Although prostate cancer (PCa) is the most commonly diagnosed non-cutaneous cancer of men in the United States [1], few prognostic biomarkers are available for routine clinical use. Serum PSA biochemical failure (BF) versus non-failure is a well-established binary outcome variable. Since clinical treatment failure (systemic progression and/or local recurrence) is essentially always preceded by BF (by a median of 8 years) [2–5], PSA non-failure has high negative predictive value (NPV) for poor clinical outcome after prostatectomy. When identifying and validating new potential tissue based biomarkers for the purpose of attempting to improve prognostication after prostatectomy, the greatest clinical risk would be to mischaracterize biologically aggressive disease as “non-aggressive,” since that would lead to erroneous underestimation of the malignant potential of the disease. For this reason, we seek tissue based biomarkers that maximally correlate with PSA failure.

Multibiomarker signatures and gene associations have been implicated in PCa progression but few markers emerge among multiple independent studies as potentially prognostic, outperform established clinicopathologic parameters in predicting BF following prostatectomy, or have become routinely used in clinical labs [6–18]. Current prognostic indicators in PCa include preoperative PSA, tumor stage, and grade [19]. Prognostic molecular biomarkers may identify pathways that can be exploited for therapy. For example, the TMPRSS2/ERG gene fusion was identified as a frequent chromosomal rearrangement present in a subgroup of prostate tumors that leads to ERG transcription factor overexpression [20]. Subsequent studies have demonstrated that inhibiting this pathway by delivery of liposomal nanovectors carrying siRNA specific for the TMPRSS2/ERG fusion transcript provides therapeutic advantage in a murine PCa model [21]. Additionally, multibiomarker signatures can be spatially quantified to aid in the development of novel imaging methods to assess PCa aggressiveness preoperatively in vivo through co-registration of postoperative pathology data with preoperative imaging data [22].

Numerous studies have described biomarkers of aggressive biologic behavior in PCa such as those correlated with aberrant hyaluronan (HA) processing [8, 11, 12], neuroendocrine phenotype [14, 23], increased tumor angiogenesis [24], and poor prognosis [7, 25]. Other groups have developed multi-biomarker signatures of PCa aggressiveness using gene expression profiling as a method for biomarker discovery; these studies have surprisingly little overlap between gene sets [6, 9, 10, 13, 15–18]. In this study, we took a combined approach of a) non-biased, cross-study examination of gene expression profiling data and b) a candidate gene approach (gene products that appear to be associated with PCa outcome in prior publications), to identify 33 potential aggressiveness biomarkers. We evaluated the association of these biomarkers with BF using immunohistochemistry and automated digital pathology techniques on tissue microarrays (TMAs) representing PCa tissue from 160 prostatectomy specimens.

While established clinicopathologic features including grade, stage, and preoperative PSA partially explain the variability in PCa outcome measured by BF after prostatectomy, further stratification of patients for more precise patient management may be possible with prognostic multibiomarker signatures. Many signatures and gene associations of PCa progression have been identified [6–18, 23–25], but few markers emerge among multiple independent studies as potentially prognostic, outperform established clinicopathologic features, or become widely used. Identification of robust prognostic biomarkers will not only improve the clinical management of PCa, but is expected to provide an important bridge between pathology data and imaging data to facilitate development of novel imaging biomarkers which can be assessed noninvasively [22]. The present study focused on evaluating the prognostic utility of 33 proteins and investigating multibiomarker signatures that were predictive of BF. Although these proteins have been previously implicated in aggressive biologic behavior of PCa from diverse studies, to our knowledge this is the first study to examine these proteins in a single prostatectomy cohort using immunohistochemistry and objective automated digital pathology methods. We found that increased levels of several proteins in malignant epithelium were significantly associated with time to BF after adjustment for established clinicopathologic parameters including HMMR, HOXC6, IGF1, MAP4K4, and SMAD4. We also demonstrated that a 3-biomarker signature including HMMR, SMAD4, and SIAH2 was predictive of BF in our cohort, warranting further investigation of this signature in larger cohorts and imaging biomarker development studies.

HMMR is a multicompartmentalized receptor for the extracellular matrix carbohydrate HA with potentially oncogenic functions such as mediating cell motility and affecting mitotic spindle integrity [35]. Aberrant HA pathway signaling is implicated in prostate tumor cell proliferation, motility, angiogenesis, and metastasis [36]. Our results indicated that elevated HMMR in malignant epithelium was associated with BF which is similar to previous reports linking HMMR expression to the development of castration-resistant PCa and metastatic disease [12, 37]. Further, low molecular weight HA fragments are catabolized from native HA by physical (reactive oxygen/nitrogen species) and enzymatic (hyaluronidase) mechanisms [38] and bind to HMMR to induce cell migration [39, 40]. These data support the underlying model of aberrant HA-HMMR signaling in aggressive PCa and are encouraging for ongoing studies in our laboratories using small molecular inhibitors to disrupt fragmented HA-HMMR interactions for therapeutic targeting.

HOXC6 was identified in our cross-study analysis of gene expression datasets as part of an 11-gene signature of PCa aggressiveness. We further demonstrated an association between HOXC6 immunohistochemistry with time to BF in our PCa cohort. Similarly, Singh et al. identified HOXC6 in a 5-gene signature that was predictive of BF [15]. HOXC6 is an androgen-regulated gene which assists during development of normal tissues and progression of PCa [41]. Mouse models provide specific evidence for the interaction of HOXC6 with multiple downstream targets including bone morphogenic protein 7 (BMP7), fibroblast growth factor receptor 2 (FGFR2), and platelet-derived growth factor receptor α (PDGFRA) that promote PCa metastasis to the bone microenvironment [42]. Anti-apoptotic roles for HOXC6 have also been described in head and neck squamous cell carcinoma in which HOXC6 directly increases the promoter activity causing overexpression of Bcl-2 [43]. Similarly, HOXC6 represses the activity of pro-apoptotic genes neutral endopeptidase (NEP) and insulin-like growth factor binding protein 3 (IGFBP3) in PCa [41]. Although androgen receptor was not directly investigated in this study, we identified an association between elevated HOXC6 and BF in our cohort which along with these studies provides rationale for further investigation of HOXC6 in the PCa androgen receptor axis.

MAP4K4 is also a pro-migratory protein involved in mammalian development and increases tumor cell motility likely through c-Jun N-terminal kinase (JNK) [44]. Our findings indicated that increased MAP4K4 was significantly associated with time to BF. MAP4K4 is similarly incorporated in numerous independent gene expression signatures which are predictive of survival in colorectal cancer [45] and recurrence in prostate cancer [16, 17]. Additionally, MAP4K4 is an independent prognostic factor for hepatocellular carcinoma and lung adenocarcinoma [46, 47]. Xenograft tumor growth in mice using a hepatocellular cell line is substantially inhibited by RNA interference of MAP4K4 [46], indicating a potential therapeutic target that may be useful for treatment of PCa.

Higher protein levels of IGF1 were associated with time to BF in our PCa cohort. This observation was discordant from our initial cross-study analysis of independent gene expression datasets identifying downregulated IGF1 as part of an 11-gene signature of PCa aggressiveness (Additional file 12: Table S11). One recent study shows evidence for decreased IGF1 mRNA in local PCa compared to benign prostate, although IGF1 may still be involved in subsequent tumor progression as this study did not evaluate advanced or metastatic disease [48]. In contrast, our immunohistochemical findings correlate with most literature on the role of increased IGF1 in cancer progression. In androgen-independent PCa, IGF1 induces tumor cell motility by activation of αvβ3 integrin via the PI3-K/Akt pathway [49]. IGF1 signaling through PI3-K/Akt and β1 integrin similarly promotes adhesion and migration in multiple myeloma cells [50].

The PI3-K pathway is also upregulated in transgenic mice expressing tissue-specific IGF1 in prostate basal epithelial cells which leads to constitutive activation of IGF1R and increased development of PCa [51]. Importantly, meta-analysis of large datasets demonstrates an association between high serum IGF1 levels and moderately increased risk of PCa which may provide an important modifiable target in PCa patients [52].

In addition to IGF1, bioinformatic analysis identifies SMAD4 as a candidate gene likely to be a primary driver of PCa progression [53]. Reports of SMAD4 expression in prostate and other cancers are complex and variable, although mutation of the SMAD4 gene appears rare during PCa progression [54]. In normal cells, SMAD4 is a critical component of the TGF-β signaling cascade and localizes to the nucleus after becoming activated to regulate TGF-β-responsive genes and inhibit cellular proliferation [55]. In this sense, SMAD4 may act as a tumor suppressor, which has been demonstrated by reduced expression of SMAD4 in PCa compared to BPH and normal prostate tissue [56]. Interestingly, in our cross-study analysis of gene expression datasets SMAD4 was present in the top 46 ranked genes (position 21) and inversely correlated with PCa aggressiveness (Additional file 4: Table S3). While Smad-dependent TGF-β signaling primarily functions to inhibit growth, the vast majority of tumors acquire resistance to these effects and tumor progression becomes stimulated by TGF-β in more advanced tumors [55]. Previous studies have linked increased SMAD4 to higher grade, stage, and DNA ploidy in PCa [57] and to infiltration of the myometrial wall in endometrioid endometrial cancer [58], which is consistent with our immunohistochemical analysis that elevated SMAD4 was significantly associated with time to BF. Similarly, SMAD4 shifts from its tumor suppressor role to an aggressiveness factor in a mouse model of breast cancer by promoting bone metastasis through TGF-β-activated expression of IL-11 [59]. Advanced primary prostate tumors often metastasize to the bone, potentially reflecting parallel events whereby bone-derived TGF-β offers an advantage to SMAD4-overexpressing prostate tumor cells and providing a possible therapeutic target in the TGF-β pathway.

HMMR, SMAD4, and SIAH2 comprised a multibiomarker signature that was predictive of BF in our cohort. SIAH2 is a RING finger ubiquitin ligase which controls the stability of multiple substrates, and under hypoxic conditions, causes ubiquitination/degradation of prolyl hydroxylase 3 and 1 [60]. This action increases the accumulation rate of HIF-1α and interacts with neuroendocrine-specific expression of FoxA2 leading to neuroendocrine PCa development and metastasis [14, 60]. SIAH2 also contributes to castration-resistant PCa by targeting a subset of inactive androgen receptors for ubiquitination which increases the activity of androgen receptor target genes implicated in PCa progression [61]. Our results which identify SIAH2 in the multibiomarker signature predictive of BF are consistent with these studies and support a functional role for SIAH2 in contributing to aggressive subtypes of PCa.

This study and its conclusions are limited by a small sample size and short follow-up for some individuals. We collected data on 160 patients of which half were followed for less than 2.3 years. As a result, only 22 of the 160 patients in our sample experienced BF even though 67 patients had extraprostatic extension, lymph node involvement, or surgical margin involvement. It is likely that longer follow-up would result in more BF instances. The small number of BF’s observed in our sample is directly related to the relatively modest AUC (AUC ~0.7 for 3- and 5-year BF) observed for our 3-biomarker signature. A larger sample size and more events would provide more statistical power for developing a biomarker signature and more precision for evaluating the performance of the new biomarker signature. This retrospective hospital-based cohort study is additionally limited by lack of information regarding (neo)-adjuvant therapy. This is an important limitation of our study, since such treatments could potentially affect patient outcomes and thus our interpretations. We plan to complete a larger validation study to further refine our biomarker signature and to obtain more precise estimates of its performance as a classifier for BF.

This study presents a unique methodology for evaluating the prognostic utility of PCa biomarkers. Because of the precious value of tissues represented in TMAs from large cohorts, we have tested 33 proteins by immunohistochemistry using this cohort of 160 patients in order to identify promising biomarkers for use in significantly larger validation studies [62]. Additionally, this study describes a 3-biomarker signature consisting of HMMR, SIAH2, and SMAD4 proteins which appears to be prognostic in our cohort. This signature may be useful in work currently underway to co-register detailed multibiomarker immunohistochemistry signature maps with multiparametric MR data, for the purpose of identifying MR biomarkers that assess PCa aggressiveness preoperatively in vivo[22].