Transcriptionally Active Androgen Receptor Splice Variants Promote Hepatocellular Carcinoma Progression

Hepatocellular carcinoma (HCC) is a leading cause of cancer mortality globally and represents the most prevalent form of primary liver cancer. Chronic viral hepatitis—Hepatitis B virus and hepatitis C virus (HCV), nonalcoholic steatohepatitis, and heavy alcohol intake are major etiologic contributors to hepatocarcinogenesis (1). HCC incidence and mortality rates are rapidly outpacing other malignancies in the United States, highlighting a substantial unmet medical need (2). Although early diagnosed tumors can be treated with liver resection, transplantation, and ablation, HCC is a profoundly invasive tumor and challenging to diagnose at early stages, bringing about restricted treatment choices and low survival rates. The multityrosine kinase inhibitors sorafenib and lenvatinib are the only approved first-line systemic therapy for advanced HCC, yet they have limited efficacy and severe side effects (3). Recently, regorafenib, cabozantinib, nivolumab, and ramucirumab were approved as second-line therapy in patients previously treated with sorafenib (4). The availability of molecular targeted therapy and immune checkpoint inhibitors is a welcome step toward improving systemic therapy for HCC. However, this will bring the challenge of patients' stratification and drug selection.

Hepatic carcinogenesis is a complex process characterized by dysregulation of both genetic and epigenetic signaling cascades occurring in the context of significant genetic heterogeneity (5). Despite this, HCC is a consistently sexually dimorphic cancer with men having 2- to 4-fold higher likelihood of developing HCC compared with women. Several reports have implicated androgen receptor (AR) in HCC to explain, in part, sexual dimorphism in HCC (6). The AR, a ligand-activated transcription factor, canonically exerts various physiologic and pathologic functions in response to binding androgens (7). AR's effects in prostate tissue are best characterized and include prostate cancer initiation and progression (7). In addition, several reports have demonstrated increased levels of AR expression in tumor versus adjacent nontumor tissue in HCC (8, 9). The AR has been shown to modulate cellular oxidative stress and DNA damage repairing systems leading to genomic instability (10) and high levels of AR expression are correlated with high metastatic potential in vitro in HCC cell models (11). In spite of compelling evidence supporting the role of AR signaling in HCC initiation and progression, no survival advantage was found following antiandrogen therapy in patients with HCC (12, 13).

Notably, strong recent evidence of AR axis activation in advanced HCC was provided in a report by Zhang and colleagues that found nuclear and likely active AR, as opposed to total cellular AR, was an independent predictor of overall survival in patients with HCC (9). This work showed that AR antagonism in HCC cells was associated with rebound activation of the AKT–mTOR pathway that further contributed to nuclear localized AR protein and AR transcriptional activity. In an effort to reconcile the failure of AR-targeted therapy in HCC with the continued evidence of AR activation in primary disease, we evaluated therapeutic resistance mechanisms well-known to the prostate cancer field. In prostate cancer, persistent activation of the AR axis in the presence of once effective AR-targeted therapy is intensively studied (14). The failure of AR-targeted therapy in prostate cancer has been explained by multiple mechanisms including but not limited to AR amplification, AR point mutations altering steroid and antiandrogen binding, and expression of alternative AR splice variants that lack the C-terminal ligand-binding domain, but retain transcriptional capacity (15). In each case, treatment-refractory, ligand-independent, and constitutively-active AR signaling can result. In this study, we report the expression of AR and its truncated splice variants in HCC primary patient samples and cell lines. Our data suggest the expression of truncated AR splice variants in HCC can mediate constitutive AR signaling and these AR spilce variants (AR-SV) are capable of driving AR signaling in the presence of antiandrogen therapy. Moreover, we characterize both androgen-mediated and androgen-independent AR transcriptional activity in diverse HCC models. Finally, we expand the understanding of AR's biological contribution in HCC progression by demonstrating its modulation of epithelial-to-mesenchymal transition (EMT) effector proteins.

Human HCC and prostate cancer cell lines HepG2, PLC/PRF/5, SNU-423, VCaP, 22Rv1, DU145, and the immortalized normal liver cell line THLE2 were obtained from ATCC. For reproducibility experiments, HCC cells' panel (HepG2/C3A, PLC/PRF/5, SNU-423, SNU-475) was obtained from ATCC (ATCC TCP-1011). The read pairs covering exon 4–8 of the AR locus from SNU-475 cell whole genome sequencing (WGS) are summarized in Supplementary Fig. S1A (details in Supplementary Methods). HCCLM3 cells were kindly provided by Thomas Schmittgen, University of Florida (Gainesville, FL). Cell lines have been tested and authenticated by Genetica Cell Line Testing using short tandem repeat DNA profiling. All cells were tested for Mycoplasma periodically using MycoAlert PLUS Mycoplasma Detection Kit (LT07-705, Lonza) and were negative. Cells were subcultured in ATCC recommended medium, DMEM or RPMI1640 (11995-065, 11875-093, Invitrogen), supplemented with 10% FBS (F0926-500ML, Sigma-Aldrich) at 37°C in a humidified atmosphere with 5% CO₂. Phenol red–free media containing 5% charcoal-stripped fetal bovine serum (csFBS) were used for studies involving steroid hormone signaling (21063-029, 11835-030, Invitrogen).

Freshly frozen primary HCC samples were acquired from two independent cohorts comprising 32 patients with HCC who underwent surgical resection. In cohort 1, HCC tumor specimens were acquired from 16 patients (12 tumor only samples used in RT-PCR analyses and 4 tumor and adjacent normal samples used in Western blot analyses) who were diagnosed with HCV-mediated HCC at The James Cancer Center – The Ohio State University (Columbus, OH) and collected through the Total Cancer Care (TCC) Protocol. In cohort 2, total RNA samples were collected from 16 patients who were diagnosed with HCV-mediated HCC at National Liver Institute, Menoufia University (Menoufia, Egypt). Both studies were conducted according to the guidelines of the Declaration of Helsinki and approved by local review boards, and written informed consent was obtained from all patients (IRB approval no: 2016E0496). Patient characteristics are provided in Supplementary Table S1.

We searched 1,019 HCC primary samples from six studies in cBioPortal database (16) for AR gene alterations potentially capable of explaining antiandrogen resistance in HCC. Genomic AR alterations were excluded as a potential cause as we determined somatic mutation frequency to be less than 1.5% in all cases, and there were no copy number variations detectable within the AR gene in these cohorts (Supplementary Fig. S1B and S1C). We next considered nongenomic potential mechanisms of resistance, such as the expression of C-terminal–truncated AR-SVs that result from alternative splicing of AR pre-mRNA. Several constitutively active, C-terminal–truncated splice variants capable of mediating resistance to antiandrogen therapy have been described in prostate cancer cells and patients (Supplementary Fig. S1D; ref. 17). In prostate cancer, approximately 20 AR-SVs have been described, and among these, variant 7 (AR-v7) is best characterized and thought to be the most clinically relevant. To the best of our knowledge, the expression and biological relevance of AR-SVs in HCC had not been previously examined in detail. To explore AR-SV expression in primary liver cancer, we surveyed HCC patient data in The Cancer Genome Atlas (TCGA) for noncanonical AR mRNA expression by reanalyzing raw RNA-seq data. To our surprise, AR-SV expression was observed in 290 of 372 (78%) of patients. Notably AR-SVs represented more than one quarter of total AR transcripts in the 100 patients with the highest total levels of AR message (Fig. 1A, left). AR-SV expression was detected in both male and female patients with overall higher levels in males (Fig. 1B, right; Supplementary Fig. S2A), but no obvious expression differences exist when the TCGA data are considered by ethnicity (Supplementary Fig. S2B) or tumor stage (Supplementary Fig. S2C). Similar to prostate cancer (Supplementary Fig. S2D), a well-characterized AR-dependent cancer, AR-total expression in the TCGA HCC cohort was not associated with effects on overall survival (Supplementary Fig. S2E). We also found no overt effects of HCC AR-SV expression on overall survival. We then hypothesized that AR-SV–mediated AR signaling in HCC could explain, in part, resistance to antiandrogen therapy in patients with HCC.

To further evaluate HCC patient AR-SV expression, we designed isoform-specific primers that anneal to a unique exon–exon junction in each previously characterized AR isoform (Supplementary Fig. S3A) and validated them using well-characterized AR-positive and AR-SV–positive prostate cancer cell lines as positive controls (Supplementary Fig. S3B and S3C). We next evaluated AR mRNA expression in total RNA from 12 primary HCV-mediated HCC samples acquired through the James Cancer Center at The Ohio State University. We quantified the expression of AR-FL and v7 using robust, specific RT-PCR assays (Supplementary Fig. S4A and S4B). VCaP and THLE2, immortalized normal human liver cells, were used as positive and negative controls, respectively. Our results showed heterogeneity in total AR expression among patients but, surprisingly, the expression of AR-FL and v7 in some patients approached the levels of AR-FL and v7 in VCaP, an AR-positive prostate cancer cell line that is known to express abundant AR relative to other prostate cancer cell lines (Fig. 1B; ref. 18). In the second cohort, we evaluated 16 primary HCV-mediated HCC samples obtained from the National Liver Institute in Menoufiya University (Menoufiya, Egypt) using a more variable reference gene (β-actin, Supplementary Fig. S4C), which resulted in similar AR-FL and v7 expression levels as compared with the first cohort (Supplementary Fig. S4D). These findings substantiated our in silico analyses demonstrating AR-SV expression in a significant subset of patients with HCC and further revealed that in addition to AR-FL, the C-terminally truncated AR-v7 mRNA is also abundantly expressed.

To extend our findings in primary HCCs to more readily studied HCC cell lines, we utilized Cancer Cell Line Encyclopedia (CCLE) mRNA data to quantify protein-coding AR transcripts in multiple immortalized HCC cell models (19). Specifically, AR transcript data from 18 HCC and 2 prostate cancer cell lines were evaluated (Fig. 1C). In our analysis, HCC cell lines were considered AR-positive or AR-negative based on CCLE-determined transcript abundance, our own Western blots, and published data (11). Although more than 20 AR isoforms have been described, we focused our in silico and subsequent analyses on experimentally confirmed or potentially constitutively active isoforms (Supplementary Table S2). We next determined the expression of AR-FL and well-described AR variants in prostate cancer namely, AR-v7, v1, v3, v4, and v12, in an initial panel of HCC cell lines using RT-PCR. These results showed that both AR-FL and AR-v7 expression were elevated in HCC cell lines with mesenchymal morphology derived from advanced disease (HCCLM3 and SNU-423) compared with morphologically epithelial HCC cell lines representing early disease (HepG2/C3A, and PLC/PRF/5) and the immortalized normal liver cell line (THLE2; Fig. 1D; Supplementary Fig. S5A). Furthermore, despite HCCLM3 cells demonstrating the most abundant AR-FL and AR-v7 expression among HCC cells, primary HCC samples exhibited much greater AR-FL and AR-v7 transcript on average (Fig. 1E). Similar to AR-v7, AR-v1 and v3 were more abundant in HCC cells relative to normal liver cells, but AR-v12 and v4 were undetectable in evaluated HCC cell lines.

Next, we evaluated AR protein expression in HCC cell lines utilizing two antibodies, an antibody against an N-terminal AR epitope that detects both AR-FL and C-terminal–truncated SVs (AR-NT) and an antibody against a C-terminal AR epitope that detects only AR isoforms containing the C-terminal portion of the protein, but not C-terminal–truncated splice variants (AR-CT). Consistent with our mRNA data, we detected both AR-FL and lower molecular weight (MW) splice variants in the AR-positive HCC cell line, HCCLM3, when an AR N-terminus directed reagent was used (Fig. 1F). However, the low MW species in AR-positive HCCLM3, and prostate cancer cell line, VCaP, were undetectable when using the AR C-terminal antibody (Fig. 1G). To verify our approach to detecting C-terminal–truncated AR-isoforms, we transfected the AR-negative HCC cell, PLC/PRF/5 with either an AR-FL or AR-v7–expressing plasmid (PLC_pARFL and PLC_pAR-v7, respectively). Prostate cancer models, VCaP and DU145, were included as AR-positive and AR-negative controls, respectively. As expected, the AR N-terminus–directed antibody detected both AR-FL in PLC/PRF/5–overexpressing AR-FL and AR-v7 in PLC/PRF/5–overexpressing AR-v7, whereas the C-terminus–directed antibody detected only AR-FL in PLC/PRF/5–overexpressing AR-FL. This was also the case when the Western blot was first performed with the C-terminal–targeted antibody and then stripped and reblotted with the N-terminal–targeting reagent (Supplementary Fig. S5B). We confirmed AR-SV mRNA levels with fresh, lower passage cells acquired as part of an HCC cell panel available from ATCC (Fig. 1H; Supplementary Fig. S5C). This analysis revealed the AR-SV–expressing HCC cell line SNU-475 that had abundant AR-v7 expression, but undetectable levels of AR-FL mRNA. Unlike HCCLM3, SNU-475 cells produce an AR species detectable with an AR-v7–specific antibody similar to AR-v7–expressing prostate cancer cells (VCaP and 22Rv1) and transfected HCC controls (Fig. 1I). Evaluation with an N-terminal AR-targeting antibody showed SNU-475 to exclusively express ARv7 protein (Fig. 1J). To better understand the origins of AR-v7 expression in the absence of AR-FL message or protein in this model, we performed WGS on SNU-475 cells (Supplementary Table S3). This experiment revealed a large 46-kb deletion in the AR locus, consistent with the loss of C-terminal AR exons, but retention of the N-terminal and cryptic exons required for AR-v7 (Fig. 1K). In total, these findings confirm the expression of C-terminal–truncated AR-SVs in a subset of HCC tumors.

In the absence of endogenous androgen (i.e., testosterone, 5α-dihydrotestosterone), the AR's location is predominantly cytoplasmic. Upon ligand binding, the AR translocates to the nucleus, interacts with androgen-responsive DNA elements, and recruits the transcriptional apparatus, resulting in AR-target gene modulation (7). Breaking from canonical AR action, C-terminal–truncated AR-SVs are found predominately in the nucleus, even in the absence of ligand (20). To determine AR subcellular localization and its trafficking in response to ligand, we compared nuclear and cytoplasmic AR abundance in SNU-423 and HCCLM3 HCC cell lines using Western blot analysis and immunofluorescence. Cell fractionation showed that under hormone-depleted conditions, AR expression in SNU-423 was mainly cytoplasmic but became nuclear upon presentation with a potent androgen agonist (R1881), whereas in hormone-depleted conditions, HCCLM3 AR was predominately nuclear (Fig. 2A).

Consistent with Western blot analysis results, immunofluorescence microscopy showed that AR staining was primarily cytoplasmic in untreated, hormone-depleted SNU-423 cells. Likewise, following stimulation with a potent androgen agonist, a strong fluorescence staining of AR protein occurred in the nuclei (Fig. 2B). In contrast, in hormone-depleted untreated HCCLM3 cells, AR protein staining was predominately nuclear irrespective of androgen treatment. Notably, treatment with the antiandrogen enzalutamide did not substantially alter AR nuclear localization in HCCLM3 cells (Fig. 2B). Subcellular location of HCCLM3 AR closely mimicked what we observed in VCaP, an AR-SV–positive prostate cancer cell line utilized as a positive control (Supplementary Fig. S5D). As in our Western blots (Fig. 1F and G), to distinguish between AR-FL and AR-SV subcellular localization, we used a C-terminal–targeted AR antibody. In hormone-depleted HCCLM3, C-terminal–reactive AR was mainly cytoplasmic, whereas when cells were treated with androgen, the AR became predominately nuclear, suggesting some amount of AR in HCCLM3 cells remains androgen responsive (Fig. 2C). To confirm the default nuclear localization of C-terminal–truncated splice variants in HCC cells, we overexpressed AR-v7 in the AR-positive HCC cell line, SNU-423. As expected, in the absence of ligand, N-terminal–reactive AR is predominantly nuclear in AR-v7–overexpressing SNU-423 relative to SNU-423 vector-only control cells (Fig. 2D). Notably, C-terminal–reactive AR species in SNU-423 were also predominantly nuclear in the presence of exogenous AR-v7. As with other AR-SV–positive HCC cells in the basal state, the AR in SNU-475 was predominantly nuclear and no AR was detectable with a C-terminal–targeting reagent providing additional evidence that these cells lack AR-FL (Fig. 2E). Consistent with our detection of AR-SV mRNA expression in primary HCCs (Fig. 1B) and the predominantly nuclear localization of AR-SVs in HCC cell lines, we detected low MW AR species in limited nuclear fractions prepared from primary HCC tissues as well (Fig. 2F). Together, these data support the constitutive activity of AR-SVs detected in primary HCC samples and HCC cells.

To evaluate the transcriptional activity of ligand-responsive AR-FL in HCC, we transfected HCC cells with either a noninducible basal promoter (pGL4.24) or an AR-responsive firefly luciferase vector (MMTV-LUC). Transfected cells were then treated with vehicle, R1881, or a combination of R1881 and antagonist (enzalutamide) with VCaP and DU145 serving as AR-positive and AR-negative controls, respectively. In cells transfected with the AR-responsive vector, there was a significant androgen-mediated, antagonist-reversible induction in SNU-423 (AR⁺) and VCaP (AR⁺) transcriptional signal, whereas AR-negative cell lines, DU145 and HepG2, did not respond to androgen. Notably, similar to AR-negative control cell lines, HCCLM3 (AR⁺) HCC cells did not show significant ligand-inducible transcriptional activity (Fig. 3A), despite being readily transfected according to internal transfection controls (Supplementary Fig. S5E). Under identical conditions, cells transfected with a noninducible basal vector did not respond to any treatment (Supplementary Fig. S6A).

To assess potential ligand-independent AR transcriptional activity in the same models, ligand-independent, basal luciferase activity was determined by comparing luciferase signal between vehicle-treated pGL4.24 control and MMTV-LUC–transfected cells. Consistent with constitutive transcriptional activity, we detected a 15-fold induction in luciferase signal in MMTV-LUC–transfected HCCLM3 and VCaP cells relative to corresponding pGL4.24 controls (Fig. 3B). This induction was significantly reversed with multiple siRNAs targeting the AR N-terminus that resulted in both AR-FL and AR-SV knockdown and suppressed the MMTV-LUC signal supporting AR's contribution to this ligand-independent transcriptional effect (Fig. 3B, inset; Supplementary Fig. S6B). In contrast, using the same analysis, the androgen-responsive HCC cell line SNU-423 demonstrated no constitutive transcriptional activity that further mirrored results from AR-negative DU145 and HepG2 cells. Instead, SNU-423′s MMTV-LUC activity was both ligand (Fig. 3B) and receptor dependent (Supplementary Fig. S6C). As with other AR-SV–positive cells, SNU-475 demonstrated strong constitutive transcriptional activity sensitive to AR knockdown (Fig. 3C).

To further interrogate AR transcriptional activity in HCCLM3 and VCaP cells, we attempted to modulate constitutive transcriptional activity with enzalutamide, which is not expected to impact C-terminal–truncated AR-SV–mediated transcription (Fig. 3D). This experiment was performed both alone or in the presence of siRNA targeting all AR species (as in Fig. 3C) revealing the relative inability of enzalutamide treatment to suppress AR transcriptional activity as compared with AR knockdown in HCCLM3 and VCaP cells.

To further demonstrate the ability of AR-SVs to drive constitutive AR activity in HCC cells, we recapitulated a ligand-independent MMTV-LUC signal in AR-SV–negative SNU-423 cells by introducing an AR-v7 expression vector. As in HCCLM3 cells, the presence of AR-SV in SNU-423 cells boosted basal transcriptional activation (Fig. 3E) and muted response to R1881 (Fig. 3F). To demonstrate that AR-FL is not required for constitutive activation of the MMTV-LUC reporter in HCC cells, the AR⁻ HCC cell line HepG2/C3A was similarly transfected with AR-v7, resulting in robust activation (Fig. 3G). Given the apparent movement of cytoplasmic AR-FL species in HCCLM3 to the nucleus following treatment with R1881 (Fig. 2C), we also tested the ability of R1881 to further increase the already high, constitutive MMTV-LUC signal in these cells (Fig. 3B). No additional stimulation was apparent (Supplementary Fig. S6D). When considered together, these results demonstrate ligand-independent and ligand-dependent AR-mediated transcription in HCC cells.

To assess the potential biological contribution of AR signaling in HCC, we performed cell proliferation assays over 24 hours comparing vehicle, R1881, or enzalutamide and R1881 in androgen-responsive SNU-423 cells. Short-term androgen treatment did not increase cell proliferation (Fig. 4A) or increase colony numbers (Fig. 4B). As no effects on proliferation were apparent, we further assessed the potential role of AR in HCC cell migration and invasion given that androgens and the AR have been reported to promote cell invasion and metastasis in HCC (11). Under similar conditions, R1881 treatment significantly increased SNU-423 and, surprisingly, HCCLM3 cell migration relative to vehicle in an effect that was reversible by cotreatment with enzalutamide (Fig. 4C). To further evaluate the contribution of AR to HCC cell migration, HCCLM3 cells were transfected with either siRNA targeting AR-FL and AR-SV species (as in Fig. 3C) or control siRNA. Basal HCCLM3 cell migration was significantly suppressed following AR knockdown consistent with the contribution of constitutively active AR signaling to HCCLM3 migration (Fig. 4D). Next, we investigated the role of AR in invasion in HCC using the Matrigel invasion assay, in which SNU-423 demonstrated androgen-dependent (Fig. 4E) and AR-dependent effects (Fig. 4F) but, unlike migration, HCCLM3 showed no difference in invasion when cells were treated with R1881 (Fig. 4G). However, matching migration, treatment with siRNA targeting AR-FL and AR-SVs reduced basal HCCLM3 cell invasion (Fig. 4H). Likewise, multiple siRNAs knocking down AR-v7 in SNU-475 cells significantly suppressed invasion (Fig. 4I). In addition, overexpression of AR-v7 in AR⁻ HepG2/C3A cells resulted in elevated invasion, despite the limited invasive capacity of these epithelial cells (Supplementary Fig. S7A). Taken together, these data suggest that both ligand-independent and ligand-dependent AR signaling are capable of modulating HCC cell migration and invasion.

In the context of prostate cancer, the AR has been shown to play a crucial role in cell migration and invasion through modulating expression of EMT effector proteins (21). However, the precise role that AR plays in HCC development remains poorly understood. To further elucidate how AR signaling might contribute to the EMT in HCC, we analyzed whole transcriptome data in 24 HCC cell lines from the CCLE database. Interestingly, multidimensional scaling of entire transcriptomes successfully separated HCC cell lines according to AR expression (positive or negative, as characterized above in Fig. 1C), supporting the biological relevance of AR-dependent transcriptional programs in AR⁺ HCC cells (Supplementary Fig. S8A). A differential gene expression analysis of AR⁺ relative to AR⁻ HCC cells returned 1,058 differentially expressed genes (Supplementary Table S4) that were further evaluated for potential biological function by gene set enrichment analysis (GSEA) using the molecular signature database (MSigDB; ref. 22). GSEA revealed significant enrichment of gene sets involved in the EMT pathway, including extracellular matrix and cell adhesion genes, among HCC AR⁺ cells (Fig. 5A).

To identify candidate direct AR target genes, the promoter regions of significantly enriched genes within the EMT pathway were searched for consensus androgen-responsive elements. Of the 8 EMT genes evaluated, we determined only SNAI2 had a strong putative ARE in its proximal promoter region (Supplementary Fig. S8B). Subsequent analyses of CCLE RNA-seq data revealed a moderate positive correlation between SNAI2 and AR expression in AR⁺ cell lines as compared with AR⁻ cell lines further supporting putative AR:SNAI2 regulatory interactions in AR-dependent HCC migration and invasion (Fig. 5B). In agreement with our in silico findings, we determined SNAI2 mRNA levels were transiently induced by androgen treatment in androgen-responsive SNU-423 cells (Fig. 5C) and androgen-dependent effects on SLUG protein levels (the product of SNAI2 mRNA) were more pronounced with nearly 4-fold induction apparent 24 hours after androgen stimulation (Fig. 5D). Consistent with consequential slug induction, CDH1 (E-cadherin) mRNA was also responsive to short-term androgen treatment, whereas other common EMT regulators were not (Supplementary Fig. S8C and S8D, respectively). Like AR, SLUG is a transcription factor that is nuclear localized when activated (23). Congruent with androgen-dependent SLUG activation, 24-hour androgen treatment in SNU-423 cells results in the relocalization of SLUG and AR protein to the nucleus in an effect partially reversible by enzalutamide coadministration (Fig. 5E).

The role of AR-dependent SNAI2 regulation in AR-mediated HCC cell invasion was further interrogated by AR or SNAI2 siRNA knockdown in SNU-423 cells. Androgen treatment promoted SNU-423 invasion (as in Fig. 4D) that was prevented by AR or SNAI2 knockdown (Fig. 5F). Notably, AR knockdown in either SNU-423 (Fig. 5F, inset) or HCCLM3 cells (Supplementary Fig. S8E) resulted in reduced levels of Slug and the EMT protein ZEB-1 (24). However, whereas AR and SNAI2 knockdown in SNU-423 resulted in an expected rebound in E-cadherin protein levels (Supplementary Fig. S8F), E-cadherin levels were not impacted by AR status in HCCLM3 (Supplementary Fig. S8E).

To evaluate the potential ability of ligand-independent AR-SV activity to modulate HCC cell invasion, we overexpressed AR-v7 in SNU-423, cells resulting in increased invasion (Fig. 5G) and primarily nuclear AR staining (Fig. 5H). Exogenous AR-v7 expression in SNU-423 cells also elevated SLUG protein levels and promoted SLUG nuclear translocation relative to controls. Constitutively active, nuclear localized AR-SVs in HCCLM3 also colocalized with nuclear slug (Fig. 5I). As with other AR-SV–positive HCC cells, siRNA-mediated knockdown of AR-v7 in SNU-475 cells resulted in suppression of slug protein (Supplementary Fig. S8G).

To verify the clinical relevance of AR-driven SNAI2 mRNA expression, we evaluated both TCGA and GTEx liver tissue expression data (25). Similar to the relationship in HCC cell lines, our analyses revealed a statistically significant positive correlation between AR and SNAI2 in primary HCC samples (TCGA, 372 patients). Notably, we failed to detect a similar correlation between AR and SNAI2 in matched normal samples (TCGA, 50 patients) and normal liver tissues (GTEx, 150 donors), suggesting AR-mediated slug activation is liver cancer specific (Fig. 5J). Taken together, these data suggest that both ligand-independent and ligand-dependent AR regulation of SNAI2/SLUG can modulate HCC cell EMT signaling.

Given the findings of Zhang and colleagues (9) that demonstrated antagonism of AR in HCC results in activation of the AKT–mTOR pathway we evaluated the role of AR-SV signaling in this cross talk. As shown by Zhang and colleagues, AR knockdown in SNU-423 cells resulted in activation of both p-AKT and p-mTOR (Fig. 6A) but agonist stimulation alone or in combination with ENZ had no effect on either protein (Fig. 6B). Interestingly, ectopic expression of AR-v7 resulted in p-AKT and p-mTOR induction in AR⁺ SNU-423, but not AR⁻ HepG2/C3A cells (Fig. 6C). Basal p-AKT and p-mTOR levels were also suppressed by knockdown of AR-v7 in SNU-475 cells (Fig. 6D). Taken together, these data suggest that, similar to AR antagonism with enzalutamide alone, AR-SV expression in mesenchymal-like HCC cells supports AKT–mTOR pathway activation.

In an attempt to reconcile the discordant observations of the partial AR dependency of HCC development (10, 26) and the total failure of therapies targeting the AR axis in patients with HCC (12, 13), we thoroughly evaluated the AR locus using publicly available primary HCC and HCC cell line data. We found no widespread genomic mutations or amplifications capable of explaining therapeutic resistance (Supplementary Fig. S1) but, we uncovered robust expression of AR-SVs, which could serve as potential mediators of persistent AR signaling in the presence of AR-targeted therapy (Fig. 1A). We confirmed AR-SV expression in two cohorts of primary HCCs (Fig. 1B; Supplementary Fig. S4) and HCC cell lines (Fig. 1D and H), and further demonstrated that C-terminal–truncated AR mRNA species are translated into C-terminal–truncated proteins (Fig. 1F and G) including AR-v7 (Fig. 1I and J) in HCC. Using multiple lines of evidence, we demonstrated AR-SV function in HCC cells is consistent with treatment-refractory, constitutive activity including default nuclear localization (Fig. 2A, B, and E), ligand-independent transcriptional activation (Fig. 3B and C), and relative insensitivity to antiandrogen treatment (Figs. 2B and 3D and F). To our knowledge, we are the first to report a detailed characterization of AR-SVs in HCC.

Depending on the experimental context, C-terminal–truncated AR isoforms are indistinguishable from AR-FL. This creates the potential for AR-SVs to have remained effectively “hidden” in previous studies of AR signaling in HCC and likely contributes to the persistent controversy surrounding the role of the AR in HCC (27, 28). Similar to AR-SVs studied in prostate cancer (29), we found multiple low MW AR species that reside in the nucleus (Fig. 2A, B, E, and F). Nevertheless, our findings are consistent with Zhang and colleagues (9) who reported the overexpression of nuclear AR in patients with HCC and associated nuclear AR expression with tumor progression and poor prognosis. Zhang and colleagues' (9) IHC was performed with the same N-terminal–targeted reagent we utilized and is therefore incapable of distinguishing nuclear C-terminal–truncated AR-SVs from AR-FL in this large HCC cohort. Intriguingly, enzalutamide treatment in prostate cancer cells is associated with both increased AR-FL and AR-SV expression (18). In this sense, unrecognized AR-SV signaling in Zhang and colleagues' work may contribute to the reported feedback activation of AKT–mTOR signaling and associated resistance to antiandrogen treatment. Irrespective of enzalutamide treatment, we found both endogenously expressed AR-v7 (Fig. 6D) and ectopically expressed AR-v7 (Fig. 6C) to activate AKT–mTOR signaling in HCC cells. Our data support further study of how AR-SVs interact with the AKT–mTOR pathway.

AR-SV expression notwithstanding, there exists considerable disagreement in recent literature concerning the expression of AR in common HCC cell models (9, 11, 30, 31). Our survey of AR expression in HCC cells (Fig. 1C, D, and H) is in strong agreement with Ao and colleagues (11) and Zhang and colleagues (9) but differs from the AR expression in Huh7 and PLC/PRF/5 cells studied by both Feng and colleagues (31) and Yu and colleagues (30). In further agreement with Ao and colleagues(11), we found AR⁺ HCC cell migration and invasion to be AR dependent (Figs. 4D, F, H, I and 5F, respectively). However, differing from Zhang and colleagues (9), we were unable to show predominantly nuclear AR in the absence of androgens in SNU-423 cells (Figs. 2B and 5E) or clear androgen-dependent HCC cell proliferation (Fig. 4A). In each case, our disparate results may be due to our preincubation of cells in hormone-depleted, csFBS to minimize basal hormone signaling prior to visualization or treatment. One notable finding from our characterization of AR expression in HCC cells was the discovery that SNU-475 cells express only AR-v7 with no detectable AR-FL expression (Fig. 1H–J). WGS revealed a large genomic deletion in the AR locus in these cells that we speculate causes exclusive use of the previously described alternative poly-A site within intron 3 to make AR-v7 (32). To our knowledge, this deletion is a novel source of AR-v7 and the first description of a human cancer cell line expressing solely constitutively active variant AR protein. These features may make SNU-475 cells uniquely useful in the study of AR-SV signaling.

Building on these in vitro observations, our in silico informatics approach revealed a strong EMT transcriptional signature in AR⁺ versus AR⁻ HCC cell lines (Fig. 5A and B) and identified SNAI2 as a putative novel direct AR target in HCC (Fig. 5C–F). SNU-423 cell invasion was both AR and SNAI2 dependent (Fig. 5F) and androgen stimulation of SNU-423 cells resulted in both increased SLUG expression (Fig. 5D) and nuclear localization (Fig. 5E). Furthermore, in both HCCLM3 and SNU-475 AR-SV–positive cells, AR knockdown was associated with reduced invasion (Fig. 4H and I) and SLUG protein expression (Supplementary Fig. S8E and S8G). This pattern of SLUG regulation is similar to reports in prostate cancer where AR:SLUG interactions have been shown to promote the androgen-independent growth of castration-resistant disease (23). The positive relationship between AR and SNAI2 expression was substantiated in primary HCCs (Fig. 5J) consistent with the previously reported highly invasive character of AR⁺ HCC (9, 33). Following resection, Nagasue and colleagues reported the 5-year survival of recurrence-free HCC patients to be 55% and 0% for AR⁻ and AR⁺ HCC, respectively. The identification of the AR as a therapeutically targetable driver of EMT in HCC has potentially broad implications for AR-targeted therapy in the adjuvant setting where systemic therapy has failed to demonstrate clinical benefit (34).

Clinically relevant AR-SV expression in prostate cancer is thought to occur in the context of androgen ablative and/or antiandrogen therapy (35, 36). Although AR-SV's role as a driver of therapeutic resistance in prostate cancer remains unresolved (17), multiple preclinical studies demonstrate AR-SV–mediated resistance to antiandrogen treatment (37, 38) and clinical studies associate AR-SV expression with resistance to hormone therapy (39, 40). The activation of AR-SV signaling in response to selective pressures of AR-targeted therapy is a plausible adaptive biological response by hormone-addicted prostate cancer cells but raises the critical question of what underlies abundant AR-SV expression in primary HCC (Figs. 1A and B and 2F). Unlike prostate cancer, hormone-ablative and antiandrogen therapy are not utilized in HCC treatment, suggesting AR-SVs in HCC are of disparate origins. One potential explanation is the high rates of hypogonadism, approaching 90% in men, associated in advanced liver disease (41). This well-known phenomenon suggests that “castrate” liver conditions could routinely accompany worsening liver disease, at least in men, and provide similar selective pressures as prostate cancer therapy in promoting AR-SV expression. One limitation of our study is that we have not yet confirmed the sequence of the predominant low MW AR species in HCCLM3 or primary HCC samples. These data as well as future studies identifying drivers of AR-SV expression are key to better understanding the role of AR-SV expression in both diseased and normal liver.

We demonstrate AR-SV expression in primary cells that exceeds even our most abundant expressing HCC cell line (Fig. 1E), but it is still an open question as to what may constitute clinically relevant amounts of AR-SV expression. As has been generally shown in prostate cancer (17), AR-SV protein expression was typically accompanied by AR-FL (Fig. 1D and F), with SNU-475 being the notable exception (Fig. 1H–J). However, we demonstrate that the ligand-independent characteristics of AR-SV–positive HCCLM3 and SNU-475 cells can be recapitulated by adding modest amounts of exogenous AR-SVs to AR-SV–negative SNU-423 cells, suggesting the expression of small amounts of AR-SVs may act effectively like a ligand by inducing transcription (Fig. 3E), muting response additional AR activation (Fig. 3F), and driving residual AR-FL to the nucleus, resulting in the engagement of disease relevant AR signaling networks (Fig. 5H) including activation of the AKT–mTOR pathway (Fig. 6C and D). Our results raise the intriguing possibility that AR-dependent HCC signaling may be readily dissociated from the requirement of circulating androgens. These findings add greatly to the potential complexity of AR signaling in HCC and support previous claims that targeting the AR, as opposed to its ligand, may be the most effective approach in HCC (27). We acknowledge that, with the exception of our in silico approaches, we have evaluated a relatively small number of primary HCC samples and expansion to more primary samples is required for a broader interpretation of our findings. Nevertheless, we report several key features of AR-SVs in HCC worthy of further investigation.

Overall, our report is the first to describe the expression, biological relevance, and potential clinical relevance of AR-SVs in HCC. Our findings support the AR's role in promoting HCC migration and invasion via well-characterized EMT effector proteins and implicates AR signaling in HCC progression (Fig. 6E). We provide an additional explanation for the failure of traditional approaches to targeting the AR axis in HCC (i.e., hormone ablation and steroid-competitive antagonism) but, importantly, also highlight the potential of effective AR-targeted therapy in AR⁺ hepatocellular carcinoma. A growing number of novel therapies designed to target AR-SV–mediated signaling have been described with several currently being evaluated in patients with prostate cancer (42). Our data suggest similar targeting of the AR axis in AR⁺ HCC could result in new therapeutic strategies for patients with HCC with precious few treatment options.

J.S. Blachly is an advisory board member for AbbVie, AstraZeneca, and KITE Pharma. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A.M. Dauki, C.C. Coss

Development of methodology: A.M. Dauki, J.S. Blachly, C.C. Coss

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.S. Blachly, S. Ezzat, M.H. Abdel-Rahman

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.M. Dauki, J.S. Blachly, E.A. Kautto, M.H. Abdel-Rahman, C.C. Coss

Writing, review, and/or revision of the manuscript: A.M. Dauki, J.S. Blachly, E.A. Kautto, S. Ezzat, M.H. Abdel-Rahman, C.C. Coss

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.M. Dauki

Study supervision: J.S. Blachly, C.C. Coss

We thank Campus Microscopy & Imaging Facility for their assistance with immunofluorescence as well as Genomic Shared Resource in The Ohio State University Comprehensive Cancer Center (supported by the OSU CCC Core Grant, P30 CA016058; principal investigator: Dr. Raphael Pollock). We also thank the Molecular Carcinogenesis and Chemoprevention Program (OSU Comprehensive Cancer Center). We are grateful to Dr. Mitch Phelps for the valuable input and review of the manuscript and Mohamed Badawi, Apollinaire Ngankeu, Shelley Orwick, Hanna Radomska, and Riley Mullins (College of Pharmacy, The Ohio State University, Columbus, OH) for technical assistance in preparation of tissue samples for analysis. This work was funded in part by the American Cancer Society Grant IRG-67-003-50 (to C.C. Coss), National Cancer Institute Sustainable Sciences Institute (to M.H. Abdel-Rahman and S. Ezzat). This work was also supported in part by NCI/NIH K12-CA133250-07 (to C.C. Coss), American Foundation for Pharmaceutical Education Pre-Doctoral Fellowship (to A. Dauki), Pelotonia Idea Award (OSU Comprehensive Cancer Center; to C.C. Coss), and the National Institute of General Medical Sciences of the NIH under award number 5T32GM068412-12 (to E.A. Kautto). The results shown here are in part based upon data generated by the TCGA Research Network (http://cancergenome.nih.gov/). We acknowledge our use of the gene set enrichment analysis, GSEA software, and Molecular Signature Database (MSigDB). We would also like to express our sincere gratitude to the participants in TCGA and GTEx for donating cancer specimens and tissues. We also thank the reviewers for their insights and helpful suggestions that resulted in an improved manuscript.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.