ABSTRACT
Prostate cancer is the most common non-cutaneous cancer among men in the UK, causing significant health and economic burdens. Diagnosis and risk prognostication can be challenging due to the genetic and clinical heterogeneity of prostate cancer as well as uncertainties in our knowledge of the underlying biology and natural history of disease development. Urinary extracellular vesicles (EVs) are microscopic, lipid bilayer defined particles released by cells that carry a variety of molecular cargoes including nucleic acids, proteins and other molecules. Urine is a plentiful source of prostate-derived EVs. In this narrative review, we summarise the evidence on the function of urinary EVs and their applications in the evolving field of prostate cancer diagnostics and active surveillance. EVs are implicated in the development of all hallmarks of prostate cancer, and this knowledge has been applied to the development of multiple diagnostic tests, which are largely based on RNA and miRNA. Common gene probes included in multi-probe tests include PCA3 and ERG, and the miRNAs miR-21 and miR-141. The next decade will likely bring further improvements in the diagnostic accuracy of biomarkers as well as insights into molecular biological mechanisms of action that can be translated into opportunities in precision uro-oncology.
ABSTRACT
The development of cancer is an evolutionary process involving the sequential acquisition of genetic alterations that disrupt normal biological processes, enabling tumor cells to rapidly proliferate and eventually invade and metastasize to other tissues. We investigated the genomic evolution of prostate cancer through the application of three separate classification methods, each designed to investigate a different aspect of tumor evolution. Integrating the results revealed the existence of two distinct types of prostate cancer that arise from divergent evolutionary trajectories, designated as the Canonical and Alternative evolutionary disease types. We therefore propose the evotype model for prostate cancer evolution wherein Alternative-evotype tumors diverge from those of the Canonical-evotype through the stochastic accumulation of genetic alterations associated with disruptions to androgen receptor DNA binding. Our model unifies many previous molecular observations, providing a powerful new framework to investigate prostate cancer disease progression.
Subject(s)
Prostatic Neoplasms , Male , Humans , Prostatic Neoplasms/genetics , Prostate/metabolism , Mutation , Genomics , Evolution, MolecularABSTRACT
There is growing evidence that altered microbiota abundance of a range of specific anaerobic bacteria are associated with cancer, including Peptoniphilus spp., Porphyromonas spp., Fusobacterium spp., Fenollaria spp., Prevotella spp., Sneathia spp., Veillonella spp. and Anaerococcus spp. linked to multiple cancer types. In this review we explore these pathogenic associations. The mechanisms by which bacteria are known or predicted to interact with human cells are reviewed and we present an overview of the interlinked mechanisms and hypotheses of how multiple intracellular anaerobic bacterial pathogens may act together to cause host cell and tissue microenvironment changes associated with carcinogenesis and cancer cell invasion. These include combined effects on changes in cell signalling, DNA damage, cellular metabolism and immune evasion. Strategies for early detection and eradication of anaerobic cancer-associated bacterial pathogens that may prevent cancer progression are proposed.
Subject(s)
Bacteria, Anaerobic , Carcinogenesis , Humans , Immune Evasion , Porphyromonas , Signal Transduction , Tumor MicroenvironmentABSTRACT
IMPORTANCE: Recent reports showing that human cancers have a distinctive microbiome have led to a flurry of papers describing microbial signatures of different cancer types. Many of these reports are based on flawed data that, upon re-analysis, completely overturns the original findings. The re-analysis conducted here shows that most of the microbes originally reported as associated with cancer were not present at all in the samples. The original report of a cancer microbiome and more than a dozen follow-up studies are, therefore, likely to be invalid.
Subject(s)
Microbiota , Neoplasms , Humans , Computational Biology , Metagenomics , Data AnalysisABSTRACT
We re-analyzed the data from a recent large-scale study that reported strong correlations between microbial organisms and 33 different cancer types, and that created machine learning predictors with near-perfect accuracy at distinguishing among cancers. We found at least two fundamental flaws in the reported data and in the methods: (1) errors in the genome database and the associated computational methods led to millions of false positive findings of bacterial reads across all samples, largely because most of the sequences identified as bacteria were instead human; and (2) errors in transformation of the raw data created an artificial signature, even for microbes with no reads detected, tagging each tumor type with a distinct signal that the machine learning programs then used to create an apparently accurate classifier. Each of these problems invalidates the results, leading to the conclusion that the microbiome-based classifiers for identifying cancer presented in the study are entirely wrong. These flaws have subsequently affected more than a dozen additional published studies that used the same data and whose results are likely invalid as well.
ABSTRACT
Results published in an article by Poore et al. (Nature. 2020;579:567-574) suggested that machine learning models can almost perfectly distinguish between tumour types based on their microbial composition using machine learning models. Whilst we believe that there is the potential for microbial composition to be used in this manner, we have concerns with the paper that make us question the certainty of the conclusions drawn. We believe there are issues in the areas of the contribution of contamination, handling of batch effects, false positive classifications and limitations in the machine learning approaches used. This makes it difficult to identify whether the authors have identified true biological signal and how robust these models would be in use as clinical biomarkers. We commend Poore et al. on their approach to open data and reproducibility that has enabled this analysis. We hope that this discourse assists the future development of machine learning models and hypothesis generation in microbiome research.
Subject(s)
Microbiota , Neoplasms , Humans , Reproducibility of Results , Machine LearningABSTRACT
There is considerable interest in urine as a non-invasive liquid biopsy to detect prostate cancer (PCa). PCa-specific transcripts such as the TMPRSS2:ERG fusion gene can be found in both urine extracellular vesicles (EVs) and urine cell-sediment (Cell) but the relative usefulness of these and other genes in each fraction in PCa detection has not been fully elucidated. Urine samples from 76 men (PCa n = 40, non-cancer n = 36) were analysed by NanoString for 154 PCa-associated genes-probes, 11 tissue-specific, and six housekeeping. Comparison to qRT-PCR data for four genes (PCA3, OR51E2, FOLH1, and RPLP2) was strong (r = 0.51-0.95, Spearman p < 0.00001). Comparing EV to Cells, differential gene expression analysis found 57 gene-probes significantly more highly expressed in 100 ng of amplified cDNA products from the EV fraction, and 26 in Cells (p < 0.05; edgeR). Expression levels of prostate-specific genes (KLK2, KLK3) measured were ~20× higher in EVs, while PTPRC (white-blood Cells) was ~1000× higher in Cells. Boruta analysis identified 11 gene-probes as useful in detecting PCa: two were useful in both fractions (PCA3, HOXC6), five in EVs alone (GJB1, RPS10, TMPRSS2:ERG, ERG_Exons_4-5, HPN) and four from Cell (ERG_Exons_6-7, OR51E2, SPINK1, IMPDH2), suggesting that it is beneficial to fractionate whole urine prior to analysis. The five housekeeping genes were not significantly differentially expressed between PCa and non-cancer samples. Expression signatures from Cell, EV and combined data did not show evidence for one fraction providing superior information over the other.
ABSTRACT
BACKGROUND: Up to 80% of cases of prostate cancer present with multifocal independent tumour lesions leading to the concept of a field effect present in the normal prostate predisposing to cancer development. In the present study we applied Whole Genome DNA Sequencing (WGS) to a group of morphologically normal tissue (n = 51), including benign prostatic hyperplasia (BPH) and non-BPH samples, from men with and men without prostate cancer. We assess whether the observed genetic changes in morphologically normal tissue are linked to the development of cancer in the prostate. RESULTS: Single nucleotide variants (P = 7.0 × 10-03, Wilcoxon rank sum test) and small insertions and deletions (indels, P = 8.7 × 10-06) were significantly higher in morphologically normal samples, including BPH, from men with prostate cancer compared to those without. The presence of subclonal expansions under selective pressure, supported by a high level of mutations, were significantly associated with samples from men with prostate cancer (P = 0.035, Fisher exact test). The clonal cell fraction of normal clones was always higher than the proportion of the prostate estimated as epithelial (P = 5.94 × 10-05, paired Wilcoxon signed rank test) which, along with analysis of primary fibroblasts prepared from BPH specimens, suggests a stromal origin. Constructed phylogenies revealed lineages associated with benign tissue that were completely distinct from adjacent tumour clones, but a common lineage between BPH and non-BPH morphologically normal tissues was often observed. Compared to tumours, normal samples have significantly less single nucleotide variants (P = 3.72 × 10-09, paired Wilcoxon signed rank test), have very few rearrangements and a complete lack of copy number alterations. CONCLUSIONS: Cells within regions of morphologically normal tissue (both BPH and non-BPH) can expand under selective pressure by mechanisms that are distinct from those occurring in adjacent cancer, but that are allied to the presence of cancer. Expansions, which are probably stromal in origin, are characterised by lack of recurrent driver mutations, by almost complete absence of structural variants/copy number alterations, and mutational processes similar to malignant tissue. Our findings have implications for treatment (focal therapy) and early detection approaches.
Subject(s)
Prostatic Hyperplasia , Prostatic Neoplasms , Clone Cells/pathology , Humans , Male , Nucleotides , Prostate/pathology , Prostatic Hyperplasia/genetics , Prostatic Hyperplasia/pathology , Prostatic Neoplasms/genetics , Prostatic Neoplasms/pathologyABSTRACT
BACKGROUND: Germline variants explain more than a third of prostate cancer (PrCa) risk, but very few associations have been identified between heritable factors and clinical progression. OBJECTIVE: To find rare germline variants that predict time to biochemical recurrence (BCR) after radical treatment in men with PrCa and understand the genetic factors associated with such progression. DESIGN, SETTING, AND PARTICIPANTS: Whole-genome sequencing data from blood DNA were analysed for 850 PrCa patients with radical treatment from the Pan Prostate Cancer Group (PPCG) consortium from the UK, Canada, Germany, Australia, and France. Findings were validated using 383 patients from The Cancer Genome Atlas (TCGA) dataset. OUTCOME MEASUREMENTS AND STATISTICAL ANALYSIS: A total of 15,822 rare (MAF <1%) predicted-deleterious coding germline mutations were identified. Optimal multifactor and univariate Cox regression models were built to predict time to BCR after radical treatment, using germline variants grouped by functionally annotated gene sets. Models were tested for robustness using bootstrap resampling. RESULTS AND LIMITATIONS: Optimal Cox regression multifactor models showed that rare predicted-deleterious germline variants in "Hallmark" gene sets were consistently associated with altered time to BCR. Three gene sets had a statistically significant association with risk-elevated outcome when modelling all samples: PI3K/AKT/mTOR, Inflammatory response, and KRAS signalling (up). PI3K/AKT/mTOR and KRAS signalling (up) were also associated among patients with higher-grade cancer, as were Pancreas-beta cells, TNFA signalling via NKFB, and Hypoxia, the latter of which was validated in the independent TCGA dataset. CONCLUSIONS: We demonstrate for the first time that rare deleterious coding germline variants robustly associate with time to BCR after radical treatment, including cohort-independent validation. Our findings suggest that germline testing at diagnosis could aid clinical decisions by stratifying patients for differential clinical management. PATIENT SUMMARY: Prostate cancer patients with particular genetic mutations have a higher chance of relapsing after initial radical treatment, potentially providing opportunities to identify patients who might need additional treatments earlier.
Subject(s)
Phosphatidylinositol 3-Kinases , Prostatic Neoplasms , Germ Cells , Germ-Line Mutation , Humans , Male , Neoplasm Recurrence, Local/genetics , Phosphatidylinositol 3-Kinases/genetics , Prostatectomy , Prostatic Neoplasms/surgery , Prostatic Neoplasms/therapy , Proto-Oncogene Proteins c-akt/genetics , Proto-Oncogene Proteins p21(ras)/genetics , TOR Serine-Threonine KinasesABSTRACT
BACKGROUND: Bacteria play a suspected role in the development of several cancer types, and associations between the presence of particular bacteria and prostate cancer have been reported. OBJECTIVE: To provide improved characterisation of the prostate and urine microbiome and to investigate the prognostic potential of the bacteria present. DESIGN, SETTING, AND PARTICIPANTS: Microbiome profiles were interrogated in sample collections of patient urine (sediment microscopy: n = 318, 16S ribosomal amplicon sequencing: n = 46; and extracellular vesicle RNA-seq: n = 40) and cancer tissue (n = 204). OUTCOME MEASUREMENTS AND STATISTICAL ANALYSIS: Microbiomes were assessed using anaerobic culture, population-level 16S analysis, RNA-seq, and whole genome DNA sequencing. RESULTS AND LIMITATIONS: We demonstrate an association between the presence of bacteria in urine sediments and higher D'Amico risk prostate cancer (discovery, n = 215 patients, p < 0.001; validation, n = 103, p < 0.001, χ2 test for trend). Characterisation of the bacterial community led to the (1) identification of four novel bacteria (Porphyromonas sp. nov., Varibaculum sp. nov., Peptoniphilus sp. nov., and Fenollaria sp. nov.) that were frequently found in patient urine, and (2) definition of a patient subgroup associated with metastasis development (p = 0.015, log-rank test). The presence of five specific anaerobic genera, which includes three of the novel isolates, was associated with cancer risk group, in urine sediment (p = 0.045, log-rank test), urine extracellular vesicles (p = 0.039), and cancer tissue (p = 0.035), with a meta-analysis hazard ratio for disease progression of 2.60 (95% confidence interval: 1.39-4.85; p = 0.003; Cox regression). A limitation is that functional links to cancer development are not yet established. CONCLUSIONS: This study characterises prostate and urine microbiomes, and indicates that specific anaerobic bacteria genera have prognostic potential. PATIENT SUMMARY: In this study, we investigated the presence of bacteria in patient urine and the prostate. We identified four novel bacteria and suggest a potential prognostic utility for the microbiome in prostate cancer.
Subject(s)
Microbiota , Prostatic Neoplasms , Bacteria/genetics , Humans , Male , Microbiota/genetics , Prostate/pathology , Prostatic Neoplasms/pathology , RNA, Ribosomal, 16S/geneticsABSTRACT
There is a clinical need to improve assessment of biopsy-naïve patients for the presence of clinically significant prostate cancer (PCa). In this study, we investigated whether the robust integration of expression data from urinary extracellular vesicle RNA (EV-RNA) with urine proteomic metabolites can accurately predict PCa biopsy outcome. Urine samples collected within the Movember GAP1 Urine Biomarker study (n = 192) were analysed by both mass spectrometry-based urine-proteomics and NanoString gene-expression analysis (167 gene-probes). Cross-validated LASSO penalised regression and Random Forests identified a combination of clinical and urinary biomarkers for predictive modelling of significant disease (Gleason Score (Gs) ≥ 3 + 4). Four predictive models were developed: 'MassSpec' (CE-MS proteomics), 'EV-RNA', and 'SoC' (standard of care) clinical data models, alongside a fully integrated omics-model, deemed 'ExoSpec'. ExoSpec (incorporating four gene transcripts, six peptides, and two clinical variables) is the best model for predicting Gs ≥ 3 + 4 at initial biopsy (AUC = 0.83, 95% CI: 0.77−0.88) and is superior to a standard of care (SoC) model utilising clinical data alone (AUC = 0.71, p < 0.001, 1000 resamples). As the ExoSpec Risk Score increases, the likelihood of higher-grade PCa on biopsy is significantly greater (OR = 2.8, 95% CI: 2.1−3.7). The decision curve analyses reveals that ExoSpec provides a net benefit over SoC and could reduce unnecessary biopsies by 30%.
ABSTRACT
Clinical management of prostate cancer is challenging because of its highly variable natural history and so there is a need for improved predictors of outcome in non-metastatic men at the time of diagnosis. In this study we calculated the model score from the leading clinical multivariable model, PREDICT prostate, and the poor prognosis DESNT molecular subtype, in a combined expression and clinical dataset that were taken from malignant tissue at prostatectomy (n = 359). Both PREDICT score (p < 0.0001, IQR HR = 1.59) and DESNT score (p < 0.0001, IQR HR = 2.08) were significant predictors for time to biochemical recurrence. A joint model combining the continuous PREDICT and DESNT score (p < 0.0001, IQR HR = 1.53 and 1.79, respectively) produced a significantly improved predictor than either model alone (p < 0.001). An increased probability of mortality after diagnosis, as estimated by PREDICT, was characterised by upregulation of cell-cycle related pathways and the downregulation of metabolism and cholesterol biosynthesis. The DESNT molecular subtype has distinct biological characteristics to those associated with the PREDICT model. We conclude that the inclusion of biological information alongside current clinical prognostic tools has the potential to improve the ability to choose the optimal treatment pathway for a patient.
Subject(s)
Biomarkers, Tumor , Prostatic Neoplasms , Male , Humans , Prostatic Neoplasms/surgery , Prostatic Neoplasms/diagnosis , Treatment Outcome , Prognosis , Prostate-Specific AntigenABSTRACT
The Prostate Urine Risk (PUR) biomarker is a four-group classifier for predicting outcome in patients prior to biopsy and for men on active surveillance. The four categories correspond to the probabilities of the presence of normal tissue (PUR-1), D'Amico low-risk (PUR-2), intermediate-risk (PUR-3), and high-risk (PUR-4) prostate cancer. In the current study we investigate how the PUR-4 status is linked to Gleason grade, prostate volume, and tumor volume as assessed from biopsy (n = 215) and prostatectomy (n = 9) samples. For biopsy data PUR-4 status alone was linked to Gleason Grade group (GG) (Spearman's, ρ = 0.58, p < 0.001 trend). To assess the impact of tumor volume each GG was dichotomized into Small and Large volume cancers relative to median volume. For GG1 (Gleason Pattern 3 + 3) cancers volume had no impact on PUR-4 status. In contrast for GG2 (3 + 4) and GG3 (4 + 3) cancers PUR-4 levels increased in large volume cancers with statistical significance observed for GG2 (p = 0.005; Games-Howell). These data indicated that PUR-4 status is linked to the presence of Gleason Pattern 4. To test this observation tumor burden and Gleason Pattern were assessed in nine surgically removed and sectioned prostates allowing reconstruction of 3D maps. PUR-4 was not correlated with Gleason Pattern 3 amount, total tumor volume or prostate size. A strong correlation was observed between amount of Gleason Pattern 4 tumor and PUR-4 signature (r = 0.71, p = 0.034, Pearson's). These observations shed light on the biological significance of the PUR biomarker and support its use as a non-invasive means of assessing the presence of clinically significant prostate cancer.
ABSTRACT
Os acromiale is a relatively common anatomic variant that can occasionally be associated with shoulder pain. Several surgical options to address a symptomatic os acromiale that has failed nonoperative treatment have been described. Published techniques, however, are often very invasive, technically challenging, and carry the risk of potential complications that can be difficult to manage. The technique presented here describes a relatively simple arthroscopic alternative, coined by the authors as the "Wallow technique" due to the fact that the arthroscopic shaver is used to rotate within and resect the os site, that results in complete resection of the os acromiale pseudoarthrosis and avoids the need for an open approach or the use of implants.
ABSTRACT
The objective is to develop a multivariable risk model for the non-invasive detection of prostate cancer prior to biopsy by integrating information from clinically available parameters, Engrailed-2 (EN2) whole-urine protein levels and data from urinary cell-free RNA. Post-digital-rectal examination urine samples collected as part of the Movember Global Action Plan 1 study which has been analysed for both cell-free-RNA and EN2 protein levels were chosen to be integrated with clinical parameters (n = 207). A previously described robust feature selection framework incorporating bootstrap resampling and permutation was applied to the data to generate an optimal feature set for use in Random Forest models for prediction. The fully integrated model was named ExoGrail, and the out-of-bag predictions were used to evaluate the diagnostic potential of the risk model. ExoGrail risk (range 0-1) was able to determine the outcome of an initial trans-rectal ultrasound guided (TRUS) biopsy more accurately than clinical standards of care, predicting the presence of any cancer with an area under the receiver operator curve (AUC) = 0.89 (95% confidence interval(CI): 0.85-0.94), and discriminating more aggressive Gleason ≥ 3 + 4 disease returning an AUC = 0.84 (95% CI: 0.78-0.89). The likelihood of more aggressive disease being detected significantly increased as ExoGrail risk score increased (Odds Ratio (OR) = 2.21 per 0.1 ExoGrail increase, 95% CI: 1.91-2.59). Decision curve analysis of the net benefit of ExoGrail showed the potential to reduce the numbers of unnecessary biopsies by 35% when compared to current standards of care. Integration of information from multiple, non-invasive biomarker sources has the potential to greatly improve how patients with a clinical suspicion of prostate cancer are risk-assessed prior to an invasive biopsy.
ABSTRACT
The highly heterogeneous clinical course of human prostate cancer has prompted the development of multiple RNA biomarkers and diagnostic tools to predict outcome for individual patients. Biomarker discovery is often unstable with, for example, small changes in discovery dataset configuration resulting in large alterations in biomarker composition. Our hypothesis, which forms the basis of this current study, is that highly significant overlaps occurring between gene signatures obtained using entirely different approaches indicate genes fundamental for controlling cancer progression. For prostate cancer, we found two sets of signatures that had significant overlaps suggesting important genes (p < 10-34 for paired overlaps, hypergeometrical test). These overlapping signatures defined a core set of genes linking hormone signalling (HES6-AR), cell cycle progression (Prolaris) and a molecular subgroup of patients (PCS1) derived by Non Negative Matrix Factorization (NNMF) of control pathways, together designated as SIG-HES6. The second set (designated SIG-DESNT) consisted of the DESNT diagnostic signature and a second NNMF signature PCS3. Stratifications using SIG-HES6 (HES6, PCS1, Prolaris) and SIG-DESNT (DESNT) classifiers frequently detected the same individual high-risk cancers, indicating that the underlying mechanisms associated with SIG-HES6 and SIG-DESNT may act together to promote aggressive cancer development. We show that the use of combinations of a SIG-HES6 signature together with DESNT substantially increases the ability to predict poor outcome, and we propose a model for prostate cancer development involving co-operation between the SIG-HES6 and SIG-DESNT pathways that has implication for therapeutic design.
Subject(s)
Biomarkers, Tumor/genetics , Prostatic Neoplasms , Transcriptome , Biomarkers, Tumor/analysis , Cohort Studies , Datasets as Topic/statistics & numerical data , Disease Progression , Gene Expression Profiling , Gene Expression Regulation, Neoplastic , Humans , Male , Microarray Analysis , Neoplasm Invasiveness , Prognosis , Prostatic Neoplasms/diagnosis , Prostatic Neoplasms/genetics , Prostatic Neoplasms/pathologyABSTRACT
An amendment to this paper has been published and can be accessed via a link at the top of the paper.
ABSTRACT
BACKGROUND: Prostate cancer exhibits severe clinical heterogeneity and there is a critical need for clinically implementable tools able to precisely and noninvasively identify patients that can either be safely removed from treatment pathways or those requiring further follow up. Our objectives were to develop a multivariable risk prediction model through the integration of clinical, urine-derived cell-free messenger RNA (cf-RNA) and urine cell DNA methylation data capable of noninvasively detecting significant prostate cancer in biopsy naïve patients. METHODS: Post-digital rectal examination urine samples previously analyzed separately for both cellular methylation and cf-RNA expression within the Movember GAP1 urine biomarker cohort were selected for a fully integrated analysis (n = 207). A robust feature selection framework, based on bootstrap resampling and permutation, was utilized to find the optimal combination of clinical and urinary markers in a random forest model, deemed ExoMeth. Out-of-bag predictions from ExoMeth were used for diagnostic evaluation in men with a clinical suspicion of prostate cancer (PSA ≥ 4 ng/mL, adverse digital rectal examination, age, or lower urinary tract symptoms). RESULTS: As ExoMeth risk score (range, 0-1) increased, the likelihood of high-grade disease being detected on biopsy was significantly greater (odds ratio = 2.04 per 0.1 ExoMeth increase, 95% confidence interval [CI]: 1.78-2.35). On an initial TRUS biopsy, ExoMeth accurately predicted the presence of Gleason score ≥3 + 4, area under the receiver-operator characteristic curve (AUC) = 0.89 (95% CI: 0.84-0.93) and was additionally capable of detecting any cancer on biopsy, AUC = 0.91 (95% CI: 0.87-0.95). Application of ExoMeth provided a net benefit over current standards of care and has the potential to reduce unnecessary biopsies by 66% when a risk threshold of 0.25 is accepted. CONCLUSION: Integration of urinary biomarkers across multiple assay methods has greater diagnostic ability than either method in isolation, providing superior predictive ability of biopsy outcomes. ExoMeth represents a more holistic view of urinary biomarkers and has the potential to result in substantial changes to how patients suspected of harboring prostate cancer are diagnosed.
Subject(s)
Cell-Free Nucleic Acids/urine , DNA Methylation , DNA/urine , Models, Genetic , Prostatic Neoplasms/genetics , Prostatic Neoplasms/urine , Adult , Biomarkers, Tumor/genetics , Biomarkers, Tumor/urine , Cell-Free Nucleic Acids/genetics , Cohort Studies , DNA/genetics , Humans , Male , Middle Aged , Multivariate Analysis , Neoplasm Grading , Prostatic Neoplasms/pathology , Risk AssessmentABSTRACT
BACKGROUND: Unsupervised learning methods, such as Hierarchical Cluster Analysis, are commonly used for the analysis of genomic platform data. Unfortunately, such approaches ignore the well-documented heterogeneous composition of prostate cancer samples. Our aim is to use more sophisticated analytical approaches to deconvolute the structure of prostate cancer transcriptome data, providing novel clinically actionable information for this disease. METHODS: We apply an unsupervised model called Latent Process Decomposition (LPD), which can handle heterogeneity within individual cancer samples, to genome-wide expression data from eight prostate cancer clinical series, including 1,785 malignant samples with the clinical endpoints of PSA failure and metastasis. RESULTS: We show that PSA failure is correlated with the level of an expression signature called DESNT (HR = 1.52, 95% CI = [1.36, 1.7], P = 9.0 × 10-14, Cox model), and that patients with a majority DESNT signature have an increased metastatic risk (X2 test, P = 0.0017, and P = 0.0019). In addition, we develop a stratification framework that incorporates DESNT and identifies three novel molecular subtypes of prostate cancer. CONCLUSIONS: These results highlight the importance of using more complex approaches for the analysis of genomic data, may assist drug targeting, and have allowed the construction of a nomogram combining DESNT with other clinical factors for use in clinical management.