Reciprocal YAP1 loss and INSM1 expression in neuroendocrine prostate cancer

Abstract

Neuroendocrine prostate cancer (NEPC) is a rare but aggressive histologic variant of prostate cancer that responds poorly to androgen deprivation therapy. Hybrid NEPC-adenocarcinoma (AdCa) tumors are common, often eluding accurate pathologic diagnosis and requiring ancillary markers for classification. We recently performed an outlier-based meta-analysis across a number of independent gene expression microarray datasets to identify novel markers that differentiate NEPC from AdCa, including up-regulation of Insulinoma-associated protein 1 (INSM1) and loss of Yes-associated protein 1 (YAP1). Here, using diverse cancer gene expression datasets, we show that Hippo pathway-related genes, including YAP1, are among the top down-regulated gene sets with expression of the neuroendocrine transcription factors, including INSM1. In prostate cancer cell lines, transgenic mouse models and human prostate tumor cohorts, we confirm that YAP1 RNA and YAP1 protein expression are silenced in NEPC and demonstrate that the inverse correlation of INSM1 and YAP1 expression helps to distinguish AdCa from NEPC. Mechanistically, we find that YAP1 loss in NEPC may help to maintain INSM1 expression in prostate cancer cell lines and we further demonstrate that YAP1 silencing likely occurs epigenetically, via CpG hypermethylation near its transcriptional start site. Taken together, these data nominate two additional markers to distinguish NEPC from AdCa and add to data from other tumor types suggesting that Hippo signaling is tightly reciprocally regulated with neuroendocrine transcription factor expression.

Introduction

Neuroendocrine prostate cancer (NEPC) is an aggressive variant of castration-resistant prostate cancer (CRPC). Primary NEPC is rare (0.2–1%), but treatment-related NEPC may be increasing in prevalence as patients develop resistance to increasingly potent hormonal therapies [1]. Pathologic diagnosis of NEPC can be straightforward when classic morphological features of small cell carcinoma are present, however there is increasing recognition of disease heterogeneity and hybrid phenotypes along a continuum with high grade prostatic adenocarcinoma (AdCa) [2]. This may be a natural outcome of the most common mechanism of NEPC development: in both therapy-induced as well as de novo disease, NEPC tumors likely develop via transdifferentiation from a common AdCa precursor [3,4]. Accordingly, transitional phenotypes between NEPC and AdCa are commonly encountered in clinical practice and pose a diagnostic dilemma.

We recently performed an outlier-based meta-analysis across a number of independent gene expression microarray datasets to identify novel markers that distinguish NEPC from AdCa [5]. We developed a 69 gene meta-signature of NEPC tumors (versus AdCa) using 6 microarray datasets. Interestingly, the gene neuroendocrine transcription factor (INSM1) was among the top genes up-regulated while the Hippo pathway oncogene YAP1 was among the top genes down-regulated in NEPC relative to AdCa. Notably, YAP1 and INSM1 have been reported to be reciprocally regulated in small cell lung cancer (SCLC) and together have been used for contemporary classification of this tumor type [6,7]. Herein, we examined INSM1 and YAP1 expression levels across numerous cancer gene expression datasets, as well as in diverse prostate cancer model systems and tumor cohorts and examined their regulation in prostate AdCa and NEPC cell lines in vitro.

Materials and methods

Cohort description.

Tumor samples were obtained prospectively following collection protocols approved by Johns Hopkins School of Medicine Institutional Review Board. We used previously described cohorts of primary small cell NEPC aggregated on tissue microarrays (TMA) and sampled in triplicate using 1 mm diameter tissue cores [5,8,9]. NEPC tumors were diagnosed using both morphology and immunohistochemistry on prostate tissue samples. The primary AdCa cases were selected from three different previously described high risk radical prostatectomy cohorts, sampled in triplicate or quadruplicate using 0.6 mm diameter tissue cores, and assembled into TMAs. These samples included patients who went on to develop metastasis and CRPC [10], patients who received adjuvant docetaxel after prostatectomy [11], and patients with primary Gleason pattern 5 prostate cancer [12]. For methylation analyses, we utilized an additional cohort of 21 fresh frozen samples of primary prostatic AdCa from radical prostatectomy samples which were described previously [13,14].

Weighted Gene Correlation Network Analysis (WGCNA).

Unsigned WGCNA [15] was applied to 6 NEPC/AdCa datasets used previously for meta-analysis [5] in order to identify consensus modules among ~4,500 genes, including the 500 most variable in each dataset and various NE-related and prostate-related genes. Consensus modules consisted of a mixture of correlated and anti-correlated genes since unsigned analysis was used. A gene was labeled as positive (or negative) if its module eigengene coefficient was positive (or negative) for every dataset and mixed (+/−) otherwise. To determine genes with the greatest intramodular connectivity, each consensus module’s genes were ranked according to kWithin intramodular connectivity for each dataset, and then average rank was calculated across datasets. DAVID and PANTHER were used for gene-ontology enrichment analysis [16,17].

Multi-Experiment Gene Expression Query Tool (MEM):

We used MEM to query genes inversely expressed with INSM1 or ASCL1 across human cancer datasets https://biit.cs.ut.ee/mem/) [18,19]. Gene ID was entered as INSM1 or ASCL1 (213768_S_AT), using the Microarray AE Current (1.12.14) dataset across 2811 Affymetrix GeneChip Human Genome 133 Plus 2.0 datasets with keyword “cancer”.

Immunohistochemistry.

YAP1, INSM1, Chromogranin A and Synaptophysin immunohistochemistry (IHC) assays were performed in a CLIA-accredited laboratory using YAP1 (clone EP1674Y, Abcam, Cambridge, UK; 1:200), INSM1 (clone A-8, Santa Cruz Biotech, Dallas, TX, USA; 1:200), Chromogranin A (LK2H10, Roche; pre-dilute) and Synaptophysin (27G12, Leica, Wetzlar, Germany; 1:400) monoclonal antibodies on the Ventana Benchmark (Roche, Tucson, AZ, USA) immunostaining system. For immunofluorescence detection, antigen retrieval was performed using citrate buffer (10 mM, pH 6.0) and HIER (heat-induced epitope retrieval) method. After primary antibody incubation overnight, sections were incubated with secondary antibodies (IgG anti-Rabbit or IgG anti-Mouse conjugated with Alexa Fluor 488 or Alexa Fluor 594, Invitrogen/ThermoFisher, Waltham, MA, USA) for 1.5 h at room temperature and mounted using ProLong Gold Antifade with DAPI (Life Technologies/Thermofisher). Chromogranin, synaptophysin, INSM1 and YAP1 immunostaining was scored dichotomously by two observers (AT, TLL) on tissue microarrays. Tumors with any amount of cytoplasmic chromogranin or synaptophysin expression in any TMA core were considered positive for these markers. Tumors with focal or complete loss on any TMA core were considered to have YAP1 loss, and any focal nuclear positivity on any TMA core was considered positive for INSM1 expression.

Cell lines and culture conditions.

Parental cell lines were obtained from ATCC (Manassas, VI, USA). LNCaP, DU145 and PC3 cells were maintained in RPMI 1640 supplemented with 10% fetal bovine serum (FBS) and penicillin/streptomycin. LNCaP-abl cells were maintained in RPMI-1640 and 10% charcoal-stripped serum (CSS) and penicillin/streptomycin. H660 were maintained in HITES medium (RPMI 1640, supplemented with insulin, transferrin, beta-estradiol, hydrocortisone and sodium selenite). VCaP-CR cells (a gift from S. Yegnasubramanian, Johns Hopkins University) were derived by passaging VCaP cells through castrated animals and further sub-culturing in RPMI-1640 supplemented with 10% CSS supplemented with 1× B-27 Neuronal Supplement (Gibco/ThermoFisher, Waltham, MA, USA). All cells were maintained at 37 °C in a humidified incubator with 5% carbon dioxide. For pharmacological unmasking experiments decitabine (DEC) and vorinostat (SAHA) were purchased from Selleckchem (Houston, TX, USA), dissolved in DMSO. H660 cells were treated for 48 h prior to RNA extraction.

Protein extraction, immunoblotting and primary antibodies.

Immunoblotting was performed as described previously [20]. The following primary antibodies were used: YAP1 (#ab52771, clone EP1674Y, Abcam; 1:1000) INSM1 (clone A-8, Santa Cruz; 1:500), MASH1/ASCL1 (#556604, clone 24B72D11.1, BD Pharmingen; 1:1000), SOX2 (#3579, clone D6D9, Cell Signaling Technologies [CST]; 1:1000), SYP (#36406, clone D8F6H, CST; 1:1000), CHGA (#85798, clone E8X7R, CST; 1:500), ENO2 (#8171, clone D20H2, CST; 1:1000), N-Cadherin (#13116, clone D4R1H, CST; 1:500), NCAM1 (#99746, clone E7X9M, CST; 1:500), LATS2 (#5888, clone D83D6, CST; 1:1000), AR (#5153, clone D6F11, CST1:1000) and GAPDH (#2118, clone 14C10, CST; 1:2000).

shRNA mediated gene silencing and plasmid transient transfections.

Stable YAP1 gene silencing in LNCaP and DU145 cells was performed using pGIPZ-based short hairpin RNA (shRNA) vectors for human YAP1 (clone VGH5518–200174078, clone VGH5518–200185118, clone VGH5518–200279049 and clone VGH5518–200282387 (Dharmacon, Lafayette, CO, USA)), using DharmaFECT kb transfection reagent (GE Healthcare, Chicago, IL, USA # T-2006–01), following which, stable YAP1 knockdown clones were generated using puromycin selection. H660 cells were transiently transfected with YAP1 (pEGFP-C3-hYAP1 [#17843, Addgene, Watertown, MA, USA]) using Lipofectamine 3000 as previously described [20].

Methylation analysis.

Patient and patient-derived xenograft samples used in this study were described previously [13,21,22]. Site specific DNA methylation analyses of the YAP1 locus were performed as described previously [21,23]. In brief, DNA samples were digested with AluI and HhaI (New England Biolabs, Ipswich, MA, USA) and methylated DNA fragments were enriched using recombinant MBD2-MBD (Clontech, Mountain View, CA, USA) immobilized on magnetic Talon beads (Clontech). Precipitated DNA containing methylated DNA fragments were eluted and subjected to quantitative real-time PCR using the IQ SYBR Green Supermix (BioRad, Hercules, CA, USA) using YAP1-F: GGACTCGGAGACCGACCT and YAP1-R: GTCTTGGGGTTCATGACG primers. Male white blood cell (WBC) genomic DNA was treated in vitro with M.SssI (New England Biolabs) to obtain a fully methylated positive control. Untreated male WBC DNA served as a negative control. For quantitative assessment of locus specific methylation levels, Ct-values of the samples of interest were normalized to Ct-values of the positive control (SssI) and calculated methylation indices (ranging from 0.0 to 1.0) were used to derive methylation heatmaps.

RNA extraction, cDNA synthesis and Real Time PCR analysis.

Total RNA from cell lines/cell pellets/xenografts in triplicate replicates was extracted using RNeasy kit (Qiagen, Hilden, Germany), following the manufacturer instructions. Using SuperScript III First-Strand Synthesis System for RT-PCR (Invitrogen) and oligo(dT)₅₀ primers, cDNA synthesis was realized from 2μg of total RNA. qPCR reactions were performed using QuantStudio 12K Flex Real-Time PCR System (Life Technologies), using primer/probes for INSM1 (Hs.PT.58.27970986.g), SYP (Hs.PT.58.27207712) (Integrated DNA Technologies (Coralville, IA, USA) and YAP1 (Hs.00902712_g1) (Life Technologies). GAPDH was used as reference gene to normalize input cDNA. The mRNA relative expression changes were determined using the 2-ΔCt method.

RNA-seq analysis.

Patient and patient derived xenograft samples and RNA-sequencing used in this study were described previously [24–26]. Expression values were extracted and plotted as log2 FPKM (fragments per kilobase per million reads.) RNA sequencing of triplicate cell line replicates was performed at Novogene, Beijing, PR China. In brief, messenger RNA was purified from total RNA using poly-T oligo-attached magnetic beads. After fragmentation, the first strand cDNA was synthesized using random hexamer primers, followed by second strand cDNA synthesis using dUTP. Sequencing was performed on the Illumina Novaseq 6000 platform (San Diego, CA, USA). Sequencing reads were mapped to the hg38 human genome using STAR.v2.7.3a [27]. All subsequent analyses were performed in R. Gene level abundance was quantitated using the GenomicAlignments Bioconductor package [28]. Differential expression was assessed using transcript abundances as inputs to the edgeR [29] and limma [30] Bioconductor packages in R, filtered for a minimum expression level using the filterByExpr function with default parameters prior to testing, and using the Benjamin-Hochberg false discovery rate (FDR) adjustment. Gene expression results were ranked by their limma statistics and used to conduct Gene Set Enrichment Analysis (GSEA) [31] to determine patterns of pathway activity utilizing the curated pathways from within the MSigDBv7.4 and custom gene sets. Single sample enrichment scores were calculated using GSVA [32] with default parameters using genome-wide log2 FPKM values as input.

Simplified AR signaling signatures, including 10 validated AR target genes (NKX3–1, KLK3, ALDH1A3, FKBP5, KLK2, TMPRSS2, PLPP1, PART1, PMEPA1, STEAP4), or simplified NEPC signatures, including a 10 neuroendocrine gene signature (CHGA, SYP, CHRNB2, PCSK1, NKX2–1, CHGB, ELAVL4, ENO2, SCN3A, SCG3) or more complex NEPC signatures, including the Beltran signature [25], the Zhang signature [33] and the Tsai signature [5] were assessed, as was a cell cycle proliferation signature [34] and an Rb-loss signature [35].

Statistical analysis.

All data were obtained from independent experiments analyzed in triplicate unless noted. Fisher’s exact test was used to compare YAP1 and INSM1 expression in AdCa versus NEPC. The statistical significance of differences between mean values was analyzed using t-tests or 1-way ANOVA as indicated. Gene expression (log2 FPKM) of phenotype groups were compared using Wilcoxon-rank tests with Benjamini–Hochberg multiple testing correction.

Results

To complement our recent gene expression meta-analysis comparing AdCa and NEPC [5], we applied weighted gene correlation network analysis (WGCNA) to meta-analysis datasets, generating 5 consensus modules, M1 to M5, in decreasing order by size (supplementary material, Table S1). M1, M2, and M3 were enriched for cell cycle (adj p=9e-79), androgen response (19 Hallmark genes versus 1–2 in other modules), and neuronal differentiation (adj p=1.4e-6), respectively. M3 contained several well-known neuroendocrine-related transcription factors within its positive genes including ASCL1, as well as INSM1. The most connected negative genes in M3 were REST, a well-known repressor of neurogenesis that has been implicated in small cell lung cancer [36] and NEPC transdifferentiation [33,37,38], and surprisingly, YAP1, a transcriptional co-activator in the Hippo signaling pathway.

As INSM1 and ASCL1 are well-characterized markers of neuroendocrine differentiation across a diverse set of human carcinomas, including NEPC [39–46], we used a multi-experiment gene expression query tool (MEM, https://biit.cs.ut.ee/mem/) [18,19] to examine which genes are inversely correlated with INSM1 or ASCL1 expression across 100 human cancer datasets. Intriguingly, we found that YAP1, as well as YAP1’s negative regulators in the Hippo pathway (LATS2 and PTPN14) were among the top inversely correlated genes with INSM1 expression, all with highly significant negative enrichment (supplementary material, Figure S1A). Gene ontology analysis confirmed a significant enrichment for the Reactome “Signaling by Hippo” module (p=0.01; R-HSA-2028269) in the list of genes inversely correlated with INSM1. When we repeated the analysis for ASCL1, we found a very similar significant negative enrichment for YAP1, PTPN14 and LATS2 (supplementary material, Figure S1A). This peculiar enrichment for both YAP1 and its negative regulators is potentially consistent with established negative feedback circuits within the Hippo pathway, wherein YAP1 activation (and subsequent TEAD-regulated transcription) induces expression of YAP1 negative regulators including LATS2 [47]. Indeed, LATS2 expression was also significantly downregulated in NEPC compared to AdCa in our previously published meta-analysis [5], and PTPN14 and AJUBA (another negative Hippo regulator) were both negative genes included in M3 of the WGCNA described above (supplementary material, Table S1).

To validate the striking inverse correlation between YAP1 and INSM1 expression in prostate tumor samples, we examined two well-characterized prostate cancer xenograft series. In both the LuCaP [22,24] (Figure 1A, supplementary material, Figure S1B) and Vancouver Prostate Centre (VPC) [3] (Figure 1B) patient-derived xenografts, high expression of INSM1 and silencing of YAP1 distinguished most NEPC samples from AdCa samples. Next, we examined three large series of metastatic castration resistant prostate cancer that have been classified for neuroendocrine (NE) phenotype by pathology [25] (Figure 1C) or for joint NE-Androgen Receptor (AR) status using RNA-seq [24,26] (Figure 1D,E). Although the AdCa samples showed a wider spectrum of YAP1 expression compared to the patient-derived xenograft models, YAP1 and INSM1 expression were strikingly reciprocally regulated in both cohorts, with NEPC or NE+ phenotype samples showing higher INSM1 and significantly lower YAP1 expression than AdCa samples (supplementary material, Figure S1C,D).

Figure 1. Reciprocal regulation of YAP1 and INSM1 gene expression in patient derived xenograft (PDX) models and human prostate cancer tumors and metastases.

(A) Microarray gene expression data for INSM1 vs. YAP1 in the LuCaP PDX models [22] (plotted as quantile normalized absolute log intensities). (B) Microarray gene expression data for INSM1 versus YAP1 in the Vancouver Prostate Centre (VPC) PDX models [3] (plotted as quantile normalized absolute log intensities). (C) Microarray gene expression data for INSM1 versus YAP1 in the Cornell NEPC and AdCa study [25] (plotted as log₂[value+1]). (D) RNAseq gene expression data for INSM1 versus YAP1 in the SU2C-PCF West Coast Dream Team mCRPC samples [26] (plotted as log₂[FPKM capture+1]). (E) RNAseq gene expression data for INSM1 versus YAP1 in the University of Washington (UW) autopsy series samples [24] (plotted as log₂[FPKM capture+1]).

To validate these gene expression findings at the protein level and examine in situ expression patterns, we used immunohistochemistry to detect YAP1 and INSM1 in formalin fixed paraffin embedded (FFPE) murine and human prostate AdCa and NEPC tissues. The TRAMP mouse carcinogenesis model, which uses the probasin (Pbsn) promoter to express the SV40 large T and small t antigens, provides a spectrum of tumor differentiation from intraductal carcinoma to AdCa to NEPC [48]. In both PIN lesions and intraductal AdCa tumors, both synaptophysin-negative, YAP1 was highly expressed and up-regulated compared to adjacent normal epithelium. However, in invasive NEPC from the same mice, YAP1 expression was markedly downregulated (supplementary material, Figure S2A).

In human prostate AdCa and NEPC samples, we compared YAP1 and INSM1 expression and found them to be most commonly mutually exclusive, even in samples with concurrent AdCa and NEPC components (Figure 2A). This reciprocal regulation of YAP1 and INSM1 in NEPC and AdCa was particularly dramatic in cases where there was a collision of both components (Figure 2B) and the mutually exclusive expression of these two markers in individual cells was also evident on dual immunofluorescence studies in mixed cases (Figure 2C).

Figure 2. Reciprocal protein expression of YAP1 and INSM1 in human prostate AdCa and NEPC tumors.

(A) Immunohistochemistry for YAP1, INSM1, NCAM (CD56) and Androgen Receptor (AR) in concurrent NEPC (top row) and AdCa (bottom row) from the same prostate sample. All images reduced from 200x magnification. (B) Immunohistochemistry for YAP1, INSM1, and AR in a mixed tumor with collision of the AdCa (arrow) and NEPC components. All images reduced from 200x magnification. (C) Immunofluorescence for YAP1 (red) and INSM1 (green) in a in a mixed tumor with collision of the AdCa (arrow) and NEPC components. Nuclei are labeled with DAPI (blue) (image reduced from 630x magnification). (D) Contingency table for YAP1 and INSM1 expression as measured by immunohistochemistry across AdCa and NEPC tumor samples on tissue microarray (TMA). YAP1- tumors had focal or complete loss of YAP1 protein, while INSM1+ tumors had focal or diffuse nuclear expression of INSM1.

To determine the sensitivity and specificity of YAP1 and INSM1 to distinguish AdCa from NEPC, we examined the expression of these proteins in 192 high risk primary AdCa (samples obtained from radical prostatectomies) and 35 prostatic small cell NEPC samples (at transurethral resection or prostatectomy). Of the AdCa patients, 63% (122/192) had Grade Group 5 prostate cancer and 60% (116/192) had a pathologic stage of pT3b or N1 (supplementary material, Table S2). This enrichment for high grade and stage AdCa cases was intentional since the differential diagnosis of AdCa vs NEPC is generally only encountered in very high grade cases. Overall, we found that YAP1 was focally or completely lost in 88% (31/35) of the NEPC cases, compared to 25% (49/192) of the adenocarcinoma cases (p<0.0001). INSM1 was expressed in 77% (27/35) of the NEPC cases, compared to 5% (9/192) of the AdCa cases (p<0.0001; Figure 2D).

Though YAP1 was lost in a significant proportion of AdCa cases (supplementary material, Figure S2B), INSM1 was only expressed in a small minority of these cases (8% or 4/49). Any INSM1 expression was seen in a much lower fraction of AdCa cases overall (and only very focally within those cases) compared to chromogranin and synaptophysin (supplementary material, Figure S2C), where positivity for one or both of the latter markers was seen in 37% (70/188) of cases (supplementary material, Table S3). In total, 44% (20/45) of AdCa cases with YAP1 loss were positive for either chromogranin or synaptophysin or both, though this was not significantly higher than the fraction of YAP1 intact cases positive for chromogranin/synaptophysin (35% or 50/143; p=0.29). In contrast, AdCa cases that were INSM1 positive were significantly more likely to be chromogranin and/or synaptophysin positive compared to those that were INSM1 negative (78% or 7/9 versus 35% or 63/179; p=0.01). Overall, YAP1 loss was 89% (31/35) sensitive and 74% (143/192) specific for NEPC, while INSM1 expression was 77% (27/35) sensitive and 95% (183/192) specific for NEPC. Taken together, either loss of YAP1 or expression of INSM1, or both was 89% (31/35) sensitive and 72% (138/192) specific for NEPC, while the joint profile with loss of YAP1 and expression of INSM1 was 77% (27/35) sensitive and 98% (188/192) specific for NEPC.

Despite its use for classification of SCLC, it is unclear whether YAP1 silencing is simply a bystander event or whether it contributes mechanistically to INSM1 expression or NE differentiation in this tumor type [49] or in prostate cancer. Although YAP1 loss was seen in the absence of INSM1 expression in AdCa, our findings above indicate that significant INSM1 expression almost never occurred in tumors with high YAP1 expression (3% or 5/147 cases overall). Thus, we hypothesized that YAP1 loss may modulate INSM1 expression in the prostate. To begin to test this hypothesis, we examined YAP1 and INSM1 expression in common prostate cancer cell lines. Similar to our results in human tumors, the H660 cell line, a well-characterized NEPC model [49] showed loss of YAP1 with high expression of INSM1, while the AR-positive and AR-negative AdCa cell lines showed variable YAP1 protein expression, but uniformly absent INSM1 expression (Figure 3A).

Figure 3. YAP1 loss may modulate INSM1 expression in vitro but is not associated with global transition to NEPC gene expression.

(A) Immunoblotting for YAP1, INSM1, AR and markers of neuro-endocrine differentiation in a panel of NEPC and AR-negative or positive AdCa prostate cell lines. (B) Immunoblotting showing increased expression of exogenous YAP1 in H660 cells with transient overexpression of YAP1-GFP at 72 h, and associated downregulation of INSM1, ASCL1 and SOX2. Normalized protein expression of YAP1, INSM1, ASCL1 and SOX2 is quantified by densitometry in (C,D) (n>3; P values by Student’s t-test). (D) RNAseq analysis of LNCaP or DU145 cells with YAP1 shRNA expression (two clones of each line examined with three replicates each) and H660 cells with transient YAP1 over-expression (three biological replicates examined). Heatmaps for YAP1 gene expression, as well as a 10 gene AR or 10 gene neuroendocrine (NE) gene expression signatures are depicted for each sample. Heatmaps for another NEPC signature [25] (“Beltran”, divided into 29 genes over-expressed [UP] and 41 genes under-expressed [down] in NEPC compared to AdCa) and the cell cycle proliferation (CCP, 31 genes) [34] score are also shown.

To test whether YAP1 loss is sufficient to induce INSM1 expression or NE differentiation in AdCa cell lines, we used shRNAs against YAP1 to generate stable clones of LNCaP and DU145 AdCa cell lines with YAP1 knockdown. YAP1 knockdown in LNCaP clones (which have low basal YAP1 protein expression and are AR-positive) resulted in a modest though significant increase of INSM1 protein levels, but without reproducible increases in other neuroendocrine markers, including synaptophysin (SYP), chromogranin (CHGA) and ASCL1 (supplementary material, Figure S3A–C). In DU145 cells (which have high YAP1 expression and are AR-negative), YAP1 knockdown (supplementary material, Figure S3D) was not sufficient to increase INSM1 expression, though SYP and CHGA were modestly increased (supplementary material, Figure S3E, S3F). Next, we examined whether YAP1 expression modulates INSM1 expression or NE differentiation in NEPC cells. We transfected H660 cells with exogenous YAP1, driving transient, high level expression (Figure 3B,C). Constitutive YAP1 expression was sufficient to significantly reduce INSM1 protein expression in vitro, with additional decreases in ASCL1 and SOX2 expression, though these fell short of statistical significance (Figure 3D). Taken together, these data indicate that YAP1 loss may modulate INSM1 expression in vitro.

Next, we asked whether YAP1 knockdown in AdCa cell lines or YAP1 overexpression in NEPC cell lines affected global expression of NEPC-related genes via RNAseq. Single sample enrichment scores were calculated via Gene Set Variation Analysis (GSVA) for two clones each of LNCaP or DU145 cells with stable YAP1 knockdown (Figure 3E). This revealed decreased expression of cell-cycle proliferation transcripts (CCP) [34] in cells with YAP1 knockdown compared to control cells as expected [50], but no discernable difference in neuroendocrine gene expression (10 gene NE signature) or a published NEPC gene expression signature [25,51] and no difference in AR signaling based on AR-target gene expression (10 gene AR signature). Similar analysis of H660 cells with constitutive YAP1 expression revealed only a weak upregulation of genes typically downregulated in the NEPC signature [25] (Figure 3E). We also used Gene Set Enrichment Analysis (GSEA) to examine enrichment scores for three published NEPC signatures [5,25,33] as well as an RB-loss signature [35] and the aforementioned cell cycle proliferation signature [34] in the same cell lines with similar equivocal results (supplementary material, Figure S3G). Taken together, these data suggest that the effects of YAP1 on INSM1 expression in vitro are not indicative of a more general transition to a neuroendocrine gene expression phenotype.

Finally, we queried the mechanism of YAP1 silencing in NEPC. The complete but often very focal loss of YAP1 immunostaining in NEPC and a subset of high grade AdCa tumor cells strongly suggested genomic silencing. However, YAP1 deletions occur in <1% of prostate AdCa and NEPC [25,52], and recurrent mutations have not been identified. To test whether YAP1 is regulated epigenetically, we investigated the methylation status of YAP1 using a previously described and extensively validated combined methylation-sensitive restriction enzyme digestion and methylated-DNA precipitation protocol (COMPARE-MS) [21,23]. As a control, human leukocytes treated with or without a CpG Methyltransferase (M.SssI) were also included (Figure 4A). Whereas the H660 cell line showed a high level of CpG island methylation near the transcriptional start site of YAP1 at the intron 1/exon 2 junction, the prostatic adenocarcinoma cell lines LNCaP and DU145 showed no detectable methylation at this locus (Figure 4A). Further analyses of human primary tumor samples revealed robust YAP1 methylation in FFPE prostatic small cell carcinomas (Figure 4A), but not in 21 fresh frozen samples of primary prostatic AdCa (Figure 4B).

Figure 4. The YAP1 locus undergoes CpG hypermethylation in small cell carcinoma, but not conventional adenocarcinoma.

(A) Methylation heat maps derived from COMPARE-MS analysis of the prostate cancer cell lines, primary small cell carcinoma (SCCa) of the prostate, conventional adenocarcinoma (AdCa), and adjacent benign prostate tissue (Benign). Male white blood cell DNA (WBC) and white blood cell DNA which was untreated or treated in vitro by CpG Methyltransferase M.SssI (WBC SSSI) were used as negative and positive controls, respectively (methylation heat map: gray – dense methylation; white – no methylation). (B) Methylation analysis in 21 primary adenocarcinomas and adjacent benign prostatic tissues shows absence of YAP1 methylation in conventional adenocarcinoma. (C) Gene expression heatmaps for synaptophysin (SYP), INSM1 and YAP1 (top) and methylation heatmaps of YAP1 (bottom) derived from 43 LuCaP PDX models (methylation heat map: gray – dense methylation; white – no methylation; expression heatmap: red – high expression; white – no expression). (D) Normalized YAP1 expression by RT-qPCR in H660 cells treated with DMSO (control), decitabine (DEC, 1 μM), vorinostat (VOR, 5 nM) or both for 72 h (n=4; P value by 1-way ANOVA).

To further investigate the association between epigenetic silencing of YAP1 and NEPC, we analyzed the correlation of YAP1 methylation and gene expression along with INSM1 and SYP gene expression in the LuCaP patient-derived xenograft series. LuCaP lines 145.1, 145.2 and 173.1 are classified as NEPC xenografts based on prior work [24], while LuCaP 77CR is described as amphicrine [24], with high neuroendocrine marker expression but intact AR expression as well. All four of these lines showed high INSM1 expression and some of the highest levels of YAP1 methylation, and all but 145.1 demonstrated concomitant YAP1 gene expression silencing (Figure 4C). LuCaP 49 and 93 are also NEPC models, however we were unable to detect significant INSM1 expression, and neither one of these models showed loss of YAP1 gene expression or YAP1 methylation in tumor passages analyzed for methylation (in contrast to earlier passages analyzed for gene expression; Figure 1A). Interestingly, one AdCa line, 86.2CR showed loss of YAP1 gene expression and moderate YAP1 methylation with minimal INSM1 expression, consistent with our findings of YAP1 loss in a subset of advanced AdCa cases. Finally, pharmacologic unmasking experiments showed independent and synergistic activity of the demethylating agent, decitabine (DEC) and the histone deacetylase inhibitor vorinostat (VOR) for increasing YAP1 mRNA expression in H660 cells (Figure 4D). Taken together these results suggest that the YAP1 locus undergoes epigenetic silencing via CpG hypermethylation in NEPC and this may in part maintain INSM1 expression in vitro.

Discussion

Hippo signaling is a highly evolutionarily conserved pathway whose best characterized role is in the regulation of organ size, via modulation of cellular proliferation and apoptosis. When Hippo signaling is inactive, unphosphorylated YAP and TAZ undergo nuclear translocation where they interact with the TEAD transcription factors to promote proliferation and cell survival. In mouse models, YAP1 is an oncogene, and accordingly, YAP1 overexpression (and LATS1/2 down-regulation) has been documented in a multitude of human neoplasms [53], including prostate cancer [54–57]. Although the mechanism of this up-regulation remains unclear, this is consistent with our findings in the TRAMP mouse model, where YAP1 was up-regulated in PIN lesions compared to surrounding benign epithelium, and consistent with the high level of YAP1 expression we found in most AdCa cases. Recent work in animal models has suggested that ERG, which is frequently up-regulated in prostate cancer due to gene rearrangements, may activate YAP1-regulated transcription leading to age-related prostate tumorigenesis [56]. Beyond its cell autonomous effects, prostate tumor cell YAP1 expression may also contribute to recruitment of myeloid-derived suppressor cells to the prostate cancer immune microenvironment in mice, further hastening tumor progression [57]. In light of this work, our finding that YAP1 gene and protein expression is silenced in the majority of NEPC cases was unexpected. This observation appeared to be independent of tissue-of-origin since YAP1 and other Hippo pathway gene expression was highly negatively correlated with neuroendocrine-related transcription factor INSM1 expression across hundreds of human tumor datasets. Indeed, several groups have reported that YAP1 expression is silenced in small cell lung cancers compared to non-neuroendocrine lung tumors in cell line and human tissue studies [6,58,59], and one of these studies also observed an inverse correlation between YAP1 and INSM1 expression [6]. Recent work in Merkel cell carcinoma, a high grade neuroendocrine carcinoma of the skin, also revealed YAP1 silencing [60]. Finally, while our work was in progress, another group found YAP1 downregulation in a small series of NEPC as well [61]. Similar to our current findings, Cheng, et al reported decreased expression of YAP1 at the RNA and protein levels in the LuCaP xenografts and in a series of 9 small cell NEPC cases, respectively [61]. Interestingly, Cheng, et al found that YAP1 was heterogeneously expressed in a subset of the primary small cell NEPC cases (<25% of cells), a finding that we rarely observed unless there was admixed AdCa (Figure 2C), and a difference potentially attributable to case selection or to unrecognized admixed non-NEPC cells.

Despite this progress, the mechanism of YAP1 silencing in neuroendocrine tumorigenesis remains unclear. Interestingly, suppression of YAP1 occurs during normal neuronal differentiation wherein expression of neuronal transcription factors Neurog2 or Ascl1 was sufficient to downregulate YAP1 mRNA levels through an undescribed mechanism and simultaneously activated Lats1/2 kinases [62]. Notably, our data do not support a similar role for LATS1/2 since we found that LATS2 expression was also significantly down-regulated in NEPC. In lung cancer cell lines, another pro-neuroendocrine transcription factor, ASCL1, was found to induce miR-375 expression, which targets YAP1 and down-regulates it [63]. Indeed, in medullary thyroid carcinoma (a neuroendocrine tumor), miR-375 is up-regulated and inversely correlated with YAP1 expression [64,65]. Although these studies implicate other pro-neuronal transcription factors in YAP1 downregulation, another study in Drosophila has proposed a more direct link between Yorkie (the YAP1 homolog) and Nerfin-1 (the INSM1 homolog) [66] wherein INSM1 binds to YAP1 and TEAD1, directly suppressing transcriptional activation of YAP1 target genes. Finally, as an additional mechanism in neuroendocrine lung cancer cell lines, knockdown of the RB1 tumor suppressor (a genomic driver of lung small cell carcinoma and NEPC) was sufficient to suppress YAP1 expression, though the specific mechanism of this effect was not elucidated [6].

A common theme of these prior studies is that they all posit that neuroendocrine transcription factors are likely upstream of YAP1 silencing. To our knowledge, few studies have tested whether YAP1 down-regulation is itself sufficient or permissive for induction of neuroendocrine differentiation. Given that we found YAP1 loss is not uncommon in high risk prostatic AdCa, and NEPC is thought to result from transdifferentiation from AdCa in many cases, we hypothesized that YAP1 loss might modulate neuroendocrine differentiation in prostate cancer. Accordingly, we found that exogenous YAP1 over-expression was associated with INSM1 protein downregulation in H660 cells, though there was no downregulation of the overall neuroendocrine differentiation program by RNAseq. Similarly, YAP1 loss was not sufficient to consistently drive a robust neuroendocrine differentiation program at the protein or transcriptional level in LNCaP or DU145 cells. Consistent with our findings, Cheng et al also noted that YAP1 suppression was insufficient to drive global neuroendocrine gene expression by RNAseq in transient knock-down experiments using PC3 cells [61].

While these in vitro data are far from confirmatory, they do suggest the potential for a positive feedback loop wherein neuroendocrine differentiation may initiate YAP1 loss and YAP1 loss further reinforces INSM1 expression. Based on the occurrence of YAP1 loss in high grade prostatic adenocarcinomas lacking neuroendocrine marker expression and the heterogeneous expression of neuroendocrine markers in our YAP1-knockdown cell lines, it seems most likely that YAP1 loss is not sufficient for neuroendocrine differentiation in the prostate. Future studies will examine the potentially post-transcriptional mechanism by which abrogation of Hippo signaling might maintain expression of neuroendocrine transcription factors such as INSM1 and ASCL1.

To our knowledge, few previous studies have suggested that promoter methylation is the mechanism of YAP1 silencing during tumorigenesis. Given the profound and sometimes quite focal loss of YAP1 mRNA and protein seen in NEPC, we hypothesized that the mechanism of loss was likely to be epigenetic. Indeed, we found that CpG methylation near the YAP1 transcriptional start site (near the intron 1–exon 2 border) was well correlated with YAP1 expression across human prostate tissues and cell lines, as well in the LuCaP patient-derived xenograft series. Interestingly, one previous study using the Infinium Epic platform in SCLC has shown that YAP1 methylation at the cg20782778 probe (which is near the 3’-end of the region we examined) is significantly correlated with YAP1 gene expression in these tumors [67], however in a small subset of three FFPE SCLC samples we examined by COMPARE-MS, we did not see dense methylation in this region (data not shown), suggesting that potential tissue-specific methylation patterns may underlie YAP1 silencing in different neuroendocrine tumor types. Notably, some studies have suggested that YAP1 silencing in SCLC may be methylation-independent and due to miR-375 expression which targets YAP1, as discussed above [63]. How the YAP1 locus becomes methylated in NEPC or SCLC remains an additional open question and will require further mechanistic investigation. Future studies may also examine whether YAP1 methylation could be used as potentially sensitive (though not specific) biomarker for NEPC in cell-free DNA.

Finally, our study adds to the growing number of immunohistochemical markers available as ancillary tests to establish an NEPC diagnosis. While in classic cases morphology is sufficient to make this diagnosis, the increasing number of hybrid or transitional cases has presented a growing challenge to practicing pathologists. At least two small prior studies have documented the potential utility of INSM1 in the differential diagnosis of NEPC and high grade AdCa [43,44] and our work validates these findings. We also find that YAP1 loss - while not specific for NEPC - appears to be highly sensitive and may be an additional marker to add to the current panel which includes chromogranin, synaptophysin, NCAM [2], FOXA2 [68], PDX1 [69] as well as Rb1 [8] and/or cyclin D1 [9], AR and prostate-lineage related markers [2].

Supplementary Material

tS1

Table S1. Consensus modules from WGCNA analysis with sign of coefficient contributing to eigengenes. There were five modules M1 to M5 in decreasing order by size. M1, M2, and M3 were enriched for cell cycle, androgen-response, and neuron differentiation respectively. The most connected negatively correlated genes in M3 were YAP1 and REST.

tS2

Table S2. Pathological features of the AdCa cohort.

tS3

Table S3. YAP1, INSM1, CHGA and SYP immunohistochemistry status in the AdCa cohort (0=lost, 1=intact)

supinfo

Figure S1. YAP1 mRNA expression is negatively correlated with INSM1 expression and down-regulated in NEPC datasets

Figure S2. YAP1 protein expression is down-regulated in murine models of NEPC and in some cases of human AdCa, though INSM1 is rarely expressed in human AdCa

Figure S3. Immunoblotting and RNAseq data from cell lines with shRNA-mediated YAP1 knockdown or transient YAP1 over-expression