Resistance to androgen receptor signaling inhibition does not necessitate development of neuroendocrine prostate cancer

Resistance to AR signaling inhibitors (ARSis) in a subset of metastatic castration-resistant prostate cancers (mCRPCs) occurs with the emergence of AR– neuroendocrine prostate cancer (NEPC) coupled with mutations/deletions in PTEN, TP53, and RB1 and the overexpression of DNMTs, EZH2, and/or SOX2. To resolve whether the lack of AR is the driving factor for the emergence of the NE phenotype, molecular, cell, and tumor biology analyses were performed on 23 xenografts derived from patients with PC, recapitulating the full spectrum of genetic alterations proposed to drive NE differentiation. Additionally, phenotypic response to CRISPR/Cas9-mediated AR KO in AR+ CRPC cells was evaluated. These analyses document that (a) ARSi-resistant NEPC developed without androgen deprivation treatment; (b) ARS in ARSi-resistant AR+/NE+ double-positive “amphicrine” mCRPCs did not suppress NE differentiation; (c) the lack of AR expression did not necessitate acquiring a NE phenotype, despite concomitant mutations/deletions in PTEN and TP53, and the loss of RB1 but occurred via emergence of an AR–/NE– double-negative PC (DNPC); (d) despite DNPC cells having homogeneous genetic driver mutations, they were phenotypically heterogeneous, expressing basal lineage markers alone or in combination with luminal lineage markers; and (e) AR loss was associated with AR promoter hypermethylation in NEPCs but not in DNPCs.

mg/kg/d. These doses have been documented to be effective in castrated mice against androgen sensitive ARPCs (3).
Serum PSA was measured as described previously (7).

Immunohistochemical (IHC) Staining
IHC staining was performed by the SKCCC Immunohistochemistry Core. Pretreatment conditions and primary antibody concentrations and source are summarized in (Supplemental Table 3). Images were taken using a Nikon DS-Qi1 Mc camera and NIS-Elements AR3.0 imaging software.

RNA Sequencing Analysis of the PDXs
RNA concentration, purity, and integrity was assessed by NanoDrop (Thermo Fisher Scientific, Inc.) and Agilent Tapestation. RNA-seq libraries were constructed from 1 ug total RNA using the Illumina TruSeq Sequencing reads were mapped to the hg38 human and mm10 mouse genomes using STAR.v2.7.3a (8).
Sequences aligning with higher specificity to the mouse genome deriving from potential contamination with mouse tissue were removed from the analysis using XenofiltR. Gene level abundance was quantitated from the filtered human alignments in R using the GenomicAlignments Bioconductor package (9) and normalized to kilobase of transcript per million mapped reads (FPKM) using edgeR (10). The expression values are shown as log2 FPKM. Sample phenotypic groups were assigned using classical multidimensional scaling (MDS) calculated with the cmdscale function in R on the expression profile of 21 genes from the combined lists of the 10-gene "NE" and "AR" signatures in (2) plus AR. The distance metric was "euclidean" calculated by the dist function on the columns (samples). The LvCaP-1, LN-95, CWR22, and PC-82 PDXs, which are AR-positive adenocarcinomas derived from hormonally-untreated primary prostate cancer patients, were used as additional canonical AR-PCa examples for comparison. These PDXs were obtained and propagated as previously described (11). LuCaP 49, 93, 145.1, and 145.2 were generated as previously described and used as representative examples of NEPC PDXs (12).

DNA Sequencing of the PDXs
A customized NGS panel targeting 355 cancer related genes was used to sequence DNA samples. The probes for capturing exon regions in these genes were manufactured by Roche NimbleGen. SeqCap EZ Library SR User's Guide (Roche, Pleasanton, CA) was followed for library preparation and capture of targeted sequences. Paired-end sequencing of 2 x 150 bp was performed on an Illumina NextSeq500. The median coverage for these 64 samples was at 324x. Paired-end reads was aligned to the GRCh37 version of the human genome using Burrows-Wheeler Aligner v0.7 to generate BAM files (13). After sorting the BAM files using samtools, PCR duplicates marked using Picard and realignment around putative gaps was performed using the Genome Analysis Toolkit (GATK) v3.2-2. Variant calling was performed with the GATK Haplotype caller. ANNOVAR (http://annovar.openbioinformatics.org/en/latest) and snpEff were used for annotating variants and for retrieving information on variants in the population-based studies such as the 1000 Genomes Project (www.1000genomes.org), NHLBI-ESP 6500 exomes, or ExAC (http://exac.broadinstitute.org/) or gnomAD (http:// http://gnomad.broadinstitute.org/), and clinical databases such as the Human Gene Mutation Database (HGMD) (14) and ClinVa (15). Pathogenicity of variants is defined based on American College of Medical Genetics and Genomics (ACMG) criteria (16). Specifically, pathogenic and likely pathogenic mutations are defined as 1) all protein truncating mutations unless their allele frequency is 5% or higher in any racial group in population databases or is reported as benign or likely benign in the ClinVar, and 2) nonsynonymous changes if their allele frequency is less than 5% and reported as pathogenic and likely pathogenic mutations in the ClinVar.

AR methylation analysis
For AR locus-specific DNA methylation analyses, a previously validated assay combining methylated-DNA precipitation and methylation-sensitive restriction enzyme digestion (COMPARE-MS) was used (17). In brief, DNA samples were digested with AluI and HhaI (New England Biolabs) and methylated DNA fragments were enrichment using recombinant MBD2-MBD (Clontech) immobilized on magnetic Tylon beads (Clontech). The precipitated DNA containing methylated DNA fragments was eluted and subjected to quantitative real-time PCR using IQ SYBR Green Supermix (Biorad) with primers specific to the CpG island in the first exon of AR F: PDXs were carried out using Infinium Methylation EPIC BeadChip arrays (Illumina) as described previously (18). Raw data were analyzed using the Minfi Bioconductor package and visualized using the IGV (19,20).

Plasmid Construction and Transfection
Generation of LN-95 total AR KO cell lines was described previously (3). Briefly, the CRISPR-Cas9 vector used for AR gene editing was pSpCas9(BB)-2A-GFP (Addgene plasmid ID: 48138). The guide RNAs were designed using CRISPR guide design tool: http://crispr/mit.edu/. Guide RNAs targeting different AR exons were listed below: sgAR-Exon1: CCGCCGTCCAAGACCTACCG. To construct the expression plasmids for guide RNAs, all guide RNA oligo pairs were annealed and cloned into the BbsI restriction site of pSpCas9(BB)-2A-GFP plasmid, and the correct plasmids were verified by sequencing from the U6 promoter using the U6-Fwd primer (GAGGGCCTATTTCCCATGATTCC). One day before transfection, 400,000 LN-95 cells were seeded in each well of 6-well plates in phenol-red free RPMI1640 supplemented with 10% charcoal-stripped FBS and 1% Penicillin/Streptomycin to achieve 60% confluence. The next day, each well was transfected with 2.5 μg of plasmid expressing the desired AR-guide RNA or empty GFP vector control with 3.75 μL of Lipofectamine 3000 (ThermoFisher, cat # L3000008).

Isolation of Clonal Cell Lines by FACS
After 48 hours of transfection, the cells were dissociated by TrypLE and resuspended into the FACS sorting buffer (PBS + 2% BSA + 20 mM EDTA). The cells were filtered into cell strainer tube, and sorted by Flow Cytometer (S3e Cell Sorter, BioRad) for GFP positive population. The sorted cells were then seeded in 100 mm dishes in phenol-red free RPMI1640 supplemented with 10% charcoal-stripped FBS and 1% Penicillin/Streptomycin, with 600 cells/dish and allowed to grow into colonies. Approximately 4 weeks later, 10-20 colonies were picked from each transfected cells using cloning discs (diam. 3.2 mm, #Z374431, Sigma-Aldrich) and transferred into 24 well plates to expand for further analyses.

Genotyping of Clonal Cell Lines
Once colonies were expanded into each well of 6-well plates, genomic DNA was isolated using QiaAmp

AR Isoforms and HOXB13 Protein Analysis
The whole cell lysates were prepared using RIPA buffer (Pierce) according to the manufacture's suggestions. Nuclear and cytosolic lysates were extracted by Nuclear and Cytoplasmic Extraction Reagents (Pierce). Protein samples concentration was quantified using BCA Protein Assay Kit (Pierce) and equal amount of protein samples were resolved on 4%-12% gradient SDA-PAGE gels and subjected to western blotting with anti-AR(441) (Santa Cruiz Biotechnology), anti-AR-V7 (Precision, AG10008) and anti-AR-CTD (Sigma, SP242 SAB5500007). For HOXB13 western blotting, whole cell lysates were resolved and probed with appropriate antibody (R&D, AF8156). IHC slides were imaged using the TissueFAXS Plus (Tissue Gnostics, Vienna, Austria) automated microscopy workstation equipped with a Zeiss Z2 Axio Imager microscope (Carl Zeiss Microscopy, LCC, Thornwood, NY, U.S.A.) and appropriate fluorescence filter sets. Exposure times for each channel (HOXB13, chromogranin and DAPI) were optimized to minimize pixel saturation while maximizing signal to noise for each fluorescent marker and the entire tissue section was then scanned with a 40X objective in an automated fashion. The digitized IF signals were then assessed using the TissueQuest 6.0 software (Tissue Gnostics) module to analyze the fluorescent images with precise nuclear segmentation on DAPI. The HOXB13 fluorescence staining intensity was captured for 97 individual cells with clear and robust cytoplasmic staining for CGA with an associated nucleus. HOXB13 fluorescence was also determined for 97 nearby basal ductal epithelial cells judged by location relative to stroma and relative to adjacent luminal epithelial cells, and judged also by nuclear size and shape in keeping with basal epithelial cell phenotype(e.g. smaller and more compact that the adjacent luminal layer, and also not possessing spindle or cylindrically shaped nuclei; as are typically seen for stromal smooth muscles and fibroblasts) and for 24 corresponding luminal epithelial cells which displayed markedly higher levels of HOXB13 in all cases. For each cell type, the ratio of total nuclear HOXB13 staining intensity to the nuclear area (DAPI staining) was determined, in order to further correct for fractional nuclei in the cut tissue section.

Immunohistochemistry (IHC) for LN-95 Cell Lines
Cell pellets were fixed in 10% neutral-buffered formalin and embedded in paraffin. Sections were cut and

RNA Sequencing of In Vitro Cell Lines
Parental LN-95, and AR-KO cells were subjected to RNA-Seq following the standard Trizol/RNeasy total RNA Prep Kit. The quality of the RNA extracted were verified by Qubit 3 (Invitrogen) and Bioanalyzer (Agilent) and sequenced using the Illumina HiSeq 2000 platform (Illumina Inc, San Diego, CA). An average of ~100 million reads per sample were generated. Sequences were aligned to UCSC hg19 genome building using TopHat, and the insertion or deletions were visualized using Integrative Genomics Viewer (IGV) (19). Read counts were obtained using HTSeq, and normalized per kilo base-pair gene length and per million reads library size (RPKM).

Statistical Analysis
Results are representative and expressed as mean +/-SEM. A p-value of <0.05, determined by Student's T-test or ANOVA test when appropriate, was considered statistically significant.

Study Approval
Tissue collection for research was approved by the Johns Hopkins University School of Medicine IRB.
Tumor specimens were acquired from mCRPC patients who signed informed consent. All animal procedures were approved by the Johns Hopkins University School of Medicine Institutional Animal Care and Use Committee.

Data Availability
The RNA-seq data for this publication has been deposited in NCBI's Gene Expression Omnibus and are accessible through accession number GSE160393 for the raw and mouse-gene subtracted PDX data, and GSE131985 for the LN-95 and AR-KO cells.