful translation of this biomarker, named Proclarix (Proteomedix), which is now CE marked¹²⁷. A study to determine if this biomarker can identify men who can safely avoid an upfront mpMRI, or men who can avoid biopsy when mpMRI is indeterminate, is also ongoing¹²⁸.

In summary, the most promising biomarkers for prostate cancer diagnosis and decision to biopsy will emerge from studies that consider their performance in combination with PSA and actively characterize their ability to detect clinically significant disease. A multimodal biomarker discovery approach also has great potential to deliver a biomarker with the desired clinical specificity and sensitivity, but needs to be accompanied by sizable validation cohorts to prove their generalizability.

Prognosis of localized prostate cancer.

A variety of MS-based approaches have been used in biomarker discovery studies for localized prostate cancer prognosis and risk stratification. These studies typically use risk groups or clinical ISUP GG as a proxy for non-aggressive and aggressive disease, in which clinical ISUP GG 2 or more indicates clinically significant disease that is likely to be aggressive.

An example of a biomarker discovery workflow in which biospecimens were profiled using shotgun proteomics and selected candidates were validated using targeted proteomics in independent cohorts is a study in which FABP5 in post-DRE urine-derived EVs was discovered and validated¹²⁹. In this study, FABP5 was found to be elevated in patients with intermediate-grade or high-grade prostate cancer (ISUP GG ≥2) compared with those with low-grade (ISUP GG 1) disease (P < 0.001) and predicted the presence of ISUP GG ≥2 tumours in treatment-naive patients (AUC = 0.86, 95% CI = 0.71–1.00, P = 0.002), outperforming serum total PSA (AUC = 0.511, 95% CI = 0.28–0.76, P = 0.87)¹³⁰. In addition, FABP5 was enriched in urine-derived EVs from patients with prostate cancer compared with those with a negative biopsy. However, the discovery and validation cohort sample sizes are small, as only 18 samples were used for discovery and 29 for validation (11 healthy participants, 5 men with ISUP GG 1 disease, 7 ISUP men with GG 2–3 disease, and 6 men with ISUP GG 4–5 disease). Further validation in a larger independent cohort of patients similarly spanning the entire risk spectrum of prostate cancer is required. As EV-based protein biomarkers are challenging to apply routinely in the clinic owing to the low-throughput EV isolation workflows¹³¹, validation of FABP5 in post-DRE urine would be ideal, as FABP5 can also be detected in post-DRE urine¹¹².

In one study, DDA-MS was used to analyse EPS for biomarker discovery and candidate biomarkers were then validated in post-DRE urine by SRM-MS¹⁰⁶, making use of the strengths of both fluids — EPS as a rich source of prostate-derived secreted proteins, and urine as an easily accessible fluid suitable for clinical implementation¹¹². Overall, 133 proteins in EPS were identified that were differentially abundant in extraprostatic (stage pT3) and organ-confined (stage ≤pT2) disease, from which 147 peptides that are unique to and represent the proteins of interest were suitable for verification by targeted MS (SRM) in post-DRE urine¹¹⁴. The performance of these SRM assays was evaluated in two independent sets of post-DRE urine¹¹⁴. Individually, 24 diagnostic peptides could distinguish between cancer and non-cancer groups (P < 0.05) and 14 prognostic peptides could distinguish between extracapsular and organ-confined groups (P < 0.10)¹¹⁴. Peptide triaging combined with machine learning then enabled the development of multi-peptide biomarker panels for prognosis (to distinguish organ-confined from extraprostatic disease) and also diagnosis¹¹⁴. A biomarker panel, consisting of peptides from the proteins 6PGL, SERA, GELS, PEDF, PARK7, 1433 S and RINI, was able to distinguish between extraprostatic (n = 29) and organ-confined (n = 61) disease (AUC = 0.74, 95% CI = 0.62–0.85, P < 0.05), substantially outperforming serum PSA (AUC 0.66, confidence interval and P value not shown). A second multipeptide biomarker panel, consisting of peptides from the proteins IDHC, SERA, IGJ, EF2 and KCRB, could distinguish men with prostate cancer (n = 90) from healthy individuals (n = 48) (AUC = 0.77, 95% CI = 0.68–0.87, P< 0.05), also outperforming serum PSA (AUC 0.67). These results demonstrate the potential of combining computationally guided proteomics with clinically annotated patient cohorts for the discovery of non-invasive biomarkers.

MS-based proteomics can be used to detect differences in the fluid proteome of patients with localized disease. However, most studies characterize small patient cohorts at the discovery stage (n < 20), limiting biomarker discovery to large protein changes, such as those between low-risk and intermediate-risk or high-risk groups¹³⁰, or between organ-confined and extraprostatic disease¹⁰⁶. Ideally, clinical follow-up monitoring of these cohorts should be extended to observe BCR after definitive treatment or even prostate cancer-specific survival, so that biomarker candidates can be selected based on an end point instead of patient risk group at diagnosis.

Monitoring and prediction after definitive treatment.

Few studies have identified biomarkers for the clinical management of localized prostate cancer after therapy. The lack of studies is probably caused by difficulties in establishing cohorts with a sufficient number of patients. Thus, biomarker discovery studies in this area are particularly limited by sample size and a lack of validation cohorts, and a range of clinical questions using diverse patient groups are also asked.

In one study, the serum of patients with prostate cancer with or without bone metastases (all patients were androgen deprivation therapy (ADT)-naive) was profiled using DDA-MS with iTRAQ-labelled samples with the aim of finding upregulated proteins in bone-metastatic prostate cancer¹³². The discovery cohort included 30 patients with bone metastases and 30 patients without, and each group was pooled for shotgun proteomics. In an independent cohort of patients with metastatic prostate cancer, non-metastatic prostate cancer or BPH (n = 50 each), CD59 was shown to be significantly elevated in the serum of bone metastases patients using ELISA (mean_{bone metastasis} = 58 ± 17 ng/ml mean_{non-metastatic} = 29 ± 6 ng/ml mean_BPH = 16 ± 5 ng/ml P < 0.05). The biomarker potential of CD59 was not directly assessed, but a future study could validate this protein along with other candidates for the prediction of metastases after definitive treatment or early detection of bone metastasis during post-treatment monitoring. The best cohort of patients for this validation would be those with high-risk localized prostate cancer, ideally with serial blood collection after definitive treatment and long-term follow-up data.

In one study, 15 paired serum proteomes were collected from patients with hormone-sensitive prostate cancer before and 4 months after ADT initiation, and, in addition, serum proteomes from patients with early (<2 years) or late (>3 years) ADT failure (n = 10 in each arm)¹³³. From DDA-MS with iTRAQ-labelled samples, 47 proteins were identified on the basis of both differential abundance in 4-month post-ADT-initiation sera compared with ADT-failure sera and differential abundance between early and late ADT-failure sera (both at false discovery rate <0.2). Similarly, in another study, serum EVs from 36 patients with metastatic prostate cancer (untreated men n = 8, men with ADT-sensitive disease n = 8 and men with mCRPC n = 20) were specifically profiled using TMT-labelled DDA-MS to find potential novel therapeutic targets in mCRPC¹³⁴. Levels of ACTN4 were found to be elevated in serum EVs from patients with mCRPC compared with those from patients with hormone-sensitive disease (fold change = 1.4, false discovery rate <0.01). These proteins are interesting, but independent validation was not undertaken for the detected proteins in either study. In addition, whether any of the candidates could be biomarkers that predict future early versus late castration resistance before or shortly after the initiation of ADT is a crucial clinical question. Answering this question would instead require a cohort of patients treated with ADT with at least 3 years of follow-up data, with serum collected before and ideally at multiple time points during therapy to see whether there are proteins that could indicate resistance.

Biomarker discovery studies for monitoring and prediction after definitive treatment are scarce despite the great clinical need in this area. More clinical cohorts and clinically relevant models that enable biomarker discovery in this area are urgently needed, and biomarker studies should continue to monitor their prospective patient populations so that fluid proteomes can be reanalysed in the future when outcome data are available.

Challenges in biomarker translation

The crucial challenge in translating fluid-based biomarkers into clinical use is the lack of large retrospective patient cohorts with biobanked fluids and matched clinical data (such as clinical and/or pathological ISUP GG, serum PSA level and mpMRI). Most biomarker discovery and validation have occurred in small cohorts, especially for non-routine fluids such as EPS and post-DRE urine. Most patient cohorts with fluid sample collection are prospective and lack long-term follow-up data to explore biomarker candidate associations with disease relapse after local therapy in time-to-event analyses. Thus, biomarker identification strategies are limited to approximating strong predictors of outcome, such as ISUP GG and clinical risk group, rather than directly identifying independent predictors of disease metastasis and patient survival. As a result, many clinical questions remain unanswered owing to the lack of fluid samples from appropriate patient cohorts. For example, to determine which men would most benefit from needle biopsy, control samples from men that have elevated PSA and negative DRE but are confirmed to not have prostate cancer by needle biopsy or long-term follow-up monitoring are needed. Matched mpMRI data for these men are needed to account for the role that mpMRI has in diagnosis and to facilitate discovery of biomarkers that predict which men would most benefit from biopsy following a negative mpMRI result. Similarly, for predicting which men would most benefit from AS, cohorts with upgrading data are needed, especially those that employ new clinical protocols incorporating mpMRI and molecular biomarkers in addition to standard-of-care pathways.

Patient-derived xenografts.

(PDX). Tumours grown from the implantation of a patient’s tumour cells into immunodeficient or immunocompromised mice.

A substantial technical challenge for clinical translation of protein biomarkers is the time and cost required to develop targeted MS assays for multipeptide biomarkers. For targeted MS assays, stable isotope-labelled standards (synthetic peptides carrying a ‘heavy’ stable isotope) can be used to systematically identify the best responding peptides and to generate analytical figures of merit, such as linearity, lower limit of detection and lower limit of quantification¹³⁵. The sequence composition, solubility and length of the peptide, as well as the complexity of the biological sample matrix, can all affect quantification accuracy; thus, rigorous assessment of each peptide during assay development is required in order to select the best performing peptides¹³⁵. Another technical challenge involves reducing sample loss in the sample preparation stage. Using robotic liquid handlers and miniaturizing reaction vessels might improve the peptide yield from low-input samples that are typically available in the clinical setting^82,136. Optimizing chromatographic performance by using etched array columns with lower flow rates can also reduce the amount of input material needed for shotgun proteomics analysis by 100-fold¹³⁷.

Beyond technical challenges in protein biomarker discovery and validation, the regulatory challenges in biomarker translation are substantial. Assay development and trial design need to meet the requirements for regulatory approval processes^120,138,139. After regulatory clearance, the clinical uptake of biomarkers is also not guaranteed, especially in prostate cancer for which multiple commercially available molecular biomarkers target the same patient population and report outcomes in overlapping end points. Differing recommendations between authoritative bodies and the lack of a single go-to option suggest that comparative studies are needed to determine the optimal application of molecular assays and how they should be combined with mpMRI and mpMRI-targeted biopsies in diagnosis and monitoring settings^14,27. Test availability, insurance coverage, guidelines for clinicians and education of the public are all factors that require additional investigation to guide strategies to improve biomarker uptake²⁷.

Future directions

Currently, immunoassays, usually in the form of immunohistochemistry panels, are the prevailing technology for protein biomarker assessment in the clinic given their low cost (approximately US $15–20 per marker)¹⁴⁰ and easy implementation in clinical laboratories. When considering biomarker candidates, they need to be amenable to translation into an assay format that enables ease of clinical implementation. In the future, targeted MS could be used as a platform for validating biomarker panels of up to 300 candidates in one analysis and transferred across laboratories. As contemporary pathology reports now contain prognostic and predictive factors in addition to diagnostic information, incorporation of novel proteomic biomarkers would provide complementary tools for patient management algorithms.

When patient fluids are not readily available, patient tissue and preclinical model systems such as prostate cancer cell lines, patient-derived xenografts (PDXs), organoids and transgenic or genetically engineered mouse models are an important resource for the identification and validation of new biomarker candidates^141–143. Tissue is the biospecimen that is most frequently used for biomarker discovery, disease classification and disease staging as it closely reflects tumour biology and structural changes can be readily visualized using histopathology. Tissue proteomic profiles can be linked to histopathological and imaging features, and other -omics data, such as genomic, transcriptomic, epigenomic and metabolomic information, to improve study of tumour heterogeneity and prioritize biologically relevant pathways or proteins^{38,47,144,145}. Proteins detected in prostate tissues are also more likely to be related to prostate biology than those only detected in prostate-associated fluids. Computational tools can help to select proteins that are secreted, contained in EVs or localized to cell membranes to prioritize candidates for validation in fluids using a variety of different algorithmic approaches^{121,146–148}.

Established prostate cancer cell lines have the advantages of being well characterized, and their ease of growth has made them especially useful tools for studying therapy resistance when the availability of patient-derived tissue is limited^149–151. Use of prostate cancer cell lines has facilitated proteomic studies to investigate the biology of prostate cancer therapy resistance and metastatic potential^152,153. However, these prostate cancer cell lines are limited in number and they do not fully represent the clinical spectrum of disease as most were derived frommetastatic tumours. For example, DU145, PC3 and 22Rv1 were developed from castration-resistant metastases, whereas LNCaP cells and VCaP cells were developed from hormone-naive metastases¹⁵⁴. Additionally, extensive passaging causes genetic drift, which can result in phenotypic differences between cells used across experiments and laboratories¹⁵⁵ and their growth in 2D culture means that their representation of the original tumours is limited. These limitations have been partly addressed by the derivation of primary PDXs in immunodeficient mice, which account for non-immune microenvironmental effects and maintain the genomic and histological features of the original tumour. However, their low engraftment rate, slow growth rate and lack of consistent growth in vivo, hinders their adoption for predictive assays¹⁵⁶. In addition, PDXs lose tumour-specific genomic features and subclonal complexity during serial passaging¹⁵⁷. Organoids are in vitro, 3D cellular systems that might provide a reasonable compromise between cell lines and PDXs, as they form quickly, maintain the mutational landscape and histological features of their original cancer, and are amenable for therapy prediction assays^158,159.

The development of prostate cancer mouse models enables the study of cancer progression throughout all stages from the precancerous state to localized cancer and ultimately to metastatic disease. The transgenic probasin-SV40 T antigen (TRAMP) model provides a model of prostate cancer progression that includes metastatic spread to lymph nodes, bones and lungs^160–162. The probasin PTEN-null mouse models represent prostate cancer progression from localized to metastatic disease; however, the tumours display androgen-independent growth, in contrast to the TRAMP model^128,163,164. Use of label-free LC-MS shotgun proteomics for the characterization of TRAMP model prostate glands enabled the identification of increased abundance of PDGFβ in TRAMP-derived prostate tumours relative to wild type littermates¹⁶⁵. Furthermore, proteomic (via iTRAQ-labelled DDA-MS) and transcriptomic (via microarray) profiling of the PTEN-null model demonstrated upregulation of immunity and inflammatory pathways in PTEN-null mice prostate tumours¹⁶⁶. Thus, the use of mouse models of prostate cancer has demonstrated early promise in identifying novel protein biomarkers, although a potential limitation in these models is their inability to fully encompass the heterogeneity of human prostate cancer¹⁶⁷. Performing sufficiently powered studies using preclinical models is challenging, but these investigations can be a starting point for directing biomarker discovery towards biologically relevant pathways, and building confidence in existing biomarker candidates¹⁶⁸.

To date, effectively, almost all existing molecular cancer biomarkers use a single type of information: DNA mutations, RNA abundance, protein abundance or some other information. Because each of these types of molecular information can be dysregulated in cancer in myriad ways, it might be useful to create single biomarkers that merge these different types of information. These so-called multimodal biomarkers are wide-spread in clinical practice, in which clinical assays are often integrated with pathological and radiological data using clinical risk-stratification tools (such as National Comprehensive Cancer Network guidelines) or nomograms^13,14. Creating integrated molecular markers might have several advantages. First, they might stabilize risk predictions by reducing technical and biological variance by pooling predictive information across multiple data types. Considerable evidence suggests that pooling predictors reduces variance, even in proof-of-concept studies as simple as incorporating both continuous and discrete RNA abundance in predictive modelling, which has been shown to outperform models that use only continuous or discrete data^45,169–171. Second, they might increase the ability of the biomarker to distinguish signal from noise by pooling the strengths and alleviating the shortcomings of each individual data type (that is relying on information from one data type in contexts where it is most reliable). Many examples of this possibility exist. For example, in prostate cancer, strong evidence shows that both genomic instability^172,173 and tumour evolutionary complexity⁴⁶ are accurate biomarkers of disease aggression, and both are more accurately detected from the genome than from the transcriptome or proteome^174,175; in contrast, the tumour microenvironment does change the prostate cancer genome¹⁷⁶, but inferring the state of the tumour microenvironment through the transcriptome or proteome is easier¹⁷⁷. Thus, one can imagine a multimodal biomarker measuring both the genome and gene expression (either or both of the transcriptome and proteome) to integrate information from genomic instability, evolutionary complexity and the tumour microenvironment for prostate cancer prognosis. Finally, the lack of correlation between molecular data types might in itself be a useful biomarker, as the lack of correlation could be an indication of additional dysregulation in the cancer signalling network, although to our knowledge, no studies have directly and systematically assessed the biomarker potential of these^47,178–180. Detailed studies of patient prostate-associated fluids with analysis of features such as matched germline DNA sequencing from blood and fluid EV cargo will be needed to quantify whether these benefits exist and outweigh the financial and implementation-complexity costs of multi-modal molecular biomarkers.

Conclusions

No fluid-based proteomics assay is currently available to supplement the clinical management of prostate cancer. Prostate-associated fluids are suitable for proteomic characterization and are ideal for non-invasive and low-cost targeted proteomics assays. Current challenges for biomarker discovery include limited discovery cohort sizes, suboptimal sample preparation and MS techniques, and lack of appropriate patient cohorts for validation. Existing performance metrics for different assays should not be compared given the variability in size and clinical features of test cohorts. With an increasing number of groups working on this problem, and increased sample collection and clinical annotation efforts, favourable candidates will probably emerge and be clinically validated, providing complementary biomarkers to existing clinical assessments for the diagnosis, risk-stratification and treatment monitoring of prostate cancer.

Key points.

• Standard-of-care clinical tools for the management of localized prostate cancer result in substantial overdiagnosis and overtreatment.

• Fluid-based protein biomarkers have the potential to complement clinical decision-making.

• Advances in mass spectrometry, such as increased scan speeds and mass resolution, have enabled the systematic discovery and validation of protein biomarkers in prostate-associated fluids.

• Appropriate sample selection for biomarker discovery and validation can improve detection of prostate-derived proteins in fluids.