The Histone Methyltransferase KMT2D is a Critical Mediator of Lineage Plasticity and Therapeutic Response in Castration Resistant Prostate Cancer

Abstract

Castration-resistant prostate cancer (CRPC) is largely dependent on the androgen receptor (AR) for growth and often exhibits hyperactive PI3K signaling, most frequently due to PTEN loss. Therapeutic pressure from anti-AR therapies can induce trans-differentiation toward an AR-independent phenotype. Recently, different subtypes of AR-independent CRPC have been redefined, with the stem cell-like (SCL) subtype emerging as one of the most prevalent. Elucidation of the epigenetic mechanisms controlling the maintenance of these distinct CRPC cell states could pave the way for effective combinatorial therapies for CRPC. In this study, we identified a key role for the histone methyltransferase KMT2D in establishing the chromatin competence necessary for the recruitment of AR and FOXA1 transcription factors (TFs) that are essential for the AR transcriptional output in AR-dependent CRPC cell lines, patient derived organoids, and patient samples. Unexpectedly, KMT2D maintained the identity of the AR-low CRPC-SCL subtype and controlled activity of AP-1 TFs such as FOSL1, which acts as a master regulator of this subtype. Single cell transcriptomics and chromatin assays underscored the role of KMT2D in sustaining a mixed lineage cell state via AP-1 and FOXA1. The combined suppression of PI3K/AKT and KMT2D reduced cell proliferation in prostate cancer cells and patient-derived organoids in both CRPC-AR and CRPC-SCL subtypes. Altogether, these results unveil KMT2D as a major mediator of the epigenetic landscape in subtype-specific CRPC, contributing to tumor growth and therapeutic response.

Introduction

Androgen receptor (AR) orchestrates an oncogenic gene program crucial for prostate cancer (PCa) proliferation and survival (1). As such, androgen deprivation therapy (ADT) has been the standard of care for patients diagnosed with metastatic PCa. However, resistance often develops, leading to castration-resistant prostate cancer (CRPC). Reactivation of AR signaling represents the most common driver of CRPC and next generation AR signaling pathway inhibitors (ARPI) are used in combination with ADT as a therapy for CRPC. However, the strong selective pressure of these inhibitors can result in the trans-differentiation of these tumors into an AR-independent phenotype, through a process commonly referred to as lineage plasticity. In general, lineage-plastic CRPC is clinically aggressive and lacking effective treatments (2,3).

It is now clear that lineage plastic CRPC can have many different phenotypes and heterogeneous tumors with mixed features (3). Based on chromatin accessibility and gene expression profiles of CRPC model systems, recent work has stratified lineage-plastic, AR-independent CRPC into 3 subcategories: CRPC with stem cell like features (CRPC-SCL), with neuroendocrine differentiation (CRPC-NE), and with active WNT signaling (CRPC-WNT) (4). CRPC-SCL tumors, expressing lower levels of AR, constitute the second most common CRPC subtype after CRPC-AR, followed by CRPC-NE, while CRPC-WNT is rare (4). Epigenetic regulators likely play a crucial role in maintaining the distinct chromatin accessibility profiles and CRPC phenotypes, yet mechanistic studies to delineate the epigenetic modifiers that are crucial for maintaining the CRPC subtypes are limited (4). Moreover, given the emergence of epigenetic regulators as promising targets for cancer therapy (5), a deeper understanding of subtype-specific regulators could pave the way for effective, rationale-based combinatorial therapies for CRPC.

Prostate cancers, like breast cancers, often exhibit aberrant activation of the PI3K pathway mainly due to PTEN loss, which occurs in 70% of metastatic PCa (6–12). Previous work by our group and others has revealed an important crosstalk between the PI3K pathway and nuclear hormone receptors (13,14). Aberrant PI3K signaling reduces dependency on nuclear hormone receptors for survival and proliferation. In turn, PI3K inhibition leads to increased ER or AR-dependent transcription in breast or prostate cancer, respectively. This reciprocal interaction limits the anti-tumor activity of a single treatment with PI3K inhibitors or nuclear receptor-targeting agents (13,14). FDA-approved combinations such as alpelisib, capivasertib, and inavolisib with anti-ER therapy, are effective in treating ER+/PI3K-driven breast cancer. In prostate cancer, a phase III clinical trial combining a second-generation AR antagonist with an inhibitor of AKT, significantly improved progression-free survival in patients with PTEN loss (15). This underscores the potential of AKT inhibitors as effective agents in combinatorial therapies for treating PI3K-driven prostate cancers.

Mechanistically, our group identified the histone methyltransferase KMT2D as a key regulator of ER/PI3K crosstalk activity upon PI3K inhibition in breast cancer through direct phosphorylation of KMT2D by the PI3K pathway effectors AKT/SGK, which negatively affects KMT2D function. KMT2D (also known as MLL4) is a member of COMPASS (Complex Of Proteins Associated with Set1) family and is a major histone methyltransferase that regulates transcription at enhancers through H3 lysine 4 mono-methylation and di-methylation (16–18).

We aim to uncover the epigenetic mechanisms that maintain androgen receptor (AR) transcriptional activity in AR-dependent prostate cancer and that drive lineage-plastic, AR-independent CRPC, with a focus on the most common AR-independent subtype, CRPC-SCL. We have identified a pivotal role for KMT2D in regulating growth and lineage phenotypes across distinct stages of PCa. Specifically, in AR-dependent prostate cancer, KMT2D modulates the activation of AR activity as a result of PI3K inhibition. In CRPC-SCL, we have discovered that KMT2D is crucial for maintaining the mixed lineage state of SCL by regulating the activity of AP-1 transcription factors. These findings may lead to the identification of novel combinatorial treatment strategies currently absent in the therapeutic arsenal against CRPC.

Materials and Methods

Experimental models

Cell lines

LNCaP cells were obtained from ATCC and cultured in RPMI 1640 (Corning) media, in plates coated with 0.01% poly L- lysine (sigma). HEK-293T cells were obtained from ATCC (ATCC^® CRL-3216) and were cultured in DMEM (Corning) media. VCAP, DU145 cells were obtained from Dr. Jelani Zarif’s Laboratory (Johns Hopkins School of Medicine) and were cultured in DMEM (Corning) media. PC3, 22Rv1 cells were obtained from Dr. Nathaniel Brennen’s Laboratory (Johns Hopkins School of Medicine) and were cultured in RPMI-1640 (Corning) media. All media was supplemented with 10% fetal bovine serum and 1% peniciliin/streptomycin.

All cell lines were cultured at 37°C under normal oxygen conditions (5% CO₂), used at low passages and periodically tested for mycoplasma contamination.

Prostate organoids

MSKPCA2, MSKPCA3, MSKPCA12 and LuCAP176, LuCAP49 prostate organoids were cultured as per established protocols (19,20). Briefly, metastatic prostate cancer organoids were isolated and digested in collagenase type II (Gibco). Organoids were cultured in advanced DMEM/F12 and passaged once a week in matrigel (Corning) via TrypLE (Gibco) dissociation.

Cell-derived xenograft studies

For the shKMT2D xenograft study, 4–6 week NSG male mice were castrated. A week later, 2 × 10⁶ PC3 cells in 1:1 PBS/Matrigel (Corning) were injected subcutaneously. Once the tumors reached ~0.1cm³, the mice were randomized into 5 mice per group, and treated with vehicle or 100mg/kg capivasertib (Selleckchem), P.O, BID, 5 days a week. At the same time, doxycycline-inducible feed (VWR) or control feed (VWR) were given to the mice. Tumor measurements and body weights were taken twice a week. After 14 days of treatment, tumors were extracted 2 hours after the last treatment and weighed. All procedures were approved by the Animal Care and Use Committee at Johns Hopkins University.

Experimental Methods

Lentiviral production

Lentiviruses expressing the empty vector (PLKO.1) or the shRNA against KMT2D (shKMT2D-1: TRCN0000013140; shKMT2D-2: TRCN0000013139) were transduced into LNCaP cells for constitutive expression of KMT2D knockdown. For a system of doxycycline inducible KMT2D knockdown, LNCaP cells and prostate cancer organoids were transduced with LT3REPIR (pRRL) vector (T3G-dsRED-mirE/shRNA-PGK-PURO-IRES-rtTA3). For lentiviral production, HEK-293T cells were transfected with pCMV-VSVG, pCMV-dR8.2 and the plasmid of interest with Lipofectamine 3000 (Invitrogen) according to the manufacturer’s instructions.

In vitro proliferation assays

All cells were pre-induced for 72 hours with −/+ 1μg/ml doxycycline prior to seeding for proliferation assays. LNCaP cells were seeded in 12 well plates in full media and treated with PI3K/AKT inhibitors −/+ 1μg/ml doxycycline treatment. Plates were fixed at days 0, 3, and 6 with 3.7% formaldehyde and quantified with 0.1% crystal violet. For the enzalutamide experiments, 5000 LNCaP cells were seeded in 96 well plates and treated with either DHT, Enzalutamide or the combination −/+ 1μg/ml doxycycline the next day. Prestoblue cell viability reagent (Thermofisher) was used to quantify the plates at day 8.

For the dose response curve in LNCaP cells with GDC0941, 6000 cells were seeded per well in a 96 well plate, serial dilutions were done starting from 8μM of GDC0941 or 4μM of capivasertib for both LNCaP and PC3 cells with −/+1μg/ml doxycycline and quantified using Cell Titer Glo (Promega) on day 7. Basal proliferation assays in LNCaP and 22Rv1 were done with a seeding of 5000 cells over a period of 6 days with KMT2D knockdown and quantified at days 1, 3, 6 using prestoblue. 1500 PC3 cells were seeded over a period of 9 days with readings taken at days 1, 3, 6, 9 using prestoblue.

The MSKPCA2 and MSKPCA3 organoid proliferation assays have been performed with pre-induction for 72 hours with 500ng/ml doxycycline. Later, 5000 cells were seeded in 24 well plates in triplicates in prostate organoid media. 1μM GDC0941 was added to the wells. Drug was refreshed every other day. On day 5, bright field pictures were taken at 4X magnification and cells were quantified using Cell Titer Glo (Promega).

Immunoblot and Immunoprecipitation

For immunoblot analysis, cell pellets were harvested, lysed with RIPA buffer supplemented with protease and phosphatase inhibitors (ThermoFisher). Protein quantification was performed with the Pierce BCA Protein Assay Kit (ThermoScientific). Whole protein lysates for high molecular weight proteins such as KMT2D (~590 KDa) were run on NuPAGE 3–8% bis-tris gradient precast gels (Invitrogen) in tris-acetate SDS running buffer (Invitrogen). PVDF membranes were used for gel transfers. For low molecular weight proteins, 4–12% bis-tris gradient precast gels (Invitrogen) were used in MOPS SDS running buffer (Invitrogen) and nitrocellulose membranes were used for gel transfers.

For the immunoprecipitation assay, cells were transiently transfected with the required plasmids using Lipofectamine 3000 (Invitrogen) for 48 hours, as described earlier. Cells were then lysed in RIPA buffer supplemented with protease and phosphatase inhibitors and incubated with Anti-V5 Agarose Gel antibody (Sigma Aldrich Cat# A7345, RRID: AB_10062721) for 2 hours. The immunocomplexes were washed thrice with RIPA buffer and a western blot was run as described above.

The membranes were probed with the following primary antibodies: anti-KMT2D (Fortis LifeSciences Cat# A300-BL1185, RB X-ALR, 1:500), phospho-KMT2D (S1331) (Eurogentec; Custom, 1:250), anti-AR [D6F11] XP (Cell Signaling Technology Cat# 5153, RRID:AB_10691711, 1:2000), anti-FOXA1 [E7E8W] (Cell Signaling Technology Cat# 53528, RRID:AB_2799438, 1:1000), anti-FOXA2 (Proteintech Cat# 22474–1-AP, RRID:AB_2879110, 1:1000), anti-V5 [D3H8Q] (Cell Signaling Technology Cat# 13202, RRID: AB_2687461, 1:1000), anti-synaptophysin [YE269] (Abcam Cat# ab32127, RRID: AB_2286949 1:1000), anti-beta actin (Cell Signaling Technology Cat# 4967, RRID: AB_330288, 1:2000), anti-vinculin (Cell Signaling Technology Cat# 4650, RRID: AB_10559207, 1:1000), Histone H3 (mono methyl K4) antibody- ChIP Grade (Abcam Cat# ab8895, RRID: AB_306847, 1:500), Histone H3 [D2B12] (Cell Signaling Technology Cat# 4620, RRID: AB_1904005, 1:1000), anti-keratin 5 (BioLegend Cat# 905504, RRID:AB_2616956, 1:10000), purified anti-keratin 8 (BioLegend Cat# 904801, RRID: AB_2565043, 1:5000), Vimentin [D21H3] XP (Cell Signaling Technology Cat# 5741, RRID: AB_10695459, 1:1000), ECL^™ Anti-Rabbit IgG HRP- linked antibody (from donkey) (Cat# NA934V; Lot 17010251, RRID: AB_772206, 1:4000), Anti-mouse IgG HRP- linked antibody (Cell Signaling Technology Cat# 7076, RRID: AB_330924, 1:2000).

Immunohistochemistry and Muliplex Immunofluorescence

Immunohistochemistry and multiplex immunofluorescence of the organoids and cell lines were performed using Ventana Discovery Ultra Auto Stainer, using published Ventana standard protocols. Briefly, the organoids/cells were deparaffinized, blocked and incubated with primary antibody for 60 minutes. The slides were incubated with OmniMap anti-Rb/Ms HRP for 16 minutes, with DAB CM and H2O2 CM for 8 minutes, and counstained with hematoxylin for 8 minutes. Antigen retrieval was done using sodium citrate pH 6. The antibodies used were purified anti-keratin 5 (BioLegend Cat# 905504, RRID:AB_2616956, 1:1000), purified anti-keratin 8 (BioLegend Cat# 904801, RRID: AB_2565043, 1:1000), Androgen Receptor [D6F11] XP (Cell Signaling Technology Cat# 5153, RRID: AB_10691711, 1:1000), KMT2D polyclonal antibody (Thermo Fisher Scientific Cat# PA5–115579, RRID: AB_2900214, 1:250), anti-FOXA1 [E7E8W] (Cell Signaling Technology Cat# 53528, RRID:AB_2799438, 1:1000) and Vimentin [D21H3] XP (Cell Signaling Technology Cat# 5741, RRID: AB_10695459, 1:100), DAPI (Carl Roth 6843.2, 1:1000).

RNA extraction, cDNA synthesis, RT-qPCR

Total RNA was extracted using the Qiagen RNeasy kit (Thermofisher) and cDNA was synthesized using the iScript cDNA synthesis kit (Bio-Rad) according to the manufacturer’s instructions. cDNA was amplified through a ViiA 7 Real-Time PCR system using SYBR select master mix (Applied Biosystems). Every sample was run in technical replicates in 384 well plates.

Histone extraction in LNCAP, PC3 and DU145 cells

Cells were seeded in regular full media conditions with −/+ 1μg/ml doxycycline. 72 hours later, cells were washed with 1X PBS and subjected to the histone extraction protocol from Abcam. Briefly, cells were lysed in Triton Extraction buffer (PBS containing 0.5% Triton X-100 (v/v), 2mM phenylmethylsulfonyl fluoride and 0.02% (w/v) sodium azide, and spun down to harvest a nuclear pellet. The nuclei were acid extracted using 0.2N HCl overnight at 4°C, pelleted and neutralized with 2M NaOH. Protein content was determined by BCA assay and immunoblots were run according to protocols described above.

Bulk RNA-seq, ChIP-seq, ATAC-seq in LNCaP cells

For AR ChIP-seq, LNCaP cells were seeded in regular full media conditions with −/+ 1μg/ml doxycycline to allow for robust attachment. 24 hours later, cells were washed twice with 1X PBS and refreshed with phenol red free media −/+ 1μg/ml doxycycline supplemented with 10% charcoal stripped serum and 5% penicillin/streptomycin. On the day of harvesting cells 72 hours later, the media was refreshed for 1 hour. Treatments with 10nM DHT and 1μM PI3K inhibitor GDC0941 coupled with 1μg/ml doxycycline were performed and the cells were incubated for 8 hours. For FOXA1, KMT2D and H3K4me2 ChIP-seq, LNCaP cells were seeded with −/+ 1μg/ml doxycycline for 72 hours and ChIP-seq was performed as previously described (21). Briefly, cells were crosslinked with formaldehyde, quenched with glycine to a final concentration at 125nM and lysed with SDS, after which they were sheared. Sheared chromatin was incubated with the appropriate antibody and protein dynabeads A/G. After decrosslinking at 65°C, DNA was eluted with AMPure beads. High throughput sequencing was performed using Illumina HiSeq 2500. The antibodies used for ChIP-seq were androgen receptor antibody [ER179(2)] (Abcam Cat# ab108341, RRID: AB_10865716), FOXA1 antibody- ChIP Grade (Abcam Cat# ab23738, RRID: AB_2104842), KMT2D antibody (Sigma-Aldrich Cat# HPA035977, RRID: AB_10670673) and Histone H3 (di methyl K4) antibody [Y47] (Abcam Cat# ab32356, RRID: AB_732924), FRA1/FOSL1 [D80B4] (Cell Signaling Technology Cat# 5281, RRID: AB_10557418).

ChIP-qPCRs were done with a rabbit IgG control (Abcam Cat# ab171870, RRID: AB_2687657).

The ChIP-qPCR primers used in the study are-

NDRG1 promoter: 5’-ATGGCCCCAGATATGTTCCA-3’, 3’-CCCAAGGTCTCAGAGCCAGT-5’

KLK3 enhancer: 5’-TGGGACAACTTGCAAACCTG-3’, 3’-CCAGAGTAGGTCTGTTTTCAATCCA-5’.

For the RNA-seq experiments, LNCaP cells were treated with 10nM DHT and 1μM GDC0941 or both with 1μg/ml doxycycline for 24 hours in hormone depleted medium. Total RNA extraction was performed using the Qiagen RNeasy kit (Thermofisher). For ATAC-seq analysis, LNCaP cells were cultured in DMSO or 1μg/ml doxycycline for 72 hours followed by treatment with 1μM PI3K GDC0941 for 24 hours. 50K LNCaP cells were detached using Accutase (Stemcell technologies) and were subjected to a published ATAC-seq protocol (22).

CUT&RUN in LNCaP cells

Cells were pretreated with 1μg/ml doxycycline for 72 hours in regular full media conditions before proceeding with the Epicypher CUTANA CUT&RUN v3 and library preparation v1.1 protocols following the manufacturer’s instructions. No E-coli spike-in was used. A validated Epicypher H3K4me1 antibody was used (Epicypher Cat# 13–0057, RRID: AB_3076424, 0.5μg).

Bulk RNA-seq, ATAC-seq in MSKPCA3 organoids

MSKPCA3 organoids were seeded in full media. On the same day, androgen deprived media was added to the cells −/+ doxycycline for 72 hours. The cells were then treated with 1μM PI3K inhibitor GDC0941 for 24 hours, after which organoids were processed for RNA-seq and ATAC-seq as previously described.

Bulk RNA-seq in MSKPCA2 organoids

MSKPCA2 organoids were androgen deprived with −/+doxycycline induction for 72 hours, followed by 24 hour induction with 10nM DHT, after which they were processed for RNA-seq as previously described.

Bulk RNA-seq, ATAC-seq in DU145 and PC3 cell lines

150K DU145 and PC3 cells were induced in regular full media conditions −/+ 1μg/ml doxycycline treatment for 72 hours, after which total RNA extraction was performed for RNA-seq using the Qiagen RNeasy kit (Thermofisher). 50K cells were processed as per the aforementioned ATAC-seq protocols (22).

RT-qPCRs in VCAP, 22Rv1, DU145 and PC3 cells

200K VCAP and 22Rv1 cells were seeded in regular full media conditions −/+ 1μg/ml doxycycline treatment. Next day, the cells were washed twice with 1X PBS and refreshed with phenol red free media −/+ 1μg/ml doxycycline supplemented with 10% charcoal stripped serum and 5% penicillin/streptomycin for 48 hours, after which the cells were treated with 10nM DHT for 24 hours. RNA extraction was then performed, after which cDNA was extracted and processed according to the protocols mentioned above. The primers used for VCAP cells are-

KLK3: 5’-TCATGCTGTGTGCTGGACGCTG-3’, 3’-CTTTCGGGCAGGGCACATGGTT-5’

NDRG1: 5’-ATCACCGGCCTCCTGCAAGAGT-3’, 3’-GAGTCCCACACAGCGTGACGTG-5’

GAPDH:5’- CCAGGTGGTCTCCTCTGACTTC-3’,3’- TCATACCAGGAAATGAGCTTGACA-5’

The primers used for 22Rv1 cells are-

KLK3: 5’-TCATGCTGTGTGCTGGACGCTG-3’, 3’-CTTTCGGGCAGGGCACATGGTT-5’

KLK2: 5’-AGTTCTTGCGCCCCAGGAGTCT-3’, 3’-TACCACCTGTCCAGAGCCCAGC-5’

NDRG1: 5’-ATCACCGGCCTCCTGCAAGAGT-3’, 3’-GAGTCCCACACAGCGTGACGTG-5’

NKX3.1: 5’-AGAACGACCAGCTGAGCAC-3’, 3’-TAAGACCCCAAGTGCCTTTC-5’

GAPDH: 5’- CCAGGTGGTCTCCTCTGACTTC-3’, 3’- TCATACCAGGAAATGAGCTTGACA-5’

150K DU145 and 200K PC3 cells were seeded in regular full media conditions with −/+ 1μg/ml doxycycline treatment. After 72 hours, RNA was extracted, cDNA was synthesized, and RT-qPCRs run according to the protocol mentioned above. The primers used for DU145 cells are-

FOSL1: 5’-AGGCCTTGTGAACAGATCAG-3’, 3’-TCATCTTCCAGTTTGTCAGTCTC-5’

FOSL2: 5’-CGGGAGCTGACAGAGAAG-3’, 3’-GGGCTAATCTTGCACACTGG-5’

JUN: 5’-AGCCCAAACTAACCTCACG-3’, 3’-TGCTCTGTTTCAGGATCTTGG-5’

JUNB: 5’-GGACACGCCTTCTGAACG-3’, 3’-CGGAGTCCAGTGTGGTTTG-5’

YAP1: 5’-AGATGGAGAAGGAGAGGCTG-3’, 3’-AGTGTTGGTAACTGGCTACG-5’

WWTR1/TAZ: 5’-AGCTCAGATCCTTTCCTCAATG-3’, 3’-TCCTGCGTTTTCTCCTGTATC-5’

ANKRD1: 5’-GGTGAGACTGAACCGCTATAAG-3’, 3’-GGCTGTCGAATATTGCTTTGG-5’

AXL: 5’-ATTTCCTGAGTGAAGCGGTC-3’, 3’-GCTGTGTAGGTCTCCATGTTTC-5’

CCND1: 5’-GCTGCGAAGTGGAAACCATCC-3’, 3’-CGCACTTCTGTTCCTCGCAG-5’

AJUBA: 5’-TGTCAATGGCTCTGTGTACTG-3’, 3’-ACTTCCCCATTGCTTGTAGG-5’

CTGF: 5’-TGCCCTCGCGGCTTACCGACTG-3’, 3’-TGCAGGAGGCGTTGTCATTGGTAAC-5’

CYR61: 5’- CAAGGAGCTGGGATTCGATG-3’, 3’-AAAGGGTTGTATAGGATGCGAG-5’

GAPDH: 5’- CCAGGTGGTCTCCTCTGACTTC-3’, 3’- TCATACCAGGAAATGAGCTTGACA-5’

Single-cell multiome library prep and sequencing

Single Cell Multiome ATAC + Gene Expression was performed with the 10X genomics system using Chromium Next GEM Single Cell Multiome Reagent Kit A (catalog no. 1000282) and ATAC Kit A (catalog no. 1000280) following Chromium Next GEM Single Cell Multiome ATAC + Gene Expression Reagent Kits User Guide and demonstrated protocol - Nuclei Isolation for Single Cell Multiome ATAC + Gene Expression Sequencing. Briefly, 100,000 cells (viability 95%) were lysed for 4min and resuspended in Diluted Nuclei Buffer (10x Genomics, PN- 2000207). Lysis efficiency and nuclei concentration was evaluated on Countess II automated cell counter by trypan blue staining. 9,660 nuclei were loaded per transposition reaction, targeting recovery of 10,000 nuclei after encapsulation. After transposition reaction nuclei were encapsulated and barcoded Next-generation sequencing libraries were constructed following the User Guide, which were sequenced on an Illumina NovaSeq 6000 system at the same time.

Computational Analysis

Bulk ChIP-seq and ATAC-seq

For ChIP-seq and ATAC-seq computational analysis, fastq files were generated containing single-end reads and paired-end reads respectively. Sequencing adaptors were removed using Trimmomatic v0.39. Trimmed reads were aligned to the hg38 human genome using Bowtie 2.4. Duplicate PCR and optical reads were removed using Picard MarkDuplicates v2.18.16. Aligned reads were extracted using SAMtools v1.9. Peaks were identified using a pipeline that adheres to standards set by ENCODE. This includes initial peak calling with MACS2 v2.1.3 with a liberal p value threshold of 0.1 normalized to the input. Replicate peaks were then assessed using an irreproducible discovery rate (p value<0.05). Those peaks passing IDR were then merged to form a peak atlas. Peak intensity was calculated using the countOverlaps function in the GenomicAlignments section of R. Peaks were annotated using ChIPseeker in the R package. The differential analysis of each condition compared to the control was done using DESeq2. For each pairwise comparison, only those peaks that were reproducible were considered for differential analysis. Moreover, peaks were required to have an FDR<0.05 (FDR, falMotise discovery rate) to be considered statistically significantly different between comparisons. Normalized DESeq2 bigwig files were used for plotting tornado plots using DeepTools. However for the LNCaP bulk ATAC-seq (adjusted p value<0.001 and log₂FC>0.5), AR ChIP-seq (FDR<0.05) in LNCaP cells or the FOSL1 ChIP-seq (FDR<0.05) in DU145 cells, IDR was not performed.

CUT&RUN

Sequencing reads were first quality-filtered and trimmed of adaptor sequences using fastq (23). The processed reads were aligned to the mouse reference genome (GRCm38) using Bowtie2 (24) under the parameters: --local --very-sensitive --no-unal --no-mixed --no-discordant -I 10 -X 800. Peak detection was carried out using MACS2 incorporating IgG controls for each antibody (25). Regions enriched for target of interest were identified using an FDR cutoff of 0.05 and a comprehensive peak atlas was constructed by merging peak regions across all samples. Read coverage across these union peaks were quantified with GenomicAlignments in R [software for computing] (26). Differential enrichment analysis was performed using DESeq2 (27), with significant peaks defined as those with an absolute value of log2FoldChange > 0.5 and FDR < 0.05. For visualization, BigWig files from biological replicates were averaged and DeepTools was used to generate tornado plots summarizing signal distributions at differential sites (28). Motif discovery on differentially enriched regions was conducted using HOMER (29).

Motif analysis

Both ATAC-seq and ChIP-seq peaks were distributed into upregulated and downregulated peaks based on their respective positive or negative fold change calculated by DESeq2 in a differential analysis. In figure 1 motifs were then identified within peaks using the HOMER suite with the script findMotifsGenome.pl. In figure 3 FIMO was used to identify motifs from the hocomocco motif database and significance determined with an FDR<0.05. We then assessed transcription factor activity using a generalized linear model as described in the following section.

Figure 1. KMT2D regulates AR transcriptional output and AR-dependent growth in prostate cancer.

(A) ATAC-seq tornado plot from LNCaP cells treated with DMSO or GDC0941 (1μM, 24 hours) showing sites with altered accessibility (adj p value<0.001, log₂FC>0.5; n = 2 biological replicates).

(B) Top de novo enriched motifs in gained and lost ATAC peaks upon GDC0941.

(C) AR binding (ChIP-seq) and chromatin accessibility (ATAC-seq) at the KLK3 promoter in LNCaP cells.

(D) Tornado plot of AR ChIP-seq peaks in LNCaP cells after combined treatment of DHT (10nM) and GDC0941 (1μM).

(E) RNA-seq heatmap of significant genes differentially expressed in LNCaP cells upon KMT2D knockdown (DOX 1μg/ml) under hormone-depleted conditions followed by DHT (10nM) or GDC0941 (1μM) treatment for 24 hours (q value<0.05; n = 2 biological replicates).

(F)-(G) GSEA of androgen response genes in LNCaP cells after treatment with DHT ± GDC0941 and after KMT2D knockdown ± DHT or GDC0941 (nominal p values and FDR adjusted p values, GSEA package).

(H) AR signature enrichment linked to KMT2D expression (high or low) in TCGA and SU2C patient cohorts. p value has been indicated using the Student’s t-test.

(I) Tornado plots showing KMT2D chromatin binding upon GDC0941 treatment (1μM, 8 hours) globally and at FOXA1 and AR sites.

(J) Tornado plot showing enrichment of H3K4me1 peaks in LNCaP cells with control and KMT2D knockdown (DOX 1μg/ml).

(K) Tornado plot of FOXA1 ChIP-seq upon KMT2D knockdown (DOX 1μg/ml) in LNCaP cells.

(L) AR ChIP-seq tornado plots in control and KMT2D knockdown cells treated with DHT (10nM) ± GDC0941 (1μM) in LNCaP cells.

(M)-(N) Prestoblue viability curves of LNCaP cells upon KMT2D knockdown (in blue, DOX 1μg/ml) ± PI3K/AKT inhibitors (GDC0941, AZD5363 (capivasertib)) (5 days, n = 4 biological replicates, representative shown).

(O) Proliferation assays of patient-derived organoid MSKPCA2 upon KMT2D knockdown ± GDC0941 (1μM) (n = 3 biological replicates; p value, student’s t-test).

(P) Kaplan-Meier survival curve of SU2C patients stratified by KMT2D activity.

Mean ± SD unless noted. ET, ethanol; DHT, dihydrotestosterone; DOX, doxycycline.

Figure 3. KMT2D is a key mediator of subtype-specific chromatin accessibility and cooperates with several different TFs that are required for maintaining the mixed-lineage state in CRPC-SCL.

(A)-(B) UMAP of single-cell RNA-seq (scRNA-seq) from MSKPCa3 organoids colored by cluster or experimental condition (control, shKMT2D) (DOX, 500ng/ml, 7 days).

(C) Fraction of control and shKMT2D cells per cluster.

(D) Single-cell expression signatures of basal epithelial, luminal epithelial including prostatic luminal club and hillock cells, along with androgen response (AR) enriched and stem cell like CRPC signatures. Colors indicate z-score of normalized enrichment transformed to [−2, 2] interval.

(E) Mean expression level (color) and percentage of expressing cells (dot size) of basal, luminal and its subtypes (club, hillock) compartment specific markers (columns) across 10 clusters (rows).

(F) Relative enrichment of compartment specific markers (p values, one-sided Wilcoxon rank-sum test).

(G) Hallmark GSEA of genes downregulated in shKMT2D cells.

(H) Multiplex IF of AR, KMT2D, CK8, CK5, DAPI and the overlay has been shown in MSKPCA3 organoids. Scale bar = 10μm.

Additionally, to detect the differential activity of TFs driving changes in chromatin accessibility profiles in response to GDC0941 treatment, we used the machine learning method gkm-SVM (30)(31), which is especially well suited for detecting subtle changes in sets of ~1000 sequences. We trained gkm-SVM on pos vs neg sequences (AUROC=0.88) and pos vs mid sequences (AUROC=0.71) and extracted Z-scores for the most predictive motifs from the gkm-weight distribution. We supported this gkm-SVM analysis by mapping the scores of these motifs in all peaks and calculating the average accessibility fold change for motif score bins.

Predictive modeling of transcription factor activity

To infer cell-state-specific TF activities, we performed a supervised modeling of chromatin accessibility data based on TF motif occurrences. A feature matrix consisting of TF motif match prediction, including only inferred motif genes that were also expressed in RNA-seq, in regions around peak summits was used as input for the generalized linear model (GLM). Each value in the matrix was the log2 fold change between KMT2D and control peak counts within the summit region after normalization. For each ATAC-seq sample, we used generalized linear regression modeling of the corrected chromatin accessibility fold changes in peak summit regions using this feature matrix, with ridge regularization, using glmnet package (32). We limited the regression analysis to peak summit regions with at least 10 normalized reads on average across all samples. We identified a hyperparameter multiplier of the ridge regularization penalty term using 5-fold cross-validation. Then we used the coefficients of this regression as a proxy for TF activity scores.

Integration of ATAC-seq with ChIP-seq

Accession files of AR ChIP-seq (GEO: GSE28264) (33) were downloaded and aligned to the hg38 human genome. Differential expression analysis was performed using DESeq2. Enrichment of genes were plotted using Deeptools using default parameters.

For the integration of MSKPCA3 ATAC-seq with MSKPCA3 ChIP-seq sites of FOSL1, TAZ, TEAD1 and YAP (GEO: GSE196732), the MSKPCA3 ATAC-seq replicate peaks were merged, and a tornado plot depicting gained and lost peaks was generated for significant differential peaks of KMT2D knockdown versus control using plotHeatmap in deepTools.

The multiBigwigSummary function in deepTools was used to generate a score matrix for FOSL1, TAZ, TEAD1, and YAP ChIP-seq data across all ATAC-seq differential peaks between KMT2D knockdown and control. A threshold was applied to retain approximately 25% of peaks with strong signals, which were then used to create a tornado plot with plotHeatmap in deepTools.

Bulk RNA-seq

Adaptor sequences were removed using Trimmomatic v0.39 and aligned to the hg38 human genome using STAR2.6.0. The reads that were uniquely mapped were counted using HTSeq0.9.1. The reads per kilobase per million (RPKM) was calculated for each gene and normalized to obtain gene expression levels.

For the MSKPCA2 RNA-seq, differential expression analysis of shKMT2D versus control was performed using DESeq2. The top 5,000 peaks were selected based on adjusted p-values. Heatmaps of shKMT2D and control samples were then generated using the ComplexHeatmap package in R.

For PC3 and DU145 RNA-seq analyses, transcript quantification was performed using Salmon. Differential expression analysis was conducted using DESeq2, and genes were classified as upregulated (log2FC ≥ 0.5, padj < 0.05) or downregulated (log2FC ≤ −0.5, padj < 0.05). DESeq2-normalized expression values of these differentially expressed genes were scaled to z-scores, and heatmaps were generated using the ComplexHeatmap package to visualize expression patterns across samples.

Signature activity score and survival analysis

The transcriptome data of TCGA prostate cancer cohorts was obtained from the cBioPortal (http://cbioportal.org). These patients were divided into groups based on the median expression of KMT2D. Androgen response and AP-1 TF-related signatures were obtained from MSigDB. R package GSVA 1.40.1 (34) was used to calculate the activity scores of these signatures. Patient samples were stratified into high or low populations based on the median value of the signature activity score. The significance of the activity scores was assessed using a t-test.

Gene expression data for the SU2C cohort was downloaded from cBioPortal (http://cbioportal.org), and gene signatures for four subtypes (AR, NE, SCL, WNT) were obtained from the article (4). Using R package Gene Set Variation Analysis (GSVA), we summarized the expression of each gene signature into a single score per patient. Then, K-means clustering was performed to classify patients based on these signature scores. The clustering analysis was initialized to create four clusters, corresponding to the four subtypes. Within these clusters, we identified groups of patients with high AR (Androgen Receptor) scores and those with high SCL (stem cell–like) scores.

To determine if the observations in the LNCaP cell line could also be observed in AR patient subtypes, we used the upregulated and downregulated genes upon KMT2D knockdown in the LNCaP cell line as signatures. AR patient subtypes were then grouped into KMT2D high and low groups based on their signature scores for the upregulated and downregulated genes. Patients with high score for upregulated genes and low score for downregulated genes likely exhibit transcriptional patterns similar to KMT2D knockdown samples, reflecting a positive transcriptional responses to the knockdown. Conversely, patients with low scores for upregulated genes and high scores for downregulated genes are more similar to WT samples. Incorporating both the directionality and magnitude of gene expression changes can accurately group patients based on transcriptional profiles. AR activity for each patient was calculated using AR-related gene signatures from MSigDB with GSVA. The AR activity between the KMT2D high and low patient groups was compared using a t-test to determine statistical significance. Survival differences between the KMT2D high and low groups were analyzed using the R package survival.

For the effect of KMT2D in SCL subtypes, we used a similar methods wtih AR subtypes, with upregulated and downregulated genes identified upon KMT2D knockdown in MSKPCA3 organoids as signatures. SCL patient subtypes were grouped into KMT2D high and low groups accordingly.

Gene set enrichment analysis (GSEA)

MSigDB was used to evaluate GSEAs towards Hallmark datasets using ClusterProfiler v3.16.0. Briefly, we specified the significance of a gene using a Benjamini Hochberg adjusted p-value (padj) computed from the p.adjust function in the R package. Based on this, GSEA plots showing the top upregulated and downregulated genes were ranked according to their normalized enrichment score. For DU145 and PC3 RNA-seq, GSEA was performed using DESeq2-normalized expression values against the MSigDB pathway database.

Single-cell RNA-seq data preprocessing

Single-cell RNA-seq data preprocessing was performed following Seurat v4 pipeline with custom modifications (35). We first merged raw counts from KMT2D knockdown (shKMT2D) and control samples and removed cells with < 200 and > 6000 genes detected, and with more than 20% mitochondrial reads. We then applied Scrublet (36) with default parameters to remove additional putative doublet cells. Combined, these procedures filtered ~10% of all cells. We subsequently iteratively applied SCTransformv2 (37) to normalize the combined dataset. First, we used % mitochondrial reads as the covariate, and estimated S and G2/M cell cycle scores using the (41) function from Seurat. Next, we applied SCTransformv2 again using % mitochondrial reads and cell cycle score difference as covariates as the normalized count, followed by PCA and 2-D projection UMAP using 30 principal components with the default Seurat workflow as a basis for subsequent analyses.

Single-cell RNA-seq analysis

To cluster single cells based on normalized RNA counts, we used Leiden clustering with resolution 0.5, 1.0, and 1.2 to derive initial clusters. We used silhouette information as a metric to define optimal clustering resolution. For each cluster, we counted the number of cells belonging to shKMT2D and control groups and used the fraction of cells in each group as the metric of enrichment. To annotate each cluster by putative cell types, we obtained marker genes from a recent single-cell study of normal human prostate (38), and tested for enrichment of each cell type in each cluster by the expression of marker genes relative to random control gene sets with matched overall expression levels, using AddModuleScore function from Seurat with default parameters. We similarly obtained marker genes for androgen response (AR) from (39) and estimated AR enrichment in each cluster, and the list of markers for subtypes of castration-resistant prostate cancer (CRPC) calculated from bulk RNA-seq of patient samples (4) to estimate enrichment of each CRPC subtype in each cluster. In each enrichment analysis, we compared the distribution of enrichment score between shKMT2D and control cells using a two-sided Wilcoxon rank-sum test. Finally, we defined marker genes for each cluster using FindMarkers function in Seurat with a negative binomial generalized linear model, requiring genes to be detected in at least 25% of cells with log fold-change >= 0.1, followed by fast GSEA (FGSEA) (BioRxiv 2021.02.01.060012) to estimate pathway enrichment using data from the MSigDB hallmark and canonical pathway sets.

Single-cell ATAC-seq analysis

For ATAC-seq analysis, we followed the default ArchR pipeline (40). Briefly, we removed doublet cells and iteratively estimated LSI embedding using 500bp peaks tiled across the genome, followed by ArchR clustering (resolution = 0.8) and UMAP projection workflows to generate ATAC-seq cell clusters and two-dimensional embeddings. We assessed clustering consistency between single-cell RNA-seq and ATAC-seq data using the confusion matrix. We mapped RNA-defined clusters onto single-cell ATAC-seq data, which was used to define pseudobulk replicates for peak calling with MACS2, followed by marker peak identification using the getMarkerFeatures function in ArchR. To obtain transcription factors enriched in shKMT2D and control cells, we similarly applied getMarkerFeatures to obtain list of marker peaks for each condition, and annotated motif enrichment for all transcription factors using the addMotifAnnotations function with the CisBP database (41). We defined enriched transcription factors as FDR adjusted p value<0.1 and log fold change>0.5. For top enriched transcription factors in either condition, we mapped their expression in each cell from paired single-cell RNA-seq data and estimated their mean expression level and fraction of cells with detectable expressions. Subsequently, we quantified their activity using chromVAR (42) and compared the distribution of transcription factor activity and expression between shKMT2D and control cells using a two-sided Wilcoxon rank-sum test similarly as above. Finally, we visualized expression and activity simultaneously for the top TFs on the same UMAP embeddings defined from single-cell ATAC-seq data.

Data and Materials Availability

Bulk sequencing and single-cell sequencing data have been deposited with the Gene Expression Omnibus under GSE296270 and GSE291282, respectively. All other raw data generated in this study are available upon request from the corresponding authors. The data analyzed in this study were also obtained from TCGA and SU2C (http://cbioportal.org). We additionally utilized publicly available AR ChIP-seq (GEO: GSE28264) and MSKPCA3 ChIP-seq datasets (GEO: GSE196732).

Results

KMT2D regulates AR transcriptional output and AR-dependent growth in prostate cancer

Given our previous work in breast cancer, we inquired whether KMT2D is also an AR modulator in prostate cancer. Important regulatory AR-PI3K crosstalk also occurs in prostate cancer (14), but the chromatin-based mechanisms leading to the activation of AR upon PI3K/AKT inhibition are unknown. To this end, we first studied the impact of PI3K inhibition on chromatin accessibility of the prostate cancer epigenome, by performing ATAC-seq (assay for transposase-accessible chromatin followed by sequencing) in PTEN-null/AR-dependent LNCaP cells upon treatment with the pan-PI3K inhibitor GDC0941. ATAC-seq identified robust changes in chromatin state upon PI3K inhibition, with the majority of sites gaining accessibility (Fig. 1A, S1A) and displaying motif enrichment for AR, and the AR cofactor FOXA1 by Homer and gkm-SVM motif analyses (Fig. 1B, S1B–C) (30). Indeed, we observed a marked occupancy of AR and FOXA1 binding in the differentially accessible sites upon PI3K inhibition (Fig. S1D). Accessibility at AR target gene loci such as KLK2, KLK3 encoding for PSA, and NKX3–1 is markedly increased upon PI3K inhibition (Fig. 1C, S1E). Consistently, AR binding at AR loci is enhanced upon combined treatment with DHT and GDC0941 (Fig. 1D). Altogether these data suggest that PI3K inhibition promotes AR activity by enhancing chromatin accessibility at AR target sites and subsequent AR chromatin binding.

We next performed RNA-seq assays in LNCaP cells that were androgen deprived for 72 hours, followed by treatment with DMSO control, DHT, GDC0941 or the combination, with and without KMT2D knockdown (Fig. S1F). Treatment with DHT, PI3K inhibitor, or the combination led to marked changes in transcription compared to the DMSO control (Fig. S1G). Gene set enrichment analysis (GSEA) identified androgen response among the top 10 gene signatures upregulated upon DHT and PI3K inhibition (Supplementary Table 1). Notably, knockdown of KMT2D led to altered transcription of thousands of genes upon DHT, PI3K inhibition, or the combination (Fig. S1H). Out of 858 genes that were either downregulated or upregulated by androgen, 543 were no longer significantly differentially regulated by KMT2D knockdown in LNCaP cells (Fig. 1E), indicating that more than 60% of all androgen-regulated genes require KMT2D for androgen regulation. KMT2D knockdown also significantly affected the PI3K inhibitor-induced transcription (Fig. 1E). Consistently, in the basal or PI3K inhibition setting, androgen response was one of the top gene signatures negatively regulated by loss of KMT2D (Fig. 1F–G and Fig. S1I, Supplementary Table 2). Canonical AR target genes like KLK3, KLK2 and KLK4 which were upregulated upon DHT or PI3K inhibition, were downregulated upon KMT2D knockdown (Fig. S1J). AR gene signatures or AR-target genes were also downregulated when KMT2D was knocked down in a human prostate cancer organoid model (MSKPCA2) (19), or in 22Rv1 and VCAP cells (Fig. S2A–H), suggesting that the role of KMT2D is conserved in AR-high/PTEN-WT contexts, as well.

Lastly, stratification of AR+ clinical samples from TCGA or Stand Up to Cancer (SU2C) cohorts into KMT2D-low or KMT2D-high based on their mRNA expression levels, demonstrated that KMT2D-low patients had lower AR activity compared to KMT2D-high patients (Fig. 1H, S2I). Altogether these data demonstrate that KMT2D knockdown negatively affects androgen activity in prostate cancer across models.

Given our finding that KMT2D regulates androgen response upon PI3K inhibition in prostate cancer, we next sought to determine how PI3K inhibition influences KMT2D binding to the chromatin. KMT2D binding by ChIP-seq at all sites, AR or FOXA1 sites showed a marked increase upon PI3K inhibition, confirming that the phosphorylation of KMT2D by AKT/SGK affects KMT2D recruitment at the chromatin. (Fig. 1I, S3A). Phosphorylated KMT2D at S1331 was also detected by western blot when we immunoprecipitated wild-type KMT2D but not the non-phosphorylatable S1331A mutant KMT2D in LNCaP cells (Fig. S3B–C).

In the setting of KMT2D loss, we next examined the chromatin recruitment of FOXA1 and AR, and the presence of mono and dimethylated histone H3 lysine 4 (H3K4me1/2), histone modifications that are catalyzed by KMT2D and are associated with FOXA1 binding at cis-regulatory elements (43). Genome-wide analysis in LNCaP cells revealed that KMT2D knockdown markedly affected H3K4me1 enriched regions (Fig. 1J), which was corroborated by a reduction in global H3K4me1 levels (Fig. S3D). KMT2D depletion also resulted in the loss of global FOXA1 binding sites as well as those shared with AR and a decrease in H3K4me2 occupancy at AR loci (Fig. 1K, S3E). Notably, KMT2D knockdown also affected the binding of AR upon DHT stimulation alone and in combination with PI3K inhibition (Fig. 1L). We validated these findings by performing ChIP-qPCR in LNCaP cells to examine AR binding at the NDRG1 promoter and KLK3 enhancer (Fig. S3F). Altogether this data demonstrates that KMT2D is a key determinant of the recruitment of the AR-FOXA1 transcriptional regulatory complex at the chromatin.

Since KMT2D knockdown suppresses AR activation, we hypothesized that loss of function of KMT2D would negatively affect the proliferation of LNCaP cells and augment the therapeutic activity of PI3K/AKT inhibitors and the anti-AR inhibitor, enzalutamide. Indeed, cell proliferation assays demonstrated that KMT2D knockdown decreased cell viability of LNCaP cells and had a greater effect in combination with PI3K inhibitors, (p110β inhibitor AZD6482, pan-PI3K inhibitor GDC0941), AKT inhibitors (GDC0068, MK2206, capivasertib), and enzalutamide (Fig. 1M–N, S3G–L).

KMT2D loss in 22Rv1 cells (Fig. S3M) and PTEN null patient-derived organoids, MSKPCA2 (Fig. 1O) also reduced cell proliferation and augmented the activity of the PI3K inhibitor GDC0941. In addition, we built a gene signature of KMT2D activity from RNA-seq of LNCaP cells and probed the association between KMT2D activity and overall survival of AR-dependent patient samples from SU2C cohort. Notably, AR-dependent SU2C patients with low KMT2D activity display better survival outcomes in comparison with patients who have high transcriptional activity of KMT2D (Fig. 1P). In addition, the KMT2D low activity score was also associated with a downregulated androgen response signature in AR-Dependent SU2C patients (Fig. S3N). Altogether, our findings in cells, patient derived organoids, and clinical samples demonstrate that KMT2D is a key determinant of AR transcription and that its silencing in preclinical models decreases proliferation and increases the antitumor activity of PI3K/AKT inhibitors in AR-dependent prostate cancer.

KMT2D regulates AP-1/AR gene expression programs and chromatin accessibility at AP-1/FOXA1 loci in CRPC-SCL

Our findings demonstrate that KMT2D establishes the chromatin competence necessary for the recruitment of AR and FOXA1 to activate AR-dependent transcription in AR-high/PTEN null prostate cancer models. In CRPC, the loss of AR dependence leads to clinically aggressive tumors with limited therapeutic opportunities. To this end, we examined the role of KMT2D in PTEN-deficient CRPC tumors with low AR expression and transcriptional output, focusing on CRPC-SCL. We prioritized CRPC-SCL because it is the most common AR-independent CRPC subtype and the most heterogeneous, including models with low or absent AR (4). To investigate this subtype, we used human prostate cancer organoid models (MSKPCA3 and MSKPCA12) derived from metastatic tumor biopsies of PTEN-null, AR-low patients, along with two AR-independent cell lines, DU145 and PC3, classified as CRPC-SCL (4). Successful knock down of KMT2D for the patient-derived organoids is shown in Fig. S4A–B. To study the impact of KMT2D on global chromatin accessibility and underlying transcription factor (TF) dynamics, we first employed ATAC-seq to investigate the chromatin landscape changes upon KMT2D knockdown and treatment with the PI3K inhibitor GDC0941 in MSKPCA3 organoids. We obtained a total atlas of 56,463 regions with 2,482 displaying significantly decreased or increased accessibility after KMT2D knockdown (FDR<0.05), and the majority (1,665 sites) losing accessibility (Fig. 2A, Fig S4C). To define the TF motifs associated with lost or gained accessible sites after KMT2D knockdown, we performed differential motif analysis of TFs that are expressed in prostate cancer (FDR<0.05). Among the TFs whose binding is predicted to decrease, we identified the pioneer factor and AR cofactor FOXA1; AR target gene RUNX1; and the nuclear receptor motif (NRE, RARA) (Fig. 2B). Indeed, a significant number of predicted FOXA1 binding sites are closed upon KMT2D knockdown (Fig. 2C). Interestingly, members of the AP-1 TF complex (FOS, JUNB), a defining feature of CRPC-SCL (4), and TFs like TEAD were the most enriched at regions that lose accessibility upon KMT2D knockdown (Fig. 2C). Integrating the MSKPCA3 ATAC-seq dataset with MSKPCA3 ChIP-seq data available for FOSL1, TAZ, TEAD and YAP from Tang et al, 2022 (4) (GEO: GSE196732) showed binding of these TFs at chromatin regions that lose accessibility with KMT2D knockdown, suggesting that KMT2D may regulate the activity of these TFs (Fig. S4D–E).

Figure 2. KMT2D regulates AP-1/AR gene expression programs and chromatin accessibility at AP-1/FOXA1 loci in AR-low/PTEN-null prostate cancer.

(A) ATAC-seq heatmap displaying differentially accessible sites in the patient-derived organoid model, MSKPCA3 after KMT2D knockdown (n = 3 biological replicates) (q value<0.05, n = 2482).

(B) Top enriched TF motifs with lost or gained accessibility upon KMT2D knockdown (ridge regression, FDR<0.01).

(C) Heatmap of differentially accessible sites upon KMT2D knockdown based on each TF motif (q value<0.05, n = 4492).

(D)-(E) ATAC-seq tornado plots of DU145 and PC3 cells upon KMT2D knockdown (n = 3 biological replicates).

(F)-(G) Top enriched de novo motifs at sites that lose accessibility upon KMT2D knockdown in DU145 and PC3 cells.

(H) FOSL1 ChIP-seq tornado plot in DU145 cells upon KMT2D knockdown (n = 3 biological replicates).

(I) Top motifs at sites that lose accessibility upon KMT2D knockdown in DU145 cells (FDR<0.05).

(J)-(M) GSEA of androgen response and AP-1 C3 motif gene sets in MSKPCA3, DU145 and PC3 after KMT2D knockdown (Nominal p values and FDR adjusted p values, GSEA package).

(N)-(P) RNA-seq expression of AP-1 complex members and YAP/TAZ target genes downregulated upon KMT2D knockdown in patient-derived MSKPCA3 organoids, DU145 and PC3 cells (p adj < 0.05).

(Q) AP-1 signatures linked to KMT2D expression (high or low) in TCGA and SU2C patient cohorts (p value, student’s t-test).

(R) AP-1 signatures linked to KMT2D transcriptional activity derived from MSKPCA3 RNA-seq changes upon KMT2D knockdown, in AR-independent SU2C patient cohort (p value, student’s t-test).

NES, normalized enrichment score. FDR, false discovery rate.

In addition to MSKPCA3, we successfully knocked down KMT2D in other established CRPC-SCL models such as DU145 (AR-neg) and PC3 (AR-neg) cell lines (Fig. S5A, S5F). Both models also exhibited a marked decrease in bulk H3K4me1 levels with KMT2D knockdown (Fig. S5B, S5G). We further assessed changes in the chromatin landscape by performing ATAC-seq which showed significant differential changes in accessibility with KMT2D knockdown in both models, with the majority of the sites losing accessibility (8,303 DU145; 22,019 PC3) (Fig. 2D–E). Consistent with the MSKPCA3 results, AP-1, TEAD and forkhead TF motifs were identified at sites that lost accessibility with KMT2D knockdown (Fig. 2F–G). Notably, FOSL1, a pioneering factor and master regulator of the CRPC-SCL subtype (4) showed reduced occupancy upon KMT2D knockdown, accompanied by an enrichment of AP-1 and forkhead TF motifs at the lost bound sites, consistent with the ATAC-seq results (Fig. 2H–I).

We next subjected control and KMT2D knockdown MSKPCA3 organoids treated with DMSO or PI3K inhibitor and the DU145 and PC3 cell lines with KMT2D knockdown to RNA-seq analyses. KMT2D knockdown led to widespread changes in the transcriptome of prostate cancer organoids (6,610 genes, q value <0.05) and cell lines (3,631 genes, q value <0.05, DU145; 2,265 genes, q value <0.05, PC3) (Fig. S4F, S5C, S5H) demonstrating a critical role for KMT2D in gene regulation in these models. Of the genes that were upregulated or downregulated by PI3K inhibitor treatment, 30% were affected by KMT2D knockdown, suggesting that KMT2D knockdown is affecting PI3K-dependent gene expression in this setting as well (Fig. S4G–I). GSEA identified the androgen response gene signature as significantly downregulated by KMT2D loss in MSKPCA3, DU45 and PC3, along with other signatures that are enriched in the stem cell like state (4) such as TGF-beta, TNF-alpha and JAK/STAT signaling pathway (20), epithelial to mesenchymal transition, inflammatory and interferon response pathways (Fig. 2J, S4J, S5D, S5I). Notably, in line with the ATAC-seq data, GSEA also identified multiple members of the AP-1 TF family to be downregulated upon knockdown of KMT2D (Fig. 2K–M, S4K, S5E, S5J). Dissecting this further, we observed that knockdown of KMT2D led to a downregulation of several YAP/TAZ target genes (AJUBA, ANKRD1, CCND1, AXL) and AP-1 TF such as FOSL1 which has been shown to regulate itself in a positive feedback loop (4) (Fig. 2N–P). GSEA also revealed downregulation of the gene signatures for YAP1 and WWTR1/TAZ and AP-1 TF family in MSKPCA3 (Fig. S4L). Similarly, downregulation of AP-1 and YAP/TAZ target genes was also observed upon KMT2D knockdown in the SCL cell lines, DU145 and PC3 (S5K–L). Moreover, using prostate cancer patient samples from the TCGA and SU2C, we found that KMT2D-low patients had lower AP-1 transcriptional activity compared to KMT2D-high patients (Fig. 2Q). We also constructed a gene signature reflecting KMT2D-dependent transcriptional activity using RNA-seq from MSKPCA3 organoids. Subsequently, we investigated the association between KMT2D activity and AP-1 and YAP-TAZ gene signatures, derived from Tang et al (4), within the same model. Our findings indicate a significant association between high KMT2D activity and elevated levels of these gene signatures in CRPC-SCL patients from the SU2C cohort (Fig. 2R). Altogether, these data across cell lines, patient-derived organoids and patient samples suggest that KMT2D plays a key role in regulating the androgen response and the AP-1 transcriptional program which has recently been identified (4) as a defining feature of the CRPC-SCL subtype.

Single-cell chromatin and transcriptomic profiling reveal KMT2D’s role in sustaining the mixed-lineage cell state of CRPC-SCL

The transition from an AR-dependent to an AR-independent phenotype, in CRPC-SCL may result in tumors with mixed features and considerable tumor heterogeneity. Our bulk RNA-seq and ATAC-seq data suggest for the first time that CRPC-SCL cell states may be epigenetically controlled by KMT2D, which maintains their gene expression repertoire and unique master TF profiles. To further investigate the phenotypic outcomes of KMT2D loss, we performed single-cell RNA-seq (scRNA-seq) and scATAC-seq analyses using transcriptome and open chromatin profiling within the same nucleus (multiome) in MSKPCA3 patient-derived organoids upon KMT2D knockdown at 7 days post-doxycycline administration. In total 17,965 cells were analyzed using this multiomics approach (Fig. S6A). Transcriptomes were filtered by Seurat (Methods, (35) and visualized by UMAP. Unsupervised clustering identified 10 distinct transcriptional states within the whole cell population (Fig. 3A, see Methods). First, we looked at the distribution of control and KMT2D knockdown cells among these clusters (Fig. 3B–C). One of the clusters (#1) was predominantly composed of KMT2D knockdown cells, whereas clusters 2, 3, 4, 5 and 8 were composed of mainly control cells (Fig. 3B–C). Additionally, we identified several smaller cell clusters (6, 7, 9 and 10) composed of cells from both experimental groups (1,752 cells out of a total of 17,965, ~9.7%) (Fig. 3B–C, S6A). We noted that the shared clusters contained signatures of cell cycle and E2F target genes, indicating that they were proliferating cells (Fig. S6B).

To elucidate the effects of KMT2D loss we focused our further analyses on clusters composed of cells of predominantly one experimental state, which was the vast majority of the cells (90%). In an effort to identify the phenotypic state of the organoids upon shKMT2D knockdown, we integrated the gene signatures used to identify normal epithelial cell types in the prostate (38). In addition to the basal, luminal, club and hillock gene signatures (38), we applied an AR gene signature (39) as MSKPCA3 organoids still express AR, albeit at reduced levels. In control cells, signatures of all normal epithelial cell types could be identified, in line with the mixed-lineage state of CRPC-SCL (Fig. 3D). In KMT2D knockdown cells, correlation to nearly all of these transcriptional signatures apart from basal cells was markedly reduced, pointing to a key role for KMT2D in regulating the mixed-lineage state of CRPC-SCL (Fig. 3D–F). Moreover, there was a loss of AR signature, indicating that KMT2D also regulates AR transcriptional output in an AR-low expression setting (Fig. 3D–G). Thus, in cells with KMT2D loss there was a significant loss of luminal identity.

We also applied transcriptomic signatures of the CRPC-SCL subtype generated by Tang et al. 2022 (4). As expected, control cells strongly associated with the signature CRPC-SCL subtype, to which MSKPCA3 is assigned. Strikingly, we noted a distinct loss of the CRPC-SCL signature in shKMT2D knockdown cells suggesting a significant role for epigenetic control by KMT2D in the maintenance of SCL (Fig. 3D–F, S6C). In addition, we also detected a stronger correlation to transcriptional signatures associated with MYC activity in cells with KMT2D knockdown (Fig. S6D–G).

Given the heterogeneity and the mixed lineage state of CRPC-SCL, we also evaluated the expression of lineage specific markers by Western blotting (Fig. S7A) and immunostaining via multiplex immunofluorescence (IF) (Fig. S7B, Fig. 3H) in CRPC-SCL organoid models (MSKPCA3, MSKPCA12, LuCAP176) and cell lines (DU145, PC3) (Fig. S7B–C, Fig.S8A–B). Additionally, we also conducted immunohistochemistry (IHC) on all the patient-derived organoid models (Fig. S9A). IHC and multiplex IF were also performed to evaluate KMT2D expression. KMT2D displayed consistent nuclear staining across all models, with no significant differences in expression between cells of different subtypes/lineages (Fig. 3H, Fig. S7B–C, S8A–B, S9A). In contrast, lineage markers CK8, CK5 and Vimentin exhibited greater heterogeneity in all CRPC-SCL models (Fig. 3H, Fig. S7A–C, S8A–B, S9A). In the AR-positive CRPC-SCL models LuCAP176, MSKPCA3 and MSKPCA12, some variability in AR expression was observed as expressed. No synaptophysin, a NE marker, was detected in any of the SCL models, while a control CRPC-NE model LuCAP49 tested positive (Fig. S7A, S9A).

To identify how the chromatin is affected by KMT2D loss, we next turned our attention to the scATAC data. A total of 16,621 cells were analyzed (Fig. S10A–B). In concordance with the scRNA-seq data the control and shKMT2D cells separated well based on experimental manipulation. Unsupervised clustering defined 10 clusters (see Methods), with 3 distinct clusters in the shKMT2D cells, 5 in the control cells and 2 small clusters composed of both experimental states (Fig. 4A–B, S10C). These clusters showed good concordance with the clusters identified in the scRNA-seq set (Fig. S10D–E). To identify potential TFs affected by the loss of KMT2D, we assessed the enrichment of TF DNA binding motifs in the differentially accessible chromatin in the knockdown setting and control cells using chromVAR (42). Additionally, we assessed the RNA expression of TFs with enriched motifs. Using this approach, we identified high activity of AP-1 in the control setting (cluster 2,4,6), but also a cell population with high FOXA1 activity (cluster 5) (Fig. 4C–E, S10F). Strikingly, the accessibility for both FOSL1 and FOXA1 was significantly lost upon shKMT2D knockdown (Fig. 4F–G, S10G–J). Instead, in line with the scRNA seq, we saw increased accessibility of the master regulator of basal cell fate, TP63 (cluster 9) (Fig. 4C–D, S10G). Additionally, we observed clusters marked by TCF4, TCF12 and ID4 accessibility and expression (cluster 9,10) (Fig. 4C–D). Given that FOXA1 and FOXA2 have identical motifs and that recent elegant work has shown that FOXA2 can rewire AP-1 activity in prostate cancer, (44), we assessed the expression of FOXA1 and FOXA2 in established CRPC-SCL patient-derived organoids such as MSKPCA3, LuCAP176, MSKPCA12 and cell lines such as DU145 and PC3. FOXA2 expression was nearly absent in the patient-derived organoid models and cell lines (Fig. S10K–M), suggesting that FOXA1 may be the key TF to drive mixed lineage identity of the SCL subtype. Together, this data suggests that KMT2D is a key mediator of subtype-specific chromatin accessibility and cooperates with different TFs that are required for maintaining the mixed-lineage state in CRPC-SCL. In addition, KMT2D sustains the luminal identity present in MSKPCA3 organoids.

Figure 4. Single-cell chromatin and transcriptomic profiling reveal a key role for KMT2D in the maintenance of the mixed-lineage cell state in CRPC-SCL.

(A)-(B) UMAP of single-cell ATAC-seq (scATAC-seq) colored by cluster or experimental condition (control, shKMT2D) (DOX, 500ng/ml, 7 days).

(C) Heatmap of TFs enriched in single-cell ATAC-seq analysis (-log10 Bonferroni-adjusted p value using the CisBP database).

(D) Scatter plot showing TF motif enrichment and ranks in differentially expressed peaks in control and shKMT2D cells (-log10 Bonferroni-adjusted p value). Colors indicate the fraction of cells expressing each TF.

(E) Violin plots of differential TF motif enrichment in control and shKMT2D cells (chromVAR).

(F) (G) Example UMAP of FOXA1 and FOSL1 motif deviation (chromVAR z-score using single-cell ATAC-seq) and density of normalized expression profiles (single-cell RNA-seq). UMAP coordinates were derived from single-cell ATAC-seq analysis.

(H)-(I) Proliferation assays of patient-derived organoid models MSKPCA3 (DOX, 500ng/ml), MSKPCA12 DOX, 200ng/ml) ± GDC0941 (1μM)/ Capivasertib (0.5μM) (7 days, n = 3 technical replicates, mean, SD). (p value, student’s t-test).

(J) Proliferation assays of PC3 cells with KMT2D knockdown (DOX, 1μg/ml) (n = 3 technical replicates, mean, SD) (p value, student’s t-test).

(K)-(L) Prestoblue viability curves of PC3 cells upon KMT2D knockdown (in blue, DOX 1μg/ml) and upon treatment with GDC0941 and AZD5363 (capivasertib) (5 days, n =4 biological replicates, representative shown, mean, SD).

(M)-(N) PC3-derived xenograft study with DOX-inducible shKMT2D ± Capivasertib (P.O., BID). Tumor measurements were made over indicated time and weights were taken (p value, two-sided Mann-Whitney U test) (n = 5 per group, mean, SEM).

(O) qPCR of AP-1 and YAP/TAZ target genes in PC3-derived xenografts ± KMT2D knockdown (using induced by DOX-feed, expression normalized to GAPDH). (p value, student’s t-test) (n = 4 biological replicates, mean, SEM).

Given the crucial role of KMT2D in subtype-specific chromatin accessibility and gene expression in CRPC-SCL, we also hypothesized that KMT2D could affect the cell viability of MSKPCA3 organoids. Proliferation assays demonstrated that KMT2D knockdown reduced the proliferation and augmented the activity of the PI3K inhibitor GDC0941 in the MSKPCA3, MKSPCA12 organoids (Fig. 4H–I, S10N), as well as capivasertib in the MSKPCA12 organoids (Fig. 4I). Moreover, KMT2D loss also impaired basal proliferation of PC3 cells and increased the activity of PI3K/AKT inhibitors (Fig. 4J–L). We next investigated the in vivo activity of capivasertib in combination with KMT2D knockdown by injecting PC3 human prostate cancer cells with doxycycline-inducible KMT2D shRNAs into nude mice. KMT2D silencing by doxycycline diet and capivasertib alone had a significant effect on the growth of tumor xenografts (Fig. 4M–N). In addition, KMT2D knockdown significantly increased the antitumor activity of capivasertib in vivo, with the combination of KMT2D loss and capivasertib providing more significant tumor regression in this model. We also confirmed the knockdown of KMT2D upon doxycycline treatment (Fig. S5M). Furthermore, in these tumors the expression of AP-1 and YAP/TAZ target genes was significantly downregulated upon loss of KMT2D (Fig. 4O).

In summary, our findings unveil KMT2D as a master regulator, intricately shaping subtype-specific chromatin accessibility and the transcriptomic landscape and cell states in CRPC-SCL, required for prostate cancer growth and therapeutic response. These results lay the groundwork for potential targeted interventions using epigenetic therapy in metastatic prostate cancer (Fig. 5).

Figure 5.

Proposed Model

Discussion

In this study, we dissected the contribution of the histone methyltransferase KMT2D in shaping chromatin accessibility, transcription factor regulation, gene expression and the dynamic functional phenotypes in distinct stages of PCa. Our efforts to probe the mechanistic contributions of KMT2D in PCa were imotivated by findings linking extensive changes in chromatin landscape with tumor cell state or lineage (4,45), our group’s identifying KMT2D as mediator of nuclear receptor function at cell-specific enhancers (8,13,46), and the pressing need for effective combinatorial therapies to improve prostate cancer treatment.

We identified a key role for KMT2D in AR-dependent PCa. In this setting, KMT2D promotes the chromatin landscape required for the recruitment of AR and FOXA1 to dictate AR-dependent transcription and prostate cancer cell proliferation. Therefore, the therapeutic efficacy of PI3K/AKT inhibitors in AR-high, PTEN-null prostate cancer is enhanced upon KMT2D knockdown. We demonstrate that KMT2D is a common mechanism controlling nuclear receptor function and nuclear receptor-PI3K pathway crosstalk at cell-specific enhancers in breast and prostate cancer through direct phosphorylation of KMT2D by effectors of the PI3K pathway, the AKT/SGK kinases (47).

We also report the discovery and characterization of an unexpected role for KMT2D in a later stage of prostate cancer evolution which is characterized by AR independence and a mixed-lineage state, CRPC-SCL (4). Recent findings suggested that AP-1 TFs such as FOSL1 and TEAD function together to drive oncogenic growth in CRPC-SCL samples (4). Our single-cell RNA-seq and single-cell ATAC-seq in patient-derived prostate cancer organoids (CRPC-SCL, AR-low) upon knockdown of KMT2D, enabled an in-depth study of the cell states and master TFs potentially regulated by KMT2D. Our analysis identified that the mixed-lineage state of CRPC, which consisted of luminal, and SCL subpopulations together with androgen response was predominantly enriched in KMT2D-intact organoids. The recently reported SCL signature (4) and established luminal and androgen signatures were mostly absent in the KMT2D loss setting with a number of AP-1 TFs like FOSL1, FOS, JUNB, and the FOXA1 TF downregulated with KMT2D loss. These data suggest that KMT2D, by regulating luminal-defining TFs like AR and FOXA1, as well as SCL-defining AP-1 transcription factors like FOSL1, plays a crucial role in preserving the mixed-lineage state in CRPC-SCL and potentially drives lineage plasticity when dysregulated. While our analysis focused on FOSL1 as the dominant AP-1 family member in CRPC-SCL (4), other AP-1 family members may also cooperate in enhancer driven reprogramming and thus represent an avenue for future studies. In AR-negative CRPC-SCL models, KMT2D specifically controls the AP-1 TF family activity, underscoring its broader role in modulating lineage-defining transcriptional networks. Future investigations should delineate the biochemical interactions between lineage-defining TFs and KMT2D across distinct stages or subtypes of CRPC including CRPC-NE and the rare CRPC-WNT. Such endeavors are imperative to unravel the mechanisms for subtype-specific gene expression and the maintenance of cellular states across subtypes.

In alignment with previous studies recognizing epigenetic regulators such as LSD1 (48), EZH2 (45,49,50), BRD4 (50), and DNMTs (51,52) as critical mediators of oncogenesis in prostate cancer, our findings underscore the importance of chromatin-based mechanisms as drivers of lineage plasticity in hormone-driven cancers. For instance, augmented activity of epigenetic modifiers such as EZH2 is observed in prostate cancer, fueling cellular reprogramming and consequent lineage plasticity (45). In addition, genomic sequencing efforts have revealed that 20% of prostate cancer samples harbor mutations, in epigenetic modifiers or chromatin remodeling genes such as KMT2D, KMT2C and genes coding for the SWI/SNF complex. These alterations, enriched in metastatic disease (53,54), may contribute to therapy resistance but remain poorly understood. Their large size - KMT2D for instance is ~600 kDa - makes biochemical characterization technically challenging and underexplored. Given that AR, the major driver of prostate cancer, along with the newly identified subtype-specific TFs, often function in conjunction with various chromatin remodelers, it is of importance to functionally characterize the role that epigenetic regulators play in prostate cancer. Understanding the chromatin-based mechanisms that instigate changes in tumor cell state, from tumorigenesis to the emergence of therapy resistance, is imperative in prostate cancer research for the discovery of new targets.

Monotherapy with anti-AR therapies is currently the standard of care for CRPC and CRPCs which lose AR are left with limited to no therapeutic options. Combinatorial therapies are not yet standard of care for CRPC, and few options are currently in clinical trials including an AKT inhibitor in combination with anti-AR therapy (NCT04493853). The recent FDA approval of the AKT inhibitor capivasertib for the treatment of metastatic ER+ breast cancers bearing mutations in the PI3K pathway (PIK3CA, PTEN, AKT) ignites hope for PI3K-driven CRPCs. As this is the first drug approval for patients with PTEN mutant tumors, it will be critical to leverage this advancement for addressing other cancer types harboring these mutations, such as prostate cancer. Lastly, the discovery of the specific role of KMT2D in the crosstalk between critical survival pathways in prostate cancer and the regulation of subtype-specific TFs provides a rationale for epigenetically informed combination therapies, perhaps small molecule inhibitors targeting KMT2D in combination with PI3K/AKT in PTEN-deficient CRPC-AR and CRPC-SCL. These findings increase the therapeutic potential and opportunities of targeting KMT2D for the treatment of PI3K-driven prostate cancer patients.

Statement of Significance.

KMT2D is a critical regulator of chromatin accessibility and transcriptional landscapes in castration-resistant prostate cancer that drives both AR-dependent and AR-independent subtypes, highlighting KMT2D as a potential therapeutic target.