2.1 Cell culture

HCT116 (RRID:CVCL_0291), SW480 (RRID:CVCL_0546) and HT29 (RRID:CVCL_0320) were purchased from ATCC. HCT116 and HT29 were cultured in McCoy's (Thermo Fisher, Waltham, MA) and SW480 in RPMI medium (Thermo Fisher), supplemented with 10% v/v fetal calf serum (PAA). All cell lines have been authenticated using SNP profiling within the last 3 years (Multiplexion, Heidelberg, Germany). Absence of mycoplasms was regularly controlled by a PCR-based test. All experiments were performed with mycoplasma-free cells.

2.2 Compounds

Vorinostat, UNC0638, UNC0642, A-366, chloroquine diphosphate salt and actinomycin D were purchased from Merck. Lys05 and panobinostat were purchased from Selleckchem. BIX-01294 and Z-VAD(OMe)-FMK were purchased from TargetMol. Human recombinant KillerTRAIL was obtained from Enzo Life Sciences and TNF-α from Peprotech.

2.3 Generation of CRISPR knockout cell library and screening

SW480 cells with stable expression of Cas9-2A-puromycin were generated using the piggybac transposon system.¹⁴ 4 × 10⁷ SW480-Cas9 cells were infected with lentiviral particles containing the sgRNA library at a multiplicity of infection (MOI) of 0.3 to 0.4 as previously described.¹⁴ 72 h after transduction, cells were reseeded into T225 flasks (Thermo Fisher) and selected with 4 μg/mL blasticidin (Thermo Fisher) for 72 h. Subsequently, medium with blasticidin was removed and cells were allowed to recover for up to 7 days. For screening, 1.4 × 10⁷ cells were seeded per replicate in T225 flasks, resulting in >1.000-fold library representation. The next day, drugs were added (0.1% v/v DMSO, 3 μM vorinostat, 8 nM panobinostat) and replaced on days 4, 11 and 16 (for vorinostat) and 4, 8 and 11 (for panobinostat). Cells were harvested at day 17 (vorinostat) or 14 (panobinostat), and pellets containing 1.2 × 10⁷ cells were used for DNA extraction and library preparation. For each experimental condition, two replicates were performed.

2.4 Isolation of genomic DNA, library preparation and sequencing

Genomic DNA was extracted from 3.6 × 10⁷ pelleted cells per replicate using the DNeasy Blood & Tissue Kit (Qiagen), including preincubation with 200 μg/ml RNAse A (Qiagen) for 5 min. Concentration of eluted DNA was determined with a Qubit fluorometer using the dsDNA HS assay kit (both Thermo Fisher Scientific). DNA amplification was done as previously described.¹⁴ Per replicate, 25 PCR reactions were performed using 1 μg genomic DNA as input. Q5 Hot Start HF polymerase (NEB) was used with the primers SEQ-F1 (TTGCTCTGGAAAACTCATTTGC) and SEQ-R1 (TTAGGGGCGGGACTATGG) and the following PCR conditions: 98°C for 2 min, 25 cycles of 98°C for 10 s, 62°C for 15 s and 72°C for 30 s, final extension at 72°C for 2 min. The PCR product was cleaned using a QIAquick PCR purification Kit (Qiagen), and 5 ng of the eluted DNA was used for a second PCR step with the primers SEQ-F2 (AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNNNTCTTGTGGAAAGGACGAAACACCG) and SEQ-R2 (CAAGCAGAAGACGGCATACGAGAT[Index]GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCAACTTCGGAATTCGTGTGCTCGAG), and the following PCR conditions: 98°C for 2 min, 15 cycles of 98°C for 10 s, 72°C for 15 s, and 72°C for 30 s, final extension at 72°C for 2 min. The product was purified with Agencourt AMPure XP beads (Beckman Coulter) at a product-to-beads ratio of 1:1.2. The purified libraries were controlled for correct size using the DNA High Sensitivity Assay on a BioAnalyzer 2100 (Agilent) and then sequenced on a MiSeq (Illumina) by 100 or 150 bp single-end sequencing, with the addition of 20% PhiX Control v3 (Illumina) at a concentration of 8 pM. The sequencing coverage and quality statistics for each sample are summarized in Table S1.

2.5 Immunohistochemistry

Antibody staining was performed as previously described¹⁵ with 1:50 diluted anti-EHMT2 antibodies (Atlas Antibodies). Staining intensity was determined in normal colon epithelia and tumor cells using previously published, custom-made CRC tissue microarrays.¹⁶ The intensity of nuclear staining was scored on a semi-quantitative scale as 0 (negative), 1 (weakly positive, >0%-33%), 2 (moderately positive, 33%-67%) and 3 (strongly positive, 67%-100%) by a trained examiner who was blinded towards all patient data. Images were acquired using a standard brightfield microscope (Leica DMRE), DFC450C microscope camera and LAS 4.12 software (all Leica Microsystems). The median intensity scores of 1 to 3 separate cancer tissue cores per patient were used for analysis.

2.6 Short-term and long-term viability assays

Short-term and long-term viability assays were conducted as published before.¹⁷ The procedure is briefly described in the Supplementary Methods section.

2.7 Digital Western Blot

Digital Western Blot (DigiWest) was conducted as published before.¹⁸ The procedure is briefly described in the Supplementary Methods section and antibodies are listed in Table S2.

2.8 Apoptosis detection by flow cytometry

HCT116 were seeded in 12-well plates at a density of 100 000 cells per well. One day after seeding, cells were treated with the indicated drugs for 72 h, then harvested using trypsin-EDTA, washed with ice-cold PBS and stained with APC-Annexin V and 7-AAD in Annexin V Binding Buffer (all Biozol Diagnostica). Flow cytometric analysis was performed with a FACS Canto II (BD). For gating and data processing, the FlowJo V10.5 software was used. Exclusion of debris was achieved by gating in SSC-area (side scatter) vs FSC-area (forward scatter). Gates for APC-Annexin V and 7-AAD were defined by of samples with single staining, using the curly quadrant gating tool implemented in FlowJo V10.5.

2.9 Cell cycle phase quantification by flow cytometry

HCT116 were seeded in 12-well plates at a density of 100 000 cells per well. One day after seeding, cells were treated with the indicated drugs for 72 h, then were harvested using trypsin-EDTA, washed with PBS and fixed with ice-cold 70% ethanol. RNA was digested using RNAse A (Qiagen) before DNA was stained with propidium iodide (Biotium). Flow cytometric analysis was performed with a FACS Canto II (BD Biosciences), and gating and data processing were performed with FlowJo V10 software (BD Biosciences). Doublet discrimination was achieved by gating in FSC-width vs FSC-area. Automated cell cycle Gaussian curve fitting implemented in FlowJo V10.5 was used for cell cycle phase identification in area histograms of propidium iodide-stained cells. Localization of G1-phase peak was constrained manually to the propidium iodide area range between 20 000 and 100 000 to facilitate automated peak detection. Proportions of cells in the respective cell cycle phases are presented as percent of all cells fitted in G1, S and G2 populations by FlowJo V10.5. Images were arranged using Adobe Illustrator (Adobe Inc.).

2.10 Quantitative PCR

Total RNA was isolated with RNAeasy Mini Kit (Qiagen) and cDNA synthesis was performed using the Revert Aid H Minus First Strand cDNA Synthesis Kit (Thermo Fischer). Quantitative PCR was performed in 384-well plates on a LightCycler 480 (Roche, Basel, Switzerland) using the Universal probe library system (Roche). Primers are listed in Table S3.

2.11 Immunoblot

Whole cell lysates were separated on 4% to 12% NuPAGE Bis-TRIS gels (Thermo Fisher) and transferred to Immobilon PVDF membranes (Merck). For selected proteins, hand-cast SDS-polyacrylamide gels (7.5%, 12.5% or 15% v/v PAA) were used. Antibodies used are detailed in Table S4.

2.12 Compound treatment and measurement of viability of patient-derived cancer organoids

PDOs were cultured in droplets of BME V2 (Trevigen) and passaged every 5-7 days. The culture medium contained basal medium (advanced DMEM/F12 with 1% v/v penicillin/streptomycin, 1% v/v HEPES buffer and 1% v/v Glutamax (all from Life Technologies), 100 ng/mL recombinant noggin, 10 nM gastrin, 50 ng/mL EGF (all from Peprotech), ×1 B27 (Thermo Fisher), 1.25 mM n-acetyl-cysteine, 10 mM nicotinamide, 10 μM Y-27632 dihydrochloride (all from Merck), 500 nM A83-01 (Tocris), 10 nM prostaglandin E2 (Santa Cruz Biotechnology) and 100 mg/mL Primocin (Invivogen). The culture medium was exchanged every 2 to 3 days. Drug treatment was essentially done as described before.¹⁹ In brief, organoids were digested with TrypLE Express (Thermo Fisher), resuspended in BME V2 plus culture medium and seeded in 384-well plates precoated with BME V2. Cell numbers between different organoid lines were normalized by measuring ATP levels using CellTiter-Glo (Promega), and a number of cells per well equivalent to 5000 photon counts was used. Organoids were allowed to grow for 3 days before drug treatment for 96 h. Cell viability was determined using CellTiter-Glo and a Mithras LB940 Microplate Reader (Berthold Technologies). Two biological replicates were performed.

2.13 Imaging of patient-derived cancer organoids

PDOs were seeded at a density of 300 to 500 cells/μl growth-factor-reduced Matrigel (Corning) or Cultrex reduced growth factor Basement Membrane Extract type R1 (R&D Systems) in 6-well plates. Organoids were allowed to grow for 72 h before compound treatment. After treatment for 72 to 144 h, microscopic images of organoids were acquired using Primovert microscope (Zeiss) and Axiocam ERc 5 s camera (Zeiss) at ×2 magnification. The images were processed by the ZEN core software (Zeiss) and arranged using Adobe Photoshop Version 22.4.2 (Adobe Inc.). Representative sections of the images were used for illustration.

2.14 Statistical analysis

For all in vitro experiments, a two-sided Student t test was used for one- and two-sample-comparisons. For multiple sample comparisons ANOVA and Holm-Šídák's multiple comparisons test were performed. Differential DigiWest signal was tested using the moderated t test implemented in the “limma” R/Bioconductor package.²⁰ Statistical analysis of immunohistochemistry stainings was done using SPSS version 24.0 (IBM). Pearson's exact Chi-Square test was used to evaluate the correlation between EHMT2 expression and clinicopathological characteristics. The Kaplan-Meier method was used to determine the median overall and tumor-specific survival with 95% confidence intervals. The log rank test was applied for testing differences between median survivals.

Quality of drug interactions was analyzed with the web application SynergyFinder 2.0.²¹ Normalized short-term viability data was uploaded to https://synergyfinder.fimm.fi/ as drug matrices. The provided baseline correction and four-parameter logistic regression curve-fitting algorithm were applied. Synergy scores of all reference models implemented in the SynergyFinder 2.0 web application (Bliss, highest single agent (HSA), Zero interaction potency (ZIP) and Loewe) were calculated. The interaction barometer terminology was used to describe the quality of drug interactions.²² The terminology aggregates the calculated scores to define the quality of interaction: strong synergy (positive Bliss and Loewe scores), weak synergy (if only Bliss or Loewe is positive), noninteraction or additivity (negative Bliss and negative Loewe, positive HSA score) and antagonism (negative HSA score).

3.1 Knockout of EHMT1/2 sensitizes CRC cells to treatment with HDAC inhibitors

HDACi are ineffective as monotherapy for CRC, but preclinical studies indicate that combining HDACi with other epigenetically active drugs can enhance their antineoplastic effects.^{23, 24} To identify epigenetic targets for rational drug combinations, we performed a pooled CRISPR/Cas9 knockout screen in SW480 cells using a custom sgRNA library directed against genes that modulate the chromatin state. We selected genes based on their annotated function and used the CRISPR Library Designer software to identify effective sgRNA.¹⁴ The resulting “Chromatin Modulator” library contains sgRNAs targeting 614 genes with 20 sgRNAs/gene and 250 nontargeting controls (Tables S5-S7). Pools of knockout cells were generated and treated with either 3 μM of the HDACi vorinostat or DMSO for 16 days. Subsequently, the abundance of sgRNA cassettes in each pool was determined by next-generation sequencing (Figure 1A). Quality control of the screen shows a nearly complete coverage of the library (99.0% of all sgRNA represented) and a high correlation between replicates (Figure S1). Comparison of sgRNA abundances between pools revealed a significant depletion for sgRNAs targeting the histone methyl transferases EHMT1 and EHMT2 upon HDACi (Figure 1B and Table S8). EHMT1 and EHMT2 are lysine methyltransferases that act in a heterodimeric complex and catalyze the mono- and dimethylation of H3K9.²⁵ We found that all sgRNAs against EHMT1 and 14 of 20 against EHMT2 were depleted, which is in contrast to the equal distribution of nontargeting sgRNAs (Figure 1D). These results were confirmed by another CRISPR/Cas9 screen using the HDACi panobinostat. In this screen, sgRNAs against EHMT1 were also significantly depleted while those against EHMT2 showed a similar, but weaker phenotype (Figure 1C,E and Table S9). Together, these results demonstrated that knockout of EHMT1 and EHMT2 sensitizes CRC cells towards treatment with HDACi. Since both genes act in a heterodimeric complex and are druggable targets, we decided to characterize this synthetic lethal interaction.

3.2 EHMT2 is differentially expressed in CRC and protein levels correlate with clinical outcome

To determine EHMT1/2 protein expression in CRC and its correlations with tumor features and clinical outcomes, we performed immunohistochemistry on cancer tissue microarrays from 1066 CRC patients (Figure 2A). The clinical characteristics of the cohort are summarized in Table S10. For EHMT1, we tested several commercially available antibodies, but the quality of staining was insufficient for quantitative assessment of protein expression. In contrast, immunohistochemistry of EHMT2 was successfully established, and expression was determined by measuring nuclear staining intensities of tumor cells (Figure 2A,B). The majority of tumors (52.6%) showed a strong EHMT2 expression, while a minority (3.3%) had no detectable protein expression (Figure 2C). Comparison of EHMT2 levels with clinical characteristics demonstrated a significant correlation with the extent of tumor size/invasion, with low EHMT2 expression predominantly found in UICC (Union for International Cancer Control) stage T3/T4 cancers (Figure 2D, Table S10). Furthermore, patients with low intratumoral EHMT2 levels (score 0-2) showed a significantly poorer overall and tumor-specific survival (Figure 2E,F), particularly in UICC stage II cancers (Figure S2). In summary, these results demonstrate that EHMT2 is differentially expressed in CRC and that low EHMT2 protein expression is a prognostic marker for poor survival.

3.3 EHMT1/2 inhibitors synergize with HDACi to reduce tumor proliferation

To further validate and characterize the interaction between HDACi and functional loss of EHMT1/2, we used four previously described, small molecule inhibitors of EHMT1/2 (BIX-01294,²⁶ UNC0638,²⁷ UNC0642,²⁸ and A-366²⁹). We tested the effect of combinations of EHMT1/2 inhibitors (EHMTi) with HDACi on tumor viability in two CRC cell lines with confirmed expression of EHMT1/2 (Figure 3A,B). In total, six different drug combinations were investigated, using four different EHMT1/2 and two different HDAC inhibitors (vorinostat, panobinostat). Compared to single agent treatment, the combination of UNC0638/vorinostat resulted in a stronger reduction of cell viability in both cell lines (Figure 3C). To analyze the quality of drug interactions, we calculated the ZIP synergy score³⁰ and used the interaction barometer terminology,²² which is based on Bliss, Loewe and HSA synergy scores. Both the ZIP score and the interaction barometer showed a synergistic interaction between UNC0638 and vorinostat after short-term treatment (Figure 3D and Table S11). Similarly, the combination of UNC0638 and vorinostat synergistically reduced cell growth in long-term proliferation assays (Figure 3E). Other combinations of HDACi and EHMTi also showed a positive ZIP score, and the quality of interaction was predominantly synergistic (Table S11). Furthermore, in long-term proliferation assays, vorinostat was synergistic in combination with all EHMTi in HCT116, and synergistic with UNC0642 and BIX-01296 in HT29 cells (Figure S3), supporting a compound class-specific effect. Given the consistent synergistic interaction of vorinostat and UNC0638, we focused on this drug combination to characterize the underlying mechanisms, using drug concentrations that show a strong phenotype in long-term proliferation assays.

3.4 Combination treatment reduces cell proliferation by drug-specific modes of action

To uncover the biological mechanisms responsible for the synergistic antiproliferative effect, we used a bead-based, high-throughput Western blot assay (DigiWest). For this exploratory approach, we designed a custom panel containing 158 (phospho-specific) antibodies (Table S2) against proteins that are involved in 11 major cancer-associated cellular processes and pathways (Figure 4A). HT29 and HCT116 cells were treated with either DMSO or vorinostat/UNC0638 and protein samples were then analyzed by DigiWest. The results showed that many proteins were induced by the combination treatment, but only few depleted (Table S12). Amongst the most up-regulated proteins (log2 fold change >0.4), we identified candidates involved in apoptosis (pro-CASP8, cleaved CASP8), cell cycle regulation (p21, p27, Cyclin A), and autophagy (LC3B-II, ATG3, ATG7; Figure 4B), indicating that these processes were targeted.

Next, we sought to discriminate the individual contribution of either vorinostat or UNC0638 to the dysregulation of the aforementioned processes. We focused on apoptosis induction first and found that treatment with UNC0638 neither increased levels of Annexin V-positive cells nor the cleavage of PARP1 or CASP3, the downstream caspase of CASP8. In contrast, vorinostat increased the number of apoptotic cells and cleavage of both proteins in a concentration-dependent manner (Figure 4C,D). However, when both compounds were combined at concentrations that synergistically reduced cell proliferation, only a minor increase of Annexin V-positive cells and no cleavage of CASP3 or PARP1 was observed (Figure 4C,D). This observation was consistent for HCT116 and HT29 (Figure S4A). To further determine if caspase-dependent apoptosis contributes to the viability effect of the combination treatment, we used the pan-caspase inhibitor Z-VAD(OMe)-FMK. While cell death caused by the apoptosis-inducing ligand TRAIL could be rescued by co-incubation with Z-VAD(OMe)-FMK, no such effect was observed when cells were treated with vorinostat/UNC0638 (Figures 4E and S4B). Taken together, these results indicated that the synergistic antiproliferative effects of low-dose vorinostat/UNC0638 combination do not rely on caspase-dependent apoptosis.

We then investigated effects on cell cycle regulation. As shown in Figure 5A, combined vorinostat/UNC0638 significantly reduced the proportion of HCT116 cells in the G1 phase and caused a cell cycle shift towards the G2 and S phase, while the effect of single agents was less pronounced. Results from the DigiWest assay demonstrated increased expression of the cell cycle regulators Cyclin A, Cyclin D1, p21 and p27 upon combination treatment (Figure 4B and Table S12). Induction of these proteins could be confirmed by immunoblot and was caused by vorinostat in HCT116 (Figure 5B). In HT29, vorinostat induced the expression of Cyclin D1 and A, but not p21 (Figure S5A). These findings demonstrated that the drug combination modulates the cell cycle and that this effect is mainly driven by vorinostat.

As the DigiWest assay revealed an up-regulation of autophagy markers, we assessed levels of LC3B-II, ATG3 and ATG7 by immunoblot. Upregulation of LC3B-II and ATG7 by the combination treatment was confirmed in HCT116 (Figure 5C) and HT29 (Figure S5B). While LC3B-II was only increased by UNC0638, expression of ATG7 was induced by vorinostat as well. In contrast, ATG3 levels were not consistently altered. Next, we investigated if changes in transcript levels of genes involved in autophagy are responsible for the induction of LC3B-II and ATG7. However, we only observed a significant downregulation of ATG12 and ATG5 levels upon combination treatment in HCT116 (Figure S6). Since induction of autophagy not only functions as a mechanism to limit cell proliferation, but also to rescue tumors from cell death,³¹ we sought to further investigate its role during UNC0638 treatment. To this end, we performed short-term viability assays combining UNC0638 with known inhibitors of late autophagy (chloroquine, Lys05), and observed a synergistic reduction of cell viability by these drug combinations (Figure S7A,B and Table S11). Together, these data indicated that autophagy induction is elicited primarily by EHMTi and that the autophagic flux is required for the survival of CRC cells, as blocking of late autophagy enhanced UNC0638-induced cell death.

A fundamental effect of EHMTi and HDACi is the alteration of specific histone marks.^{25, 32} Since we observed synergistic antiproliferative effects when vorinostat and UNC0638 were combined at low doses, we asked if histone modifications are already affected at these concentrations. To address this question, we analyzed global H3K9 mono- and dimethylation (H3K9me1 and 2) and acetylation (H3K9ac). In both HCT116 and HT29 cells, treatment with either UNC0638 or vorinostat alone resulted in a modest reduction of H3K9me2 (Figure S8A,B). A decrease of H3K9me1 levels was not consistently observed. In contrast, H3K9ac was induced by both inhibitors. Interestingly, the drug combination caused an even stronger induction of H3K9ac in HCT116. Together, these results indicate that low concentrations of vorinostat and UNC0638 are sufficient to affect histone marks and that the combination treatment can result in an enhanced induction of H3K9ac.

In summary, we demonstrated that combining EHMTi and HDACi reduces CRC cell viability by drug-specific modulation of cell cycle progression and autophagy. On an epigenetic level, both inhibitors could alter specific histone modifications.

3.5 HDAC and EHMT1/2 inhibition synergistically reduce viability of patient-derived CRC organoids

Patient-derived cancer organoids closely reflect many characteristics of primary cancers, including their response to drugs.³³ We systematically established PDO lines from CRC tissues for in vitro testing of compounds.¹⁹ To determine the response of PDOs to the combination of UNC0638 and vorinostat, we treated a panel of 11 lines representing diverse tumor stages.¹⁹ (Figure 6A). We found that the individual response of PDOs to single agent treatment was very heterogenous (Figure 6B). We then assessed the quality of drug interaction using the ZIP score, which showed that the drug combination was synergistic in the majority of PDO lines (Figure 6C and Table S11). Interestingly, synergistic effects were observed in PDO lines that were either insensitive (line 090) or highly sensitive (line 135) to UNC0638 (Figure 6D), and could be confirmed by microscopy-based imaging (Figure 6E). The synergistic antiproliferative effect in PDOs was also observed for the combination of vorinostat with two other EHMTi inhibitors, A366 and UNC0642 and (Figure S9A,B). Finally, we sought to investigate if the cellular processes altered by vorinostat/UNC0638 in cell lines were similarly modified in PDOs. To this end, we selected the PDO line 090 in which the drug combination was highly synergistic (Figure 6E). In this model, cycle markers (p27, p21) were induced by single vorinostat and UNC0638 treatment, but not by the drug combination (Figure S9C). In contrast, UNC0638 caused a clear induction of LC3B-II, which was even enhanced by the combination treatment (Figure S9D). We also observed that vorinostat increased H3K9ac while UNC0638 reduced H3K9me2 levels (Figure S9E). In summary, we showed that combining EHMT1/2 and HDAC inhibitors synergistically reduced proliferation of CRC PDOs, which was associated with similar cellular and epigenetic changes as observed in cell lines.