Reduced Expression of Histone Methyltransferases KMT2C and KMT2D Correlates with Improved Outcome in Pancreatic Ductal Adenocarcinoma

Pancreatic ductal adenocarcinomas (PDAC) make up the majority (>90%) of all pancreatic malignancies and are associated with particularly poor overall survival (1). Patients typically present with invasion and metastases at diagnosis, limiting the opportunities for curative surgical resection. The introduction of next-generation sequencing approaches has accelerated our understanding of the recurring coding mutations present in PDAC (2–6). There appears to be a founder population of cells that have accumulated activating mutations in KRAS (>90%; ref. 6), alongside loss-of-function mutations in TP53 (50%–75%; refs. 7–10) and SMAD4 (∼55%; refs. 10, 11). In addition, a significant number of other recurring copy number changes and mutations targeting components of the epigenome have been identified, including the histone lysine (K) methyltransferases KMT2C (MLL3) and KMT2D (MLL2; refs. 2–6). Intriguingly, these KMT2C and KMT2D mutations appear to identify a group of patients with better outcome relative to those with wild-type configuration (5), suggesting that depletion of these methyltransferases may either define less aggressive forms of PDAC, or serendipitously improve the efficacy of existing therapies, where the mechanisms underlying this effect are not known.

The KMT2 family of histone lysine methyltransferases consists of KMT2A (MLL1/ALL1), KMT2B (MLL2/MLL4), KMT2C (MLL3/HALR), KMT2D (MLL2/ALR/MLL4), KMT2E (MLL5), KMT2F (SET1A), KMT2G (SET1B), and KMT2H (ASH1L; ref. 12). These family members, with the exception of KMT2E and KMT2H, act as catalytic subunits within mammalian COMPASS-like complexes to catalyze the addition of methyl groups to a lysine residue on the amino tail of histone H3 (H3K4; ref. 13). H3K4 exists in unmethylated, monomethylated (H3K4me1), dimethylated (H3K4me2), and trimethylated (H3K4me3) states, where H3K4me1 is typically associated with enhancers and H3K4me3 with promoters (14). These KMT2 complexes appear to have different substrate specificities to catalyze the formation of H3K4me1 (KMT2C and KMT2D; refs. 15, 16), H3K4me1/me2 (KMT2A and KMT2B; refs. 17, 18), and H3K4me1/me2/me3 (KMT2F and KMT2G; ref. 19).

Our focus here is restricted to two of these methyltransferases identified as potential key players in PDAC. Loss of KMT2C and KMT2D in cancer is expected to impact upon gene expression; however, such changes appear to be cell type–dependent, with both negative and positive effects on cell proliferation reported (15, 20–28). We set out to understand how these methyltransferases impact upon PDAC biology, and whether they may present novel opportunities for patient stratification, personalized therapies, or even therapeutic targets.

Human tumor cell lines PANC-1 and Capan-2, and the immortalized human pancreatic ductal epithelial cell line, HPDE, were cultured in DMEM (Sigma Aldrich); BxPC-3, SUIT-2, RWP-1 and COLO 357 in RPMI1640 medium (Sigma Aldrich); and CFPAC-1 cells in Iscove's modified Dulbecco's medium with 25 mmol/L HEPES (Lonza) and 2 mmol/L l-glutamine (Sigma Aldrich). PANC-1, Capan-2, HPDE, BxPC-3, SUIT-2, RWP-1, and CFPAC-1 were obtained from ATCC. All human cell lines were obtained between 2008 and 2012, and authenticated between 2011 and 2016 using small tandem repeat profiling conducted by LGC standards and ATCC. DT6606, DT6586, and TB32034 cell lines, derived from the LSL-Kras^G12D/+;Pdx-1-Cre (KC; ref. 29; DT6585 and DT6606) and LSL-Kras^G12D/+;LSL-Trp53^R172H/+;Pdx-1-Cre (KPC; ref. 30; TB32043) mice models of PDAC, were cultured in DMEM (Sigma Aldrich). DT6606, DT6585, and TB320343 were kindly provided by David Tuveson (Cold Spring Harbor Laboratory). For each cell line, medium was supplemented with 10% heat-inactivated FBS (Life Sciences Solutions), 100 U/mL penicillin, and 100 μg/mL streptomycin (Sigma Aldrich). All cell lines were resuscitated from authentic stocks, cultured for less than two months at 37°C and 5% CO₂, and routinely screened for mycoplasma.

Cells were plated in 6-well plates with 2.5 mL growth medium without penicillin and streptomycin for forward lipid transfection with Silencer Select siRNAs (Life Sciences Solutions) and Lipofectamine RNAiMAX (Life Sciences Solutions; Supplementary Table S1). For human cell transfection, a final concentration of 8.3 nmol/L per well was achieved by combining 25 pmol siRNA with 5 μL Lipofectamine in 500-μL Opti-MEM I medium (Life Sciences Solutions) for 15 minutes, before adding to wells for 48 hours. The protocol was the same for murine cell transfection, with 150 pmol siRNA used instead to achieve a final concentration of 50 nmol/L. In all experiments, loss of KMT2D/Kmt2d or KMT2C/Kmt2c expression was confirmed by Western blot analysis or real-time PCR.

To prepare protein lysates, culture medium was removed, cells were washed twice with Dulbecco's PBS (DPBS), and RIPA buffer [Sigma-Aldrich; supplemented with protease inhibitor cocktail I (Roche) and 1:100 phosphatase inhibitor cocktail II (Sigma-Aldrich)] was added for 30 minutes on ice. Lysates were harvested with cell scrapers and cellular debris removed by 30-minute centrifugation at 16,000 rcf and 4°C. Protein quantification was determined in a bicinchoninic acid (BCA) assay [4% w/v copper (II) sulphate (Sigma-Aldrich) diluted 1:50 in bicinchoninic acid (Sigma-Aldrich)]. Samples were run on NuPAGE Novex 3%–8% Tris-Acetate or 4%–12% Bis-Tris gels (Life Sciences Solutions) and transferred to preactivated polyvinylidene difluoride membranes by iBlot dry transfer (Life Sciences Solutions). Membranes were blocked for an hour with 5% w/v BSA (Sigma-Aldrich), or 5% w/v nonfat dry milk (Marvel), in Tris-buffered saline with Tween 20 (TBST) and probed with primary antibodies diluted in blocking buffer overnight at 4°C. Membranes were washed with TBST, incubated with rabbit or mouse horseradish peroxidase–conjugated secondary antibodies diluted in blocking buffer, for an hour at room temperature, and washed with TBST (Supplementary Table S2 for primary and secondary antibodies). Staining was visualized by incubation with Amersham Enhanced Chemiluminescence (ECL), or ECL prime, Western blotting detection reagents (GE Healthcare) and developed using Super Rx X-ray films (Fujifilm) and a Konica Minolta SRX-101A medical film processor. Equivalence of protein loaded for each sample was confirmed by β-actin staining.

The eight cell lines were seeded in 6-well plates at either 5 × 10⁴ cells/well (SUIT-2) or 8 × 10⁴ cells/well (all other cell lines). After a 48-hour transfection, cells were washed with DPBS and 3 mL of fresh penicillin- and streptomycin-free media were added. Cells were cultured for 72 hours before detachment with trypsin (Sigma Aldrich) and counted using a Vi-Cell XR automated cell viability analyzer (Beckman Coulter).

Cells seeded in triplicate wells of 6-well plates at either 1.5 × 10⁵ cells/well (SUIT-2), or 2 × 10⁵ cells/well (all other cell lines), were transfected for 48 hours. Three wells per treatment were harvested and analyzed across three time points (48 hours after transfection, 16-hour treatment with 400 ng/mL nocodazole, and 24 hours after medium was replaced). Cells were pelleted by 5-minute centrifugation at 16,000 rcf and 4°C and permeabilized with 1 mL ice-cold 70% ethanol. Cells were stored at −80°C before washing with ice-cold DPBS and staining with propidium iodide [PI; 50 μg/mL PI (Sigma-Aldrich) with 100 μg/mL RNase A (Qiagen) in DPBS] for 15 minutes at 37°C. The YG610/20 filter on the Fortessa II flow cytometer was used to examine DNA staining in 30,000 cells.

To validate selective knockdown of KMT2C or KMT2D mRNA by siRNAs, total RNA was isolated from cells, after 48-hour transfection, using the RNeasy Mini Kit with an on-column DNase digestion (Qiagen). Extracted RNA was quantified using a Nanodrop spectrophotometer and reverse transcribed into cDNA using a High-Capacity cDNA Reverse Transcription Kit (Life Sciences Solutions). Duplex real-time PCRs were performed using iTaq Universal Probes Supermix (Bio-Rad) and TaqMan gene expression assays (Life Sciences Solutions) to examine expression of the gene of interest and the 18S ribosomal RNA internal housekeeping standard (Supplementary Table S3). Data were analyzed using a 2^−ΔΔC_t method (31) to examine relative expression of each gene, where mRNA expression levels were normalized to levels of 18S rRNA for the targeted siRNAs, and then expressed relative to normalized mRNA expression levels of the control siRNA treatment.

Material for RNA sequencing (RNA-seq) was generated from cells seeded at either 1 × 10⁵ cells/well (SUIT-2) or 1.6 × 10⁵ cells/well (PANC-1 and COLO 357) in 6-well plates and transfected for 48 hours. Total RNA was isolated (as above), quantified, and integrity measured using a RNA 6000 nano assay kit on a 2100 BioAnalyzer (Agilent). A Ribo-Zero Gold kit (Illumina) was used to deplete rRNA from RNA samples and cDNA libraries were prepared using the TruSeq stranded total RNA library preparation kit (Illumina). Libraries prepared for sequencing were validated on the 2100 BioAnalyzer with a DNA 1000 kit (Agilent) and again after randomizing the samples into seven pools with a high sensitivity DNA analysis kit (Agilent). Paired-end sequences (reads) of 100 bp in length were generated using seven lanes of a HiSeq 2000 (Illumina). RNA-seq data have been deposited in Gene Expression Omnibus (GEO) under the accession number GSE75327 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE75327).

After FASTQ data quality check using FastQC, raw reads were aligned to the reference genome hg19 using Tophat2 (32). An average of 27.2 M aligned paired-end reads (range 19.7–36.4 M), corresponding to an average of 75.0% (range 59.2%–82.2%) concordant pair alignment rate, were reported (Supplementary Table S4). The number of reads uniquely aligned (mapping quality score q > 10) to the exonic region of each gene were counted using HTSeq (33), based on the Ensembl annotation (version 74). KMT2C and KMT2D siRNA datasets were first analyzed independently. Only genes that achieved at least one count per million (CPM) mapped reads in at least three samples were included, leading to 15,912 and 15,818 filtered genes in total for the respective KMT2C and KMT2D siRNA datasets. These genes were classified into 15 RNA species, with protein-coding transcripts representing 81.8% and 82.3%, respectively. Read counts were further normalized using the conditional quantile normalization (cqn) method (34), accounting for gene length and GC content. Differential expression analysis was then performed using the edgeR package (35), employing the generalized linear model (GLM) approach, for each siRNA versus its control pairwise comparison, adjusting for baseline differences between the cell lines, with an additive model design as “model.matrix(∼cellline+siRNA_treatment). ” For each pairwise comparison, the significantly differentially expressed (DE) genes were selected using a double threshold of false discovery rate (FDR) < 0.05 and an absolute fold change of at least two. Common DE genes were then identified between different siRNA versus control comparisons within the KMT2C and KMT2D siRNA datasets. On the basis of log2 fold changes of siRNA treated over control for all filtered genes, gene-set enrichment analysis (GSEA) was performed for each comparison using the GSEA tool to identify the canonical pathways gene sets from the Molecular Signatures Database (MSigDB-C2 v5.0; ref. 36). The gene ontology (GO) biological process (BP) enrichment analysis was also performed for DE genes using the PANTHER classification system (37).

Two large PDAC datasets, the International Cancer Genome Consortium (ICGC; ref. 38) and The Cancer Genome Atlas (TCGA; ref. 3), with both gene expression and clinical follow-up data available, were used to complete survival analysis. Data on 87 patients from the ICGC dataset were previously compiled and processed (39). Level 3 gene expression data for TCGA dataset were downloaded via TCGA data portal (https://tcga-data.nci.nih.gov/tcga/). Only annotated and confirmed PDAC patients were selected (108 in total). RNA-seq by Expectation-Maximization (RSEM) normalized expression data for 20,501 genes were obtained. For each gene, low and high expression groups were determined using the method described previously (40). Briefly, each percentile of expression between lower and upper quartiles was used in the Cox regression analysis and the best performing threshold of percentile was determined. As an example, the selection of high and low expression groups for KMT2C and KMT2D in the ICGC and TCGA datasets are shown (Supplementary Fig. S1A).

Survival modeling and Kaplan–Meier (KM) analysis was undertaken using R statistical environment (“survival” package). Overall survival (OS) was defined as time from diagnosis to death, or to the last follow-up date for survivors. Log-rank test was used to calculate the KM P values. The Cox proportional hazards model was fitted to every gene independently.

Two additional PDAC gene expression profiles (GEP) and clinical follow-up datasets, namely “Stratford” and “BCI_Zhang_merged,” compiled and processed previously (39), were also included for validation studies.

The three murine cell lines (DT6606, DT6585, and TB32043) were seeded in 6-well plates at a density of 4 × 10⁴ cells/well. After 48-hour transfection, cells were washed with DPBS, detached with trypsin, and replated in 96-well plates at 1 × 10⁴ cells/well (DT6585 and TB32043), or 5 × 10³ cells/well (DT6606). After adherence, medium was replaced with DMEM containing different concentrations of 5-fluorouracil (5-FU; Accord Healthcare). After 72-hour incubation, WST-1 reagent (Roche) was added to each well and the optical density (OD) measured at 440 nm (reference 630 nm) after 3 hours. To generate log–dose response curves, percentage of cell viability was calculated using the maximal OD as 100% viability.

Inactivation of KMT2C and KMT2D arises through a combination of gene deletion and/or mutation in PDAC (2–6). To assess whether expression levels of these methyltransferases are also linked to patient outcomes, we used gene expression profile (GEP) data from the ICGC and TCGA patient series to compare clinical features of patients with tumors expressing different levels of KMT2C and KMT2D. In the ICGC dataset, we observed that low levels of KMT2C and KMT2D expression were independently associated with better OS (KMT2D, median 19.9 vs. 11.8 months, log-rank P =0.001, Fig. 1A; KMT2C, median 15.9 vs. 9.2 months, log-rank P = 0.029, Fig. 1B). Combined low-level expression of KMT2C and KMT2D also correlated with longer survival (median 15.9 vs. 9.2 months, log-rank P = 0.044, Supplementary Fig. S1B). A similar trend was also observed in TCGA dataset, for KMT2C and combined KMT2C/D expression; however, these results did not reach statistical significance (Supplementary Fig. S1C). Overall, these observations are consistent with recent genetic data demonstrating that 12 of 101 patients with KMT2C or KMT2D loss-of-function mutations have a superior outcome compared with patients with a wild-type configuration (5).

To determine the cellular effects of depleting KMT2C and KMT2D, three siRNAs were used to silence both methyltransferases in a panel of eight human pancreatic cell lines; three derived from primary tumors (PANC-1, BxPC-3, Capan-2), four from metastatic sites (SUIT-2, RWP-1, CFPAC-1, COLO 357), and one immortalized from the human pancreatic ductal epithelium (HPDE; Supplementary Fig. S2A and S2B). KMT2D silencing by two siRNAs located to different regions of the transcript (exons 39 and 48) resulted in a reduction of KMT2D protein (Fig. 2A) and, consistent with previous studies in medulloblastoma and colorectal cancer cells (15), reduced cell proliferation significantly in all cell lines tested (Fig. 2B). A third siRNA targeting KMT2D produced a less pronounced effect on growth inhibition and was not included in further analyses. KMT2C silencing resulted in significant, albeit less pronounced, reductions in cell proliferation for the three cell lines tested (Supplementary Fig. S2C and S2D).

To examine whether these changes in cell proliferation were accompanied by cell-cycle anomalies, analysis of KMT2D-silenced cells was performed using propidium iodide and a G₂–M blocking agent, nocodazole. In these experiments, cells treated with control siRNA accumulated within the G₂–M phases, contrasting with an observed G₀–G₁ block in KMT2D-depleted cells (Fig. 3A). When the cell-cycle profile of PANC-1 cells was evaluated over several days at 24-hour intervals, KMT2D silencing led to a decrease in the number and proportion of cells in G₀–G₁ with a concomitant increase in the apoptotic fraction (Fig. 3B and Supplementary Fig. S2E). Taken together, these experiments suggest that KMT2D contributes towards normal cell-cycle progression, with loss of this methyltransferase leading to cell-cycle arrest and subsequent apoptosis.

To better understand the mechanism(s) responsible for the cellular effects of KMT2C/D, we assessed the transcriptional changes in response to KMT2C and KMT2D depletion in three PDAC cell lines (PANC-1, SUIT-2, and COLO 357). These cell lines were selected to include a range of background mutations and both primary and metastatic tumors. KMT2C and KMT2D were targeted individually across the three cell lines using siRNAs, with each targeted siRNA resulting in a measurable and specific reduction of KMT2C or KMT2D expression, compared with the scrambled control (Fig. 4A and B). Total RNA was isolated 48 hours after siRNA treatment and transcriptomes were assessed by RNA-seq using 100-bp paired-end deep sequencing (Supplementary Table S4). Resultant RNA-seq profiles confirmed knockdown of both KMT2C and KMT2D methyltransferases in their corresponding experiments (Supplementary Fig. S3A and S3B).

We focussed our attention on differentially expressed protein-coding genes shared between the individual siRNAs and all three PDAC cell lines, noting correlations in log2 fold changes for all quantified genes within the KMT2C and KMT2D siRNA datasets between different siRNAs compared with control (correlation coefficients ranging from 0.39 to 0.60, Supplementary Fig. S3C). As our data support a greater role for KMT2D than KMT2C, in our analysis, we report data in this order. Taking each methyltransferase separately to start, we observed a total of 567 and 759 DE genes for KMT2D siRNAs 1 and 2, respectively, with 124 genes common to both siRNAs and the three lines (Fig. 4C and D; Supplementary Table S5). Fold changes for the DE genes ranged from 0.14 (downregulated, C2orf54) to 24.43 (upregulated, KRT6B), with 40 genes exhibiting a decrease, and 84 an increase, in gene expression. KMT2C silencing resulted in differential expression across 790 genes, with 31 common across all three siRNAs tested (Fig. 4C and D; Supplementary Table S6). For these 31 common DE genes, the fold changes ranged from 0.18 (downregulated, AKR1B10) to 6.51 (upregulated, ANO3), with 27 genes demonstrating an increase, and four a decrease, in expression. There was a striking overlap between DE genes in the KMT2C and KMT2D siRNA datasets, with 19 genes common to both methyltransferases, each demonstrating consistent directional and fold changes between the two methyltransferases (Supplementary Table S6).

Four of these genes were selected for validation based on their previous associations with pancreatic cancer [c-MET (2, 41), Calumenin (42), Claudin-1 (43, 44), and PTPN14 (45)], with Western blot analysis confirming expression changes of each at the protein level (Fig. 4E and Supplementary Fig. S4A). ABCB1, which encodes an ATP-dependent efflux pump and was decreased by KMT2D siRNA treatment, was also included in the validation panel, as its expression has been linked with an increase in drug resistance in pancreatic cancer (46). Decreased expression of ABCB1, identified by RNA-seq to be specific to KMT2D silencing in the two metastatic cell lines, was confirmed by Western blot analysis (Supplementary Fig. S4B). Global levels of H3K4me1/me2/3 remained largely unchanged at 48 hours after KMT2D siRNA transfection in PANC-1 cells; however, after 120 hours, KMT2D depletion was accompanied by a global reduction in H3K4me1/3 and, to a lesser extent, H3K4me2 and total H3 (Supplementary Fig. S4C).

In light of the overlapping data between the transcriptomes of KMT2C- and KMT2D-depleted lines, GSEA was employed to explore the significant differences between targeted and control siRNAs in all curated canonical pathways. In support of our proliferation and cell-cycle studies, six pathways (cell-cycle, cell-cycle mitotic, DNA replication, DNA repair, mitotic M-M/G₁ phases, and Fanconi pathways) were significantly downregulated (FDR < 0.05) in all siRNA experiments compared with controls (Fig. 5A and B). Other pathways relating to cell-cycle checkpoints and apoptosis were also affected, although significance was not reached for all five pairwise comparisons. Three pathways of note (telomere maintenance, meiotic recombination and chromosome maintenance) were highly downregulated in the KMT2D siRNA datasets, but to a lesser extent in the KMT2C siRNA experiments (Fig. 5A and B).

Using the ICGC and TCGA human pancreatic cancer datasets, we next assessed the biological process enrichment of genes that strongly correlated (Pearson correlation P < 0.001) with KMT2C/D expression levels (Supplementary Table S7). To this end, the PANTHER classification system identified a significant enrichment of the cell cycle [adjusted (adj.) P = 1.45e−02 for ICGC dataset; adj. P = 1.02e−06 for TCGA], mitosis (adj. P = 1.62e−03 for TCGA), and DNA repair (adj. P = 1.61e−02 for TCGA) pathways for genes that positively correlated with combined KMT2C/D expression. In addition, translation was found to be the most overrepresented biological pathway for genes, where expression showed significant negative correlation with KMT2C/D (adj. P = 2.21e−04 for ICGC and adj. P = 5.76e−30 for TCGA). The pathways observed to correlate with KMT2C/D expression in human pancreatic cancer datasets were therefore consistent with the most significantly downregulated and upregulated pathways following KMT2C/D knockdown in our PDAC cell lines (Fig. 5A).

Among the 94 and 257 cell-cycle genes (significantly positively correlated with combined KMT2C/D expression) from the ICGC and TCGA datasets, respectively (Supplementary Table S8), we identified three (NCAPD3, CDKL1, and EIF2AK4) as being significantly downregulated in at least one KMT2C/D siRNA experiment (Supplementary Table S9). Furthermore, patients with low-level expression of these three genes were found to have better OS rates in both the ICGC and TCGA datasets (Fig. 5C and Supplementary Fig. S5A). We focused primarily on NCAPD3, a subunit of the condensin II protein complex involved in chromosome condensation, due to its significant positive correlations with KMT2C/D expression in the ICGC and TCGA datasets (r = 0.36, P = 5.56e−04 and r = 0.38, P = 5.19e−05, respectively, Supplementary Fig. S5B), and significant associations with clinical outcome in two additional human pancreatic cancer cohorts (log-rank P = 0.042 for Stratford dataset and P = 2.12e−04 for BCI_Zhang_merged dataset, Supplementary Fig. S5C). The candidacy of CDKL1 and EIF2AK4 was not supported in these two additional cohorts (Supplementary Fig. S5C). Interestingly, NCAPD3 showed significant downregulation in our KMT2D siRNA datasets compared with control (FDR < 0.05), whereas its expression remained almost unchanged in the KMT2C siRNA datasets, findings that were confirmed at the protein level (Fig. 5D). These data are consistent with GSEA pathway analysis (Fig. 5B), which highlighted a predominant role for KMT2D, but not KMT2C, in chromosome maintenance.

The dependence of PDAC cells on KMT2D for survival and the cell-cycle inhibition renders it an appealing therapeutic target. We postulated that lower expression, and/or mutation, could increase sensitivity to commonly used chemotherapies that target cell-cycle processes, providing one potential mechanism for favourable outcome seen in patients with lower KMT2D expression. This hypothesis was tested using murine Kmt2d siRNAs in three cell lines derived from the KC (DT6585 and DT6606; ref. 29) and KPC (TB32043; ref. 30) genetically engineered mouse models of pancreatic cancer (Fig. 6A). We were unable to perform these experiments in human cell lines as loss of KMT2D affected their cell proliferation (Fig. 2), something that was not seen in the three murine cell lines (Supplementary Fig. S6A). In both KC and KPC lines we observed an increase in their sensitivity to the nucleoside analogue 5-FU, with exposure to 10-fold less 5-FU capable of eliciting the same reduction in cell viability when used in combination with Kmt2d depletion (Fig. 6B). Interestingly, this increased sensitivity to 5-FU was specific to Kmt2d loss (Supplementary Fig. S6B–S6D), in accordance with our human data where, relative to KMT2D, KMT2C loss had weaker effects on cell proliferation (Supplementary Fig. S2D) and less of an impact on patient survival (ICGC data; Fig. 1).

The identification of mutations in several key methyl and acetyltransferases are now an established feature of many different cancers (47). Mutations in these enzymes are deemed insufficient for cancer initiation alone, with many of these lesions associated with various developmental disorders (48), where mutations in KMT2D are responsible for Kabuki syndrome (49). KMT2C and KMT2D represent a second tier of mutations in PDAC occurring at a lower frequency compared with mutations in KRAS, TP53, and SMAD4. The actual frequencies of KMT2C and KMT2D mutations in PDAC are still debatable (2–6). This is further complicated by different mechanisms of inactivation that include chromosomal deletion and variations in the nature and location of mutations, resulting in missense or truncation, and therefore loss of the functional enzymatic methyltransferase SET domain from the carboxyl-terminal.

In our study, we reasoned that inter-patient fluctuations in expression of these methyltransferases might impart significant changes in outcome. We were intrigued by our initial studies comparing matched GEP and outcome in publically available PDAC series that uncovered a strong favorable signal linked with low expression of these histone modifiers; something that has also been previously reported for KMT2D in breast cancer (24). It is reassuring that expression of these methyltransferases is consistent with recent studies that link mutations in these genes, and another family member (KMT2A), with improved overall and progression-free survival in PDAC (5).

The molecular mechanisms by which these methyltransferases contribute to PDAC development and/or influence patient outcome are likely to be complex, with the H3K4me1-3 chromatin marks potentially altering expression of many target genes. Our silencing experiments, which resulted in near complete loss of these methyltransferases and led to a marked reduction in proliferation effects of KMT2D in eight pancreatic cell lines, were consistent with previous studies using gene editing in other tumor settings (15), supporting some overlapping function across tumor types. Indeed, our studies go one step further by also implicating KMT2C in cell proliferation, suggesting that these proteins have somewhat complementary roles in PDAC, and may well explain the observation that KMT2D and KMT2C mutations arise independently of each other (5, 50).

We therefore went on to explore the potential downstream effectors of KMT2C and KMT2D loss in pancreatic cancer, focusing on changes common to both methyltransferases. Our RNA-seq experiments and subsequent GSEA analysis highlighted changes in DNA replication, repair, and cell cycle, echoing the results of our initial in vitro observations. To identify the more robust gene signatures, we focused primarily on changes consistent across the three cell lines, and siRNA experiments. This highlighted several proteins previously implicated in PDAC biology including c-MET, Calumenin, Claudin-1, and PTPN14, all of which were upregulated in our experiments and are known to impact on a range of distinct pathways and processes (41, 42, 44, 51). Indeed, all four were included in the 19-gene signature that was common to KMT2D and KMT2C in our siRNAs profiles. This gene signature, however, did not predict prognosis in any of the four PDAC GEP datasets tested (ICGC, TCGA, Stratford, or BCI_Zhang_merged) based on consensus clustering and OS (data not shown), leading us to speculate that these genes may contribute instead to PDAC development. Moreover, patient clustering based on this 19-gene signature alone did not show a significant enrichment in high or low KMT2C/D expression (data not shown). While unexpected, this may reflect the inherent complexity in direct comparisons between primary and cell line data, and may also be a measure of the complete loss of protein induced by siRNA, a feature unlikely to be reflected physiologically, where other compensatory mechanisms might be induced. When we instead focused on genes significantly correlated with KMT2C/D expression from the ICGC and TCGA datasets, it was encouraging to find that significant overrepresentation of the cell-cycle signature positively correlated with KMT2C/D gene expression. We could refine this cell-cycle compartment by comparing against our RNA-seq expression profile, identifying several common genes with concurrent direction of fold change and magnitude. An examination of the contribution of each gene in turn, and their clinical associations, highlighted a potentially novel role for the NCAPD3 subunit of the condensin II complex in PDAC biology. This chromosome condensation protein demonstrated a 1.3–4.1 fold reduction in expression across the three tested cell lines for KMT2D, and encouragingly was a good predictor of outcome in all four GEP series. Having confirmed these findings at the protein level, NCAPD3 is now the focus of ongoing experiments.

Further work remains to determine the mechanisms by which loss or reduced activity of these methyltransferases associates with improved patient outcome. In some pilot experiments, we have examined changes in sensitivity to chemotherapy when Kmt2d, or Kmt2c, were depleted. For these experiments, we used cell lines derived from murine models of PDAC whose proliferation, unlike human cell lines, remained unaffected by methyltransferase depletion. Here we noted that Kmt2d silencing increased sensitivity to the antimetabolite 5-FU, suggesting that favorable outcome linked with KMT2D low expression might be attributable to an improved response to chemotherapy. This effect was not associated with a change in the levels of Abcb1 (data not shown), which has previously been shown to be a mediator of 5-FU response (52). This overall effect was specific to Kmt2d, as murine cells were not sensitized to 5-FU upon Kmt2c depletion, perhaps reflecting the weaker impact of low KMT2C expression on patient outcome.

In summary, we have identified roles for KMT2D, KMT2C, and a new role for NCAPD3 expression as prognostic predictors in PDAC. The data support the incorporation of combined KMT2C/D mutation and gene expression into existing risk stratification models. In addition, we report that loss of KMT2D, and to a lesser extent KMT2C, impacts on the cell-cycle and DNA replication pathways, leading to a reduction in cell proliferation. Overall, our studies point to therapeutic benefits of targeting these methyltransferases in PDAC, especially in those patients that demonstrate higher KTM2C/D expression.

H.M. Kocher reports receiving a commercial research grant from Celgene and is a consultant/advisory board member for Baxalta. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J.B.N Dawkins, E. Maniati, H.M. Kocher, J. Fitzgibbon, R.P.Grose

Development of methodology: J.B.N Dawkins, J. Wang, E. Maniati

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.B.N Dawkins, J. Wang, E. Maniati, J.A. Heward, C. Chelala

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.B.N Dawkins, J. Wang, E. Maniati, C. Chelala, J. Fitzgibbon

Writing, review, and/or revision of the manuscript: J.B.N Dawkins, J. Wang, E. Maniati, J.A. Heward, H.M. Kocher, S.A. Martin, C. Chelala, F.R. Balkwill, J. Fitzgibbon, R.P.Grose

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.B.N Dawkins, J. Wang, E. Maniati, L. Koniali, J. Fitzgibbon, R.P.Grose

Study supervision: F.R. Balkwill, J. Fitzgibbon, R.P.Grose

The authors thank all members of the Fitzgibbon and Balkwill laboratories for advice and guidance of this project. The authors also thank Dave Tuveson and his laboratory for providing the KC and KPC cell lines and to our other colleagues and collaborators that provided access to the human pancreatic cell lines.

This work was supported by grants from Cancer Research UK (C587/A12888, C587/A16354), the European Research Council (ERC322566), and Barts and The London Charity (467/1307).

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