Loss of SNAI2 in Prostate Cancer Correlates With Clinical Response to Androgen Deprivation Therapy

Clinical Trial

We performed an open-label nonrandomized neoadjuvant phase II study (ACTRN12612000772842) in men with prostate cancer with high-risk features (PSA > 20 ng/dL, predominant cancer Gleason pattern 4 or above, or clinical stage ≥ cT2c) with no evidence of bony metastatic disease (Data Supplement). Patients received degarelix 240/80 mg subcutaneously every 4 weeks, abiraterone acetate 500 mg orally daily titrating upward every 2 weeks by 250 mg to a final dose of 1,000 mg daily, bicalutamide 50 mg orally daily, and prednisolone 5 mg orally twice daily. Treatment was for 24 weeks (six cycles), and oral medications were continued up until the day of surgery. Sample size was calculated on the basis of an anticipated complete pathologic response (complete response [CR]) rate and used Simon's¹² two-stage phase II trial design. On the basis of a historical CR rate of approximately 5% with conventional castration and the assumption that a CR rate of 25% would be of clinical interest for examination in a larger phase III trial setting, a maximum of 17 patients needed to be accrued to have 80% power to detect a significant difference at α = .05 level. From the initial nine patients, at least one observed CR was necessary to trigger recruitment to the second stage, with at least three CRs in the final cohort required to reject the null hypothesis. All study investigations had institutional review board approval.

All patients provided written informed consent before enrollment. All study interventions and investigations were approved by the Human Research Ethics Committee at Melbourne Health (HREC 2012.220).

Fresh Tissue Banking

Frozen section confirmed fresh tumor and benign tissue were collected at the time of prostatectomy as previously described.¹³

Pathologic Assessment

Specimens were routinely fixed and processed and tumor volume calculated as previously described.¹⁴ CR was defined as no identifiable residual tumor (ypT0). No response was defined as absence of histologic evidence of tumor involution. Partial response was defined as identifiable residual tumor with variable histologic evidence of tumor involution. Although not prespecified in the study Protocol, for consistency with recent reports, minimal residual disease was defined as a residual tumor volume ≤ 0.2 cc¹⁵ with no high-grade elements present (Gleason score ≤ 3 + 3). Tumor volumes were not corrected for cellularity.

Experimental Methods

A detailed description of the experimental procedures used is provided in the Data Supplement.

Objective Responses to Profound AR Inhibitors Are Highly Variable and Not Associated With Established CRPC Drivers

Seventeen patients were enrolled and completed neoadjuvant treatment, with one patient declining to proceed with prostatectomy. Despite all patients demonstrating a decrease in serum PSA > 95% (Data Supplement), objective pathologic responses were highly variable (Figs 1A and 1B). Four of 16 evaluable patients treated had sensitive cancers, with no remaining tumor (n = 1) or minimal residual disease (n = 3) identified in the prostatectomy specimen. In contrast, an identical number of patients had no measurable treatment effect observed. The remaining patients (n = 8) had partial tumor responses, with significant volumes of residual tumor but histologic evidence of at least some tumor regression. Heterogeneity in tumor response could not be explained by differences in the pretreatment disease characteristics of the patient groups (Data Supplement), and there was no correlation between PSA nadir and residual tumor volume (Data Supplement). The risk of disease recurrence, however, was closely linked to the extent of tumor regression (Data Supplement).

Reactivation of AR signaling has been demonstrated to be an important driver of progressive castration-resistant disease.¹ We therefore assessed AR expression in pre- and post-treatment samples by immunohistochemistry and found it to be detectable and localized to the nucleus in the majority of persistent tumors, suggestive of ongoing activation (Fig 1C), although there was no difference in receptor staining or localization based on the level of treatment response (Fig 1D). A number of different mechanisms have been shown to reactivate AR signaling in castration-resistant disease. These include overexpression of the AR (frequently associated with AR amplification),³ which confers supersensitivity to subcastrate signaling androgen levels, the emergence of constitutively active receptor splice variants (in particular ARv7)⁵, mutations in the ligand-binding domain that permit promiscuous ligand binding, and an increase in intratumoral androgen synthesis.¹⁶ In addition, the presence of ARv7 in pretreatment samples has been reported to predict response to newer hormonal therapies such as abiraterone.¹⁷ However, we found no significant levels of signaling androgens in persistent tumors (Fig 1E), nor did we find an increase in the expression of the AR or its splice variants (Fig 1F and Data Supplement) or any mutations in AR coding regions, indicating that the mechanisms regulating castration persistence are distinct from those involved in castration resistance. In addition, ARv7 expression in pretreatment biopsy specimens was negligible compared with the full length receptor and did not predict treatment response (Data Supplement).

Tumor Cells Surviving Acute Pathway Inhibition Demonstrate an Incomplete Epithelial-to-Mesenchymal Transition Phenotype

Accumulating evidence suggests that transcriptional reprogramming is a key event in acute resistance to therapy in tumor cells, usually associated with the emergence of a therapy-resistant phenotype.^18-22 To probe this, we profiled global transcription in fresh-frozen residual tumors by RNA sequencing and observed clear differences in the transcriptional profile of resistant tumors compared with matched fresh-frozen untreated samples (Fig 2A; PC1 and PC2 capture 29.87% and 18.63% of the variance, respectively). Enrichment analysis revealed significant upregulation of gene sets associated with phenotype change, including stem-like and epithelial-to-mesenchymal transition (EMT), as well as genes regulated by chromatin modification (Fig 2B). Androgen-regulated genes were significantly downregulated, despite the presence and nuclear location of AR in persistent tumors, as were gene sets associated with cell proliferation (Fig 2B). We then interrogated the expression of key differentiation and regulatory genes in a number of complementary RNA-seq experiments (Fig 2C and Data Supplement) and found consistent upregulation of both markers of EMT and stemness in resistant tumor tissue. For EMT markers, similar levels of expression were observed in treated benign glands, but not in progressive castration-resistant samples, suggesting that this adaptive response is enriched in prostate epithelium surviving short-term AR pathway inhibition, although the number of CRPC samples analyzed is small. In contrast, upregulation of stem cell markers was observed in both disease states and in treated benign samples. To confirm that EMT was occurring, we measured the levels of the epithelial differentiation marker E-cadherin, loss of which is a key defining feature of EMT, and found consistent loss of expression in persistent tumors (Fig 2D). However, significant variability in the level of expression within individual cells was observed. Similar levels of E-cadherin downregulation were also identified in persistent luminal cells within benign glands (Data Supplement), which also expressed predominantly nuclear AR (Data Supplement), confirming that the adaptive response that permits epithelial survival is not cancer cell–specific. Despite high staining for the key mesenchymal marker vimentin in the stroma of persistent tumors, we observed no instance of epithelial staining, indicating that resistant tumors are in an intermediate EMT state (Data Supplement).²³

Lack of Clonal Change in Nonresponders Suggests Intrinsic Resistance

To investigate if this cell plasticity was related to the emergence of new driver lesions at the DNA level, we performed whole-genome sequencing calling single-nucleotide variants (SNVs), somatic copy number aberrations (SCNAs), and structural variants on paired pre- and post-treatment tumor samples (Figs 3A and 3B; Data Supplement). Successful sequencing was obtained from all residual tumors (n = 15, as one patient had no residual tumor), but only from 7 of 16 diagnostic specimens because of limited tumor tissue input and formalin-fixed paraffin-embedded (FFPE) induced DNA degradation. ETS fusions were identified in 5 of 15 patients, with the remaining patients falling into the Other category in The Cancer Genome Atlas.²⁴ Although we did observe enrichment of specific genomic drivers with treatment in individual patients (for example, an RB1 loss of heterozygosity event in NAT02), there was no consistent association between known drivers and tumor response, and in general, post-treatment samples had fewer detected somatic aberrations. Given that one potential mechanism of drug resistance is the clonal expansion of a distinct cell population intrinsically resistant to ADT,²⁵ we performed clonal analysis with the PyClone algorithm²⁶ using SNV and SCNA data across longitudinal patient samples (Figs 3C and 3D). We also performed an orthogonal clonal analysis by comparing SCNA segments in sample pairs. Of the seven patients with paired pre- and post-treatment WGS data, clonality was resolvable in four. To test for potential SNV artifacts in the FFPE samples, which may affect clonal analysis, we compared base pair transitions at different allele frequencies between matched fresh-frozen and FFPE samples. We observed no significant C>T/G>A enrichment in the 1%-10% or 10%-25% VAF ranges (P = .3352 and P = .7985, respectively, using a one-sided paired t test; Data Supplement). In the majority of patients, although evidence of subclones was identified in both pre- and post-treatment specimens, no significant changes in clonal makeup were identified in resistant tumors in patients NAT09 and NAT03 (Figs 3Di-3Diii and Data Supplement). NAT13 (partial responder) showed some evidence of a shrinking clone in the SCNA data (Data Supplement); however, this was not supported in the SNV analysis. In particular, no new genomic drivers were observed, indicating that tumor cell populations present before treatment were intrinsically resistant to ADT or had the intrinsic ability to adapt to the castrate state. In contrast, in one patient who demonstrated an almost complete pathologic response, we identified significant changes in the subclonal makeup, with evidence for the enrichment of a treatment-resistant subclone in both the SNV and SCNA data (Fig 3Div and Data Supplement).

Pretreatment SNAI2 Copy Number Status Correlates With Tumor Response

Given that SCNA events are more important in driving localized prostate cancer progression than SNVs and that copy number loss is more frequently observed than copy number gain,²⁴ we reasoned that a copy number loss event would be the most likely somatic aberration defining treatment sensitivity. We therefore performed an overlap analysis of copy number loss events present subclonally in pretreatment specimens but not detectable in residual persistent tumors, focusing on the three patients with the best response (NAT2, NAT3, and NAT12) for whom paired pre- and post-treatment WGS data were available, identifying 45 candidate genes (Fig 4A). We reasoned that if loss of expression was important for treatment sensitivity, then overexpression might drive treatment resistance. We checked the expression levels of these 45 genes in matching RNA-seq data and found that only two were significantly overexpressed (logfold change > 1; FDR < 0.05) at reasonable transcript abundance (log counts per million > 3) (Fig 4B) and had consistent expression across the different RNA-seq data sets (Fig 4C). Unexpectedly, one of the prioritized genes was SNAI2, which encodes the transcription factor Slug, a key regulator of EMT that directly represses E-cadherin expression²⁷ and is expressed in residual tumor epithelium (n = 13) (Fig 4D and Data Supplement). Using fluorescent in situ hybridization (FISH) to directly measure gene copy number at the cellular level (Fig 4E), we confirmed that treatment with profound ADT resulted in the loss of cells deficient for SNAI2 (Figs 4F and 4G) and that the prevalence of SNAI2-deficient cells in pretreatment biopsy specimens correlated with tumor response (Fig 4H).