Abstract
Purpose:
To develop and validate a new tissue-based biomarker that improves prediction of outcomes in localized prostate cancer by quantifying the host response to tumor.
Materials and Methods:
We use digital image analysis and machine learning to develop a biomarker of the prostate stroma called quantitative reactive stroma (qRS). qRS is a measure of percentage tumor area with a distinct, reactive stromal architecture. Kaplan Meier analysis was used to determine survival in a large retrospective cohort of radical prostatectomy samples. qRS was validated in two additional, distinct cohorts that include international cases and tissue from both radical prostatectomy and biopsy specimens.
Results:
In the developmental cohort (Baylor College of Medicine, n = 482), patients whose tumor had qRS > 34% had increased risk of prostate cancer-specific death (HR 2.94; p=0.039). This result was replicated in two validation cohorts, where patients with qRS > 34% had increased risk of prostate cancer-specific death (MEDVAMC; n = 332; HR 2.64; p=0.02) and also biochemical recurrence (Canary; n = 988; HR 1.51; p=0.001). By multivariate analysis, these associations were shown to hold independent predictive value when compared to currently used clinicopathologic factors including Gleason score and PSA.
Conclusions:
qRS is a new, validated biomarker that predicts prostate cancer death and biochemical recurrence across three distinct cohorts. It measures host-response rather than tumor-based characteristics, and provides information not represented by standard prognostic measurements.
Keywords: Cancer, prostate, stroma, host response, biomarker
1. Introduction
Prostate cancer is one of the leading causes of morbidity and mortality in men [1]. It is a heterogeneous disease with a wide range of outcomes: some tumors remain localized and indolent while others metastasize and quickly cause morbidity and mortality. Unfortunately, it is difficult to predict which outcome one may expect.
The best predictive tools at our disposal combine clinical and pathologic data. For instance, the D’Amico [2] and Stephenson [3] nomograms, and CAPRA score [4] are widely used and validated models for predicting biochemical recurrence after radical prostatectomy. These models incorporate Gleason score, pre-operative serum PSA, clinical stage, age, and biopsy features in various combinations. They are helpful yet imperfect, and a study by Tarek Bismar et. al. found that these nomograms failed to contain information that predicted the outcome in almost half the patients in their cohort [5]. This suggests that there are clinical and pathologic factors affecting outcomes that are not represented by standard predictive models.
Currently, the host response to tumor presence is not routinely evaluated. When an epithelial tumor invades adjacent connective tissue, there is a vigorous host response that approximates physiologic wound healing [6]. This leads to an altered stromal architecture, which is well characterized in prostate cancer and called the “reactive stroma [7].” A semi-quantitative grading scale of the reactive stroma has been established, with reactive stroma comprising 0–5% of the tumor area being termed reactive stroma grade (RSG) 0, 5–15% as RSG 1, 15–50% as RSG 2, and >50% as RSG 3, respectively [8]. Compared to other reactive stroma grades, RSG 3 in particular is associated with earlier biochemical recurrence [8] and increased risk of death due to prostate cancer [9]. These results have been externally validated in a large population based study from Norway [10] and a study using the Canary cohort [11]. The main limitation of RSG 3 is reproducibility, as a specialized pathologist must estimate the percentage area of the tumor that contains reactive stroma.
This study seeks to re-examine reactive stromal grading by developing a completely quantitative biomarker called quantitative reactive stroma (qRS). We believe that a fully quantitative biomarker is more reproducible and clinically applicable. Instead of RSG 0–3, we have narrowed the evaluation to a single, binary result: qRS positive or negative. We show that qRS predicts prostate cancer-specific death and recurrence, and validate these results in two large international cohorts that include both biopsy and radical prostatectomy specimens (Figure 1).
Figure 1: Study design –
A training cohort of 50 patients was used to train the tissue segmentation software to recognize tissue components. This algorithm was used to quantify reactive stroma in the Baylor College of Medicine cohort. An optimal biomarker cutoff was established as qRS 34%, which was used in two validation cohorts.
2. Materials and Methods
2.1. Definition of Reactive Stroma in Prostate Cancer
The reactive stroma of prostate cancer is characterized by an expanded extracellular matrix with collagen deposition and myofibroblast proliferation. The preponderance of collagen creates a fibrillary and lighter pink appearance and alters gland architecture. These changes distinguish it from healthy prostate, BPH, and non-stromogenic prostate cancer. A more detailed description of the histologic characteristics of the reactive stroma can be found in our previous studies [8] [12].
2.2. Cohorts Studied
The tissue samples used were from three different sources: The Baylor College of Medicine Prostate Cancer database (BCM cohort), the Canary Prostate Cancer Tissue Microarray (Canary) cohort, and the Michael E. DeBakey VA Hospital (MEDVAMC) cohort (Table 1).
Table 1.
Characteristics of the three cohorts analyzed including patients demographics, outcomes, and diagnostics procedure by which tissue was collected.
Cohort Name | Baylor College of Medicine | MEDVAMC | Canary Retrospective Cohort |
---|---|---|---|
Total # of Patients | 850 | 4598 | 1116 |
Demographics | |||
Caucasian | 747 (87.9%) | 2543 (55.3%) | - |
Hispanic | 63 (7.4%) | 110 (2.4%) | - |
African American | 30 (3.5%) | 1536 (33.4%) | - |
Asian/Middle Eastern | 10 (1.2%) | 5 (0.1%) | - |
No response | - | 372 (8%) | - |
Follow-up Time | 3–15 years | ≥ 3 years | |
Outcomes | |||
Recurrence | - | - | 495 (44.4%) |
Death from PCa | 179 (21%) | 864 (18.7%) | - |
Adjuvant Therapy | None | None | None |
Diagnostic Procedure | Radical Prostatectomy | Biopsy | Radical Prostatectomy |
In the BCM cohort, patients were included if they had localized prostate cancer at the time of radical prostatectomy and if they had at least 3 years of follow-up. The cohort consists of 850 patients with follow-up time from 3–15 years. Of the 850 patients in the cohort, 747 identify as Caucasian (87.9%), 63 as Hispanic (7.4%), 30 as African American (3.5%), and 10 as Asian American or Middle Eastern (1.2%). Death from prostate cancer occurred in 179 patients (21%). No patients received adjuvant chemotherapy or radiation therapy. An index tumor with the highest Gleason score was selected, with 2 mm cores and tissue microarray created from that tumor.
In the MEDVAMC cohort, patients were included if they had localized prostate cancer at the time of prostate biopsy. Patients had at least 3 years of follow-up. The cohort consists of 4598 patients. Of the 4598 patients in the cohort, 2543 identify as Caucasian (55.3%), 110 as Hispanic (2.4%), 1536 as African American (33.4%), 5 as Asian or Middle Eastern (0.1%) and 372 had no response to racial identification (8%). Death from prostate cancer occurred in 864 patients (18.7%). Of the cases selected for this study, none had received adjuvant chemotherapy or radiation therapy. An index tumor with the highest Gleason score was selected, with 2 mm cores and tissue microarray created from that tumor.
The Canary cohort has been previously described (Am J Surg Patol., 2016) [11]. The Canary cohort consists of 1116 patients across six institutions, with multi-year follow-up and outcomes of biochemical recurrence, time to detection of metastases, and death due to prostate cancer. The study includes tissue derived from radical prostatectomy surgical specimens. The study included samples from men with (a) recurrent prostate cancer; (b) nonrecurrent prostate cancer; and (c) unknown outcome due to inadequate follow-up time (ie, censoring). Recurrent prostate cancer is defined by (1) a single serum PSA level >0.2 ng/mL more than 8 wk after radical prostatectomy; and/or (2) receipt of salvage or secondary therapy after radical prostatectomy; and/or (3) clinical or radiologic evidence of metastatic disease after radical prostatectomy. Non-recurrent prostate cancer is defined as disease with none of the indicators of recurrence for at least 5 years after radical prostatectomy. Participants with no evidence of recurrent prostate cancer but <5 years of follow-up after radical prostatectomy (ie, censored) were also eligible for the study.
2.3. Digital Image Analysis and Machine Learning Algorithm
The tissue microarrays received from each cohort were converted to high resolution digital images by deconvolution microscopy using a Nuance™ Multispectral Microscope. We then used image segmentation via the Inform™ analysis software package to train an algorithm to recognize cellular components of the reactive stroma. Training involved manually designating areas of smooth muscle, reactive stroma, and epithelium on 50 digitized cases from the BCM cohort that were not included in the subsequent statistical analysis. This is how the Inform™ software learned to recognize tissue components so the calculation of percentage reactive stroma within the tumor could be automated (Figure 2). Once trained, the algorithm was very consistent and reliable: 50 randomly chosen tissue samples were determined to have the same (to tenth of a decimal) respective qRS across five measurements.
Figure 2: Segmented Tissue -.
Hematoxylin and eosin deconvoluted images on the left. Corresponding segmented images on the right. Smooth muscle fibers are colored in red while reactive stroma is colored purple. Individual cell nuclei and cytoplasm are identified in each compartment.
2.3. Cutoff Selection
The minimum p-value approach [14] was used to determine optimal cutoff points to dichotomize the qRS biomarker into high and low risk patient groups within the BCM cohort. A threshold of qRS 34% was determined to be the optimal threshold that maximized the difference in outcomes between each group. This cutoff was used in all three cohorts, including the Canary and VA cohorts, to validate its use in risk stratification.
2.4. Statistical Analysis
Analysis for the BCM and MEDVAMC cohorts were done at UT Houston by MH Lee, and the Canary cohort was analyzed in the central statistical analysis core at MD Anderson. Kaplan-Meier actuarial analysis and the log-rank test were used to evaluate the univariate associations between our biomarkers and outcomes. Test of proportional hazards assumption was performed and no statistically significant evidence of violations was found. Hazard ratios (HR) from Cox proportional hazard models and odds ratios (OR) from logistic regression are expressed with 95% confidence intervals and p-values. For all analysis, p-value < 0.05 is considered significant. Analyses were completed using SAS 9.4 software (SAS Institute Inc., Cary, NC).
2.5. IRB
All studies were approved by the University of Texas McGovern Medical School IRB under protocol HSC-MS-13–0094.
3. Results
3.1. High qRS and Risk of Prostate Cancer Death in BCM Cohort
After training the computer algorithm to accurately measure qRS, the remaining cases in the BCM cohort were analyzed. One of the primary outcomes measured in this cohort was prostate cancer-specific death. Of the patients in the cohort, 378 of 482 (78%) cases had qRS > 34%. Mean survival in this group was 162.8 months compared to 188.5 months for patients with qRS < 34% (Figure 3A). By multivariate analysis, qRS > 34% was associated with 2.9 fold increased risk of prostate cancer death (HR 2.94; P = 0.039) after adjusting for Gleason score, PSA, and other commonly used prognostic factors (Figure 3B).
Figure 3: BCM-TMA Survival Curve and Logistic regression –
(A) Patients whose tumors have qRS > 34% have shorter disease-specific survival compared to patients with less reactive stroma in the tumor microenvironment. (B) Multivariate analysis confirms statistical significance that is independent of other standard clinicopathologic variables.
3.2. Validation of qRS in Canary and MEDVAMC Cohorts
We analyzed tissue in two additional cohorts using the same cutoff of qRS = 34%. In the MEDVAMC cohort, 216 of 332 (65%) biopsies had qRS > 34%. Mean survival in this group was 112 months compared to 164.9 months (Figure 3A). By multivariate analysis, qRS > 34% was associated with 2.8 fold increased risk of prostate cancer death (HR 2.78; P = 0.02) after adjusting for Gleason score and PSA (Figure 3B).
In the Canary cohort, 261 of 988 (26%) cases had qRS > 34% (Figure 3C). By multivariate analysis adjusting for surgical margins, extracapsular extension, and seminal vesicle invasion, patients with qRS > 34% had significantly earlier biochemical tumor recurrence (HR 1.51; P = 0.001) (Figure 3D).
4. Discussion
There is a growing body of evidence that shows a reactive stromal response can predict poor prostate cancer outcomes. In this study, we are able to expand the total amount of data significantly by using machine learning and digital analysis. Importantly, this study includes external validation of the qRS biomarker, both with respect to the cohorts used and the use of both radical prostatectomy and preoperative prostate biopsy tissue.
We believe that incorporating quantitative reactive stroma grading into prognostic models will improve prediction of prostate cancer outcomes. It is commonly accepted that tumor progression results from a coevolution of the tumor with its microenvironment7 but most cancer biomarkers focus on the cancer cell proper without regard to the influence of the host response in the development and progression of disease. Prostate cancer is no exception, with Gleason score and PSA being epithelial-driven markers. Clinical evidence demonstrating the utility of reactive stroma grading has been summarized in a recent review [15], which outlines domestic and international case-control studies with outcomes that include increased biochemical recurrence rates [8][10][11][16], decreased survival [9] [10], and progression to hormone resistance [17]. Biologic studies [18] and mathematical models [19] also suggest that the stroma plays a role in determining tumor development and aggressiveness.
Reactive stroma grading is thoroughly studied, and qRS is a reproducible and validated biomarker of the host response to prostate cancer. The history of outcomes prediction in prostate cancer has trended towards inclusion of an increasing number of biologic and clinical factors. We propose that including qRS in the standard prostate cancer prediction models would create a more comprehensive set of biomarkers that improves prognostic accuracy by accounting for the role of the reactive stroma in tumor biology.
5. Conclusions
qRS is a novel biomarker that predicts risk of prostate cancer-specific death and biochemical recurrence after radical prostatectomy and preoperative biopsy. This biomarker has the potential to add information to standard predictive models and clarify the heterogeneity seen in prostate cancer outcomes.
Figure 4: Validation in VA and Canary Cohorts -.
(A) Patients from the VA cohort whose tumors have qRS > 34% have shorter disease-specific survival compared to patients with qRS less than 34%. (B) This risk is confirmed by multivariate analysis. (C) Patients from the Canary cohort whose tumors have qRS > 34% have shorter time to biochemical recurrence (PSA > 0.02 ng/ml) than patients with qRS less than 34%. (D) This risk is confirmed with multivariate analysis.
Outcomes in localized prostate can are difficult to predict
Evaluation of the host response and tumor microenvironment is not utilized in current predictive models
Quantification of the reactive stroma shows that vigorous stromal response increases risk of recurrence and prostate cancer-specific death
These findings are validated in two external cohorts
Funding disclosures:
This work was supported by National Institutes of Health grants RO1 CA140734-03, P50CA097186, R01CA165573, R01CA181605 and an award from the Canary Foundation. The NIH/NCI SPORE in Prostate Cancer: P50CA58236, and the NIH/NCI U01 CA196390. The U.S. Department of Defense Prostate Cancer Research Program (PCRP), DoD Award #s W81XWH-14-2-0182 and W81XWH-14-2-0183 Prostate Cancer Biorepository Network (PCBN) and NIH P30 CA006973.
Footnotes
Conflict of Interest: authors disclose no financial conflicts of interest
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
References
- 1.Siegel RL, Miller KD, Jemal A. Cancer statistics, 2016. CA Cancer J Clin 2016;66:7–30. [DOI] [PubMed] [Google Scholar]
- 2.D’Amico AV, Whittington R, Malkowicz SB, et al. Biochemical outcome after radical prostatectomy, external beam radiation therapy, or interstitial radiation therapy for clinically localized prostate cancer. JAMA. 1998; 280: 969–74. [DOI] [PubMed] [Google Scholar]
- 3.Stephenson AJ, Scardino PT, Eastham JA, et al. Preoperative nomogram predicting the 10-year probability of prostate cancer recurrence after radical prostatectomy. J Natl Cancer Inst 2006; 98: 715–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Cooperberg MR, Pasta DJ, Elkin EP, et al. The University of California, San Francisco Center of the Prostate Risk Assessment score: a straightforward and reliable preoperative predictor of disease recurrence after radical prostatectomy. J Urol 2005; 173: 1938–42. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Bismar TA, Humphrey P, Vollmer RT. Information content of five nomograms for outcomes in prostate cancer. Am J Clin Pathol 2007; 128: 803–7. [DOI] [PubMed] [Google Scholar]
- 6.Hanahan D, Weinberg RA. Hallmarks of cancer: the next generation. Cell 2011; 144: 646–74. [DOI] [PubMed] [Google Scholar]
- 7.Barron DA, Rowley DR. The reactive stroma microenvironment and prostate cancer progression. Endocrin Relat Cancer. 2012; 19: 187–204. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Ayala GE, Tuxhorn JA, Wheeler TM, et al. Reactive stroma as a predictor of biochemical-free recurrence in prostate cancer. Clin Cancer Res. 2003; 9: 4792–801. [PubMed] [Google Scholar]
- 9.Ayala GE, Muezzinoglu B, Hammerich KH, et al. Determining prostate cancer-specific death through quantification of stromogenic carcinoma area in prostatectomy specimens. Am J Pathol. 2011; 178: 79–87. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Saeter T, Vlatkovic L, Waaler G, et al. The prognostic value of reactive stroma on prostate needle biopsy: a population-based study. Prostate. 2015; 75: 662–71. [DOI] [PubMed] [Google Scholar]
- 11.McKenney JK, Wei W, Hawley S, et al. Histologic grading of prostatic adenocarcinoma can be further optimized: analysis of the relative prognostic strength of individual architectural patterns in 1275 patients from the Canary Retrospective Cohort. Am J Surg Pathol. 2016; 40: 1439–56. [DOI] [PubMed] [Google Scholar]
- 12.De Vivar AD, Sayeeduddin M, Rowley D et al. Histologic features of stromogenic carcinoma of the prostate. Hum Pathol. 2017; 63: 202–211. [DOI] [PubMed] [Google Scholar]
- 13.Hawley S, Fazli L, McKenney JK. A model for the design and construction of a resource for the validation of prognostic prostate cancer biomarkers: the Canary Prostate Cancer Tissue Microarray. Adv Anat Pathol. 2013; 20: 39–44 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Chen Y, Huang J, He X, et al. A novel approach to determine two optimal cut-points of a continuous predictor with a U-shaped relationship to hazard ratio in survival data: simulation and application. BMC Med Res Methodol 2019; 19: 96. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Miles B, Ittmann M, Wheeler T, et al. Moving beyond Gleason scoring. Arch Pathol Lab Med. 2019; 143: 565–570. [DOI] [PubMed] [Google Scholar]
- 16.Billis A, Meirelles L, Freitas LL, et al. Adenocarcinoma on needle prostatic biopsies: does reactive stroma predict biochemical recurrence in patients following radical prostatectomy? Int Braz J Urol. 2013; 39: 320–7. [DOI] [PubMed] [Google Scholar]
- 17.Wu JP, Huang WB, Zhou H, et al. Intensity of stromal changes is associated with tumor relapse in clinically advanced prostate cancer after castration therapy. Asian J Androl. 2014; 16: 710–4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Valkenburg KC, de Groot AE, Pienta KJ. Targeting the tumour stroma to improve cancer therapy. Nat Rev Clin Oncol. 2018; 15: 366–81. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Frankenstein Z, Basanta D, Franco O, et al. Stromal reactivity differentially drives tumour cell evolution and prostate cancer progression. Nat Ecol Evol. 2020; 4: 870–84. [DOI] [PMC free article] [PubMed] [Google Scholar]