Histone Lysine Methylation Dynamics Control EGFR DNA Copy-Number Amplification

Chromosomal instability is a hallmark of cancer cells (1). These abnormalities can include entire chromosome events or they can be localized to site-specific chromosomal regions (2). For example, the chromosome 1q12–q25 (1q12–q25) region is regularly amplified in tumors (3–9). This amplification event is often associated with drug resistance, as a number of drug-resistance genes (e.g., MCL1, CKS1B) reside within this chromosomal region (3–9). Amplification of these regions can occur as frequently as the well-documented oncogene amplifications MYC and EGFR in certain tumor types (e.g., 1q21.3 at 21% vs. MYC at 26% in liver cancer; ref. 10). However, it is important to note that amplifications are not always permanently integrated (2). A recent study estimated that approximately 50% of tumors contain extrachromosomal DNA (ecDNA) amplifications for the EGFR and MYC genes (11). The extrachromosomal nature of these copy gains provides the cell an opportunity to select either for or against these amplifications, which will affect cell growth and drug response. For example, extrachromosomal amplification of EGFR results in increased sensitivity to targeted therapies. However, following prolonged treatment with an EGFR inhibitor, the ecDNA copies of EGFR are reduced, leading to therapy resistance (12). In the case of methotrexate therapy, the dihydrofolate reductase (DHFR) gene is amplified and provides resistance (13–16). DHFR amplifications can occur as integrated and/or extrachromosomal events (13–16). Therefore, extrachromosomal amplifications promote tumor heterogeneity and tumor adaptation, both of which are major contributors to drug resistance (2, 11). Elucidating the cellular physiology and molecular mechanisms that promote oncogene-associated extrachromosomal events will have a profound impact on our understanding of tumor heterogeneity and drug resistance.

The mechanisms by which extrachromosomal amplification events occur are still poorly understood; however, recent studies have demonstrated a critical role for epigenetic states and chromatin-modifying enzymes in controlling site-specific rereplication and, in turn, DNA copy-number amplification (10, 17–19). For example, overexpression or stabilization of the H3K9/36 tridemethylase KDM4A, and the direct modulation of chromatin states (H3K9 and K36 methylation) promotes transient site-specific DNA copy gains (TSSG) within the Chr1q12–q21 region (17–20). These DNA copy gains are transiently generated during S phase and are lost in late S or early G₂ phase of the cell cycle (18). Indeed, KDM4A interacts with components of the replication machinery, facilitating rereplication at the TSSG sites (18). Consistent with these findings, we illustrated that targeting KDM4 family members through H3K4 methylation can result in TSSGs (10). This study reveals that lysine methyltransferases and demethylases have a high degree of specificity and work in concert to modulate site-specific DNA copy gains in the genome. These studies highlight the possibility that clinically relevant oncogenes exhibiting plasticity in their copy-number gains (i.e., EGFR; ref. 12) could be regulated by a comparable mechanism. It also leads one to question whether extrinsic cellular cues are also able to facilitate these oncogenic amplifications (21). Elucidating mechanisms by which chromatin modulators and epigenetic states regulate oncogenes would identify new therapeutic avenues to control copy-number heterogeneity and drug responses.

EGFR DNA amplification tends to result in poor prognosis for patients with EGFR-amplified cancer (22). EGFR-targeted therapies have been developed in recent years (23) and EGFR amplifications have been shown to associate with varying degrees of patient response across various amplified tumors (24–29). EGFR DNA amplification is prevalent across a number of different cancer types, with up to 54% of patients exhibiting amplification in some tumor types (e.g., glioblastoma multiforme; ref. 10). An important clinical challenge with EGFR amplification is the plasticity of the amplification (12). Therefore, there is a major clinical need to resolve the mechanisms driving EGFR amplification.

In this study, we demonstrate that chromatin-modifying enzymes and their associated epigenetic states control amplification of the EGFR locus. Specifically, we demonstrate that directly interfering with H3K9 and H3K27 methylation promotes EGFR amplification. Furthermore, we establish a critical interplay between H3K4/9/27 lysine methyltransferases and demethylases in either promoting or blocking EGFR amplification. For example, KDM4A overexpression promotes EGFR copy gains in conjunction with three H3K4 methyltransferases: KMT2A/MLL1, SETD1A, and SETD1B. In addition, we demonstrate that suppression of specific H3K9 KMTs and the H3K27 KMT EZH2 generates EGFR amplification. Consistent with these genetic experiments, we demonstrate for the first time that chemical inhibitors targeting histone lysine methyltransferases and demethylases (KMT–KDMs) are able to rheostat EGFR copy number and, in turn, growth factor and EGFR inhibitor responses. Finally, we demonstrate that extrinsic cellular cues (hypoxia and EGF) promote EGFR amplification by modulating the KMT–KDM network that controls EGFR copy number. Taken together, our study uncovers both chromatin modifiers and extracellular signals that control EGFR amplification and demonstrate that epigenetic therapies could hold a key to modulating EGFR copy-number heterogeneity in cancer, which has significant clinical implications.

Previous analysis demonstrated that up to 54% of primary tumors across the pan-cancer dataset of The Cancer Genome Atlas (TCGA) harbor EGFR amplifications of which some were shown to harbor extrachromosomal amplification (10, 11). To further explore EGFR amplification heterogeneity across and within tumors, we assessed the range of EGFR DNA copy gains and the associated EGFR RNA expression levels in each of the tumors in the pan-cancer TCGA dataset (7,069 samples; Fig. 1A). The analyses revealed significant plasticity in EGFR DNA copy number across tumor types, ranging from 2.4 to 8 copies (red) to more than 8 copies (blue; Fig. 1A). We also observed tumors with a loss of EGFR (black; Fig. 1A). As the DNA copy number increased, there was an increase in EGFR RNA levels (Fig. 1A). Therefore, promoting more EGFR DNA copies tends to associate with increased EGFR transcripts in tumors.

Because there is a range of EGFR DNA copy gains across tumors (Fig. 1A) and others have observed amplification plasticity (12), we assessed whether perturbation of epigenetic states associated with transient site-specific DNA amplifications could promote EGFR DNA copy gains (10, 18). Prior studies demonstrated that the introduction of histone 3.3 lysine to methionine (H3.3 K–M) mutants into cells could interfere with the associated methylation at that specific lysine (30). In fact, we demonstrated that transducing cells with different H3.3 K–M mutants could interfere with associated methylation and, in turn, promote amplification at specific regions in the genome (18, 30). For example, 1q12h was copy gained with the introduction of H3.3 K9M and H3.3 K36M (18), whereas 1p32.3 was amplified with only H3.3 K36M (10). These data illustrate the localized impact that modifying specific lysine methylation states has on the predilection of regions to undergo amplification versus whole-genome instability and amplification (10, 18). For these reasons, we tested whether introduction of H3K–M mutants into the immortalized retinal pigmental epithelial (RPE) cells that have a nearly diploid genome (18, 31) and no documented mutations in the EGFR gene could allow us to identify residues important in repressing EGFR amplification (Supplementary Fig. S1A). Because cell-cycle arrest (e.g., G₁–S; ref. 18) has been shown to block copy gains driven by epigenetic perturbation, we evaluated cell-cycle profiles for each mutant tested (Supplementary Fig. S1B). Our analyses did not reveal major changes in cell-cycle profiles between cells expressing the different H3.3 K–M mutants (Supplementary Fig. S1B); therefore, the cells were subsequently processed for DNA FISH to determine the copy-number status of EGFR.

Individually H3.3 K9M and H3.3 K27M resulted in significant increases in EGFR DNA copy number, without changing the copy number of chromosome 7 and 8 centromeres (7C and 8C, respectively; Fig. 1B and C), the 7p telomere (7p tel, 7p22.2; Fig. 1D; Supplementary Fig. S1C), or at a region adjacent to EGFR (IKZF1, 7p12.2; Fig. 1E; Supplementary Fig. S1D). These data suggest that EGFR is undergoing site-specific copy gain as opposed to whole chromosome 7 or chromosome 7p arm copy gain events, which is consistent with prior studies illustrating that exposure to K9M could promote site-specific amplifications on chromosome 1 (10, 18). To date, no other region undergoing site-specific copy gain has been shown to be controlled by H3K27M (10, 18). We also observed that approximately 1% to 2% of cells in the K9M and K27M mutants had higher EGFR amplification levels than wild-type H3.3 (≥5 copies; Fig. 1F and G). Consistent with this increase in EGFR copy number, these cells also had a modest increase in EGFR transcript levels when compared with cells expressing a wild-type H3 (Fig. 1H). Of interest, H3.3 K4M and H3.3 K36M did not promote EGFR copy gains.H3.3 K36M has been shown to promote copy gains of other regions (e.g., 1q12h, 1q21.2, 1p32.3; refs. 10, 18), which further highlights the specificity of epigenetic states in modulating amplification sensitivity in the genome.

Consistent with these observations, our chromatin immunoprecipitation sequencing (ChIP-seq) coupled with publicly available Hi-C data in RPE cells (32) suggest that the genomic vicinity of the EGFR locus has a specific chromosome structure and pattern of chromatin modifications (Fig. 1I). The EGFR gene body corresponds to a subdomain within a larger 500 kilobase (Kbp) 3-D interaction domain, which also includes another gene, SEC61G. Both gene bodies are enriched in H3K36me3 (Fig. 1I). This chromatin interaction domain is flanked by two wide approximately 100-Kbp regions enriched in H3K27me3 (Fig. 1I, marked with blue shadow). The H3K27me3 region immediately adjacent to the 3′ end of the EGFR gene forms a chromatin interaction domain reminiscent of the small Polycomb-generated domains described by Kundu and colleagues (33). This subdomain on one boundary of the EGFR-containing domain interacts with the H3K27me3 region on the opposite boundary of the EGFR-containing domain, resulting in the streak of interactions above this domain in the Hi-C map (Fig. 1I). This looping interaction between distant H3K27me3 regions is similar to the Polycomb-mediated looping interactions described previously (33). Thus, EGFR resides within a 3-D interaction domain whose boundaries are marked by two wide regions of H3K27me3 enrichment that form a looping interaction with each other (Fig. 1I).

At a larger scale, the EGFR-containing domain, together with the adjacent H3K27me3-enriched boundary regions and two other interaction domains to the 3′ of EGFR, are part of an approximately 1 Mb region that is flanked on both sides by wide areas of H3K9me3 enrichment (Fig. 1I). To assess the significance of these associations, we evaluated a panel of cells analyzed by ENCODE (34, 35). In all cases, they contain similar configurations of H3K9me3 and H3K27me3 at the noted locations in the RPE cells (Fig. 1I–L).Cells that did not express EGFR had broader H3K27me3 in between the domains we documented in RPE cells (Fig. 1K and L). Furthermore, this pattern was conserved at the modification and organizational level when analyzing a mouse lymphoblast cell line, C12-LX (Fig. 1L; refs. 34, 35). Finally, we analyzed publicly available H3K27me3 data in cells expressing H3.3 K27M (36) and observed a loss of the H3K27me3 domains that flank EGFR (Fig. 1M), which was consistent with a direct effect on the locus and, in turn, amplification observed in Fig. 1B and C. Taken together, these epigenomic profiles and our genetic experiments using mutated histones are consistent with the possibility that this topologic structure and H3K9/27 methylation are suppressing the propensity of the EGFR locus to undergo amplification and increased expression.

The KDM4 family of histone lysine demethylases catalyze H3K9 and K36 demethylation (37, 38). In addition, this enzyme family directly regulates DNA amplifications associated with H3K9 and K36 demethylation (10, 18). Therefore, we tested whether overexpression of the KDM4 family of enzymes could phenocopy the increase in EGFR copy number observed upon H3K9 methylation interference. The GFP-tagged KDM4 family members were transiently overexpressed for 24 hours followed by DNA FISH analysis for EGFR DNA copy number (Supplementary Fig. S2A–S2F). Of the overexpressed KDM4 family members, only transient KDM4A overexpression promoted increased EGFR DNA copy number, which highlights the specificity of KDM4A in regulating the EGFR locus (Fig. 2A). Furthermore, catalytically inactive KDM4A (H188A; ref. 38) and a mutant lacking the Tudor domains that recognize H3K4 methylation (Tudor Del; ref. 39) were unable to promote EGFR copy gains (Fig. 2B; Supplementary Fig. S2G–S2I). These data were also confirmed with RPE cells that stably overexpressed KDM4A (Fig. 2C and D; Supplementary Fig. S2J and S2K). Chromosome 7 or 8 centromere loci, 7p Tel, as well as other previously tested loci were not affected by KDM4A overexpression (Fig. 2A and D; Supplementary Fig. S2I; refs. 10, 18).

Consistent with these findings, analysis of available KDM4A ChIP-seq data (40) revealed that KDM4A was enriched across the EGFR locus between the H3K9me3 blocks (1.6 Mb; Fig. 2E, top: KDM4A in blue and H3K9me3 in black). These observations were consistent with the previous analyses that noted that KDM4A occupied the EGFR promoter region (40), which occurs within this larger binding domain that we have identified. In addition to promoting copy gains, KDM4A overexpression also increased EGFR transcripts as determined by both RNA sequencing and quantitative RT-PCR (Fig. 2E and F). These findings are entirely consistent with a recent report that demonstrated a role for KDM4A in regulating EGFR gene expression (40). Taken together, these data illustrate that KDM4A directly regulates this locus and plays a fundamental role in generating EGFR DNA copy gains and in regulating EGFR expression.

Because KDM4A overexpression promoted EGFR copy gains and increased RNA expression, we assessed the impact that KDM4A overexpression had on EGFR inhibitor sensitivity. We observed increased sensitivity to both Lapatinib and Gefitinib in KDM4A-overexpressing cells when compared with control cells (Fig. 2G and H). On the basis of these data, we further reasoned that EGF supplementation would promotecell proliferation and migration in KDM4A-overexpressing cells.Consistent with this hypothesis, KDM4A-overexpressing cells had a modest but significant increase in cell migration as compared with control cells in scratch assays and increased cell proliferation in response to EGF (Fig. 2I and J; Supplementary Fig. S2L). Furthermore, the increased cell proliferation upon EGF treatment in KDM4A-overexpressing cells was completely suppressed by siRNA-mediated depletion of EGFR (Fig. 2K; Supplementary Fig. S2M), which emphasized the importance of increased EGFR gene expression to observe the increased cell proliferation.

Because EGFR amplifications are observed in lung tumors (10, 22) and cancer cell lines (i.e., HCC827; refs. 41, 42), we tested whether KDM4A could be contributing to the observed EGFR amplifications in lung cancer cells. We used both KDM4A siRNA depletion and chemical inhibition [KDM4i (QC6352); ref. 43] of the KDM4 family to assess the impact of KDM4A on EGFR amplification. Specifically, we used DNA FISH to test whether the classically used EGFR-amplified HCC827 lung cancer cells would have reduced amplifications upon KDM4A depletion or inhibition. HCC827 cells, in addition to exhibiting extensive EGFR amplification, have an acquired activating mutation within the EGFR tyrosine kinase domain (Exon 19 deletion; ref. 41). As previously noted, HCC827 cells had very high EGFR DNA amplification levels that form large EGFR gene cluster clouds (Fig. 3A). The amplifications are so abundant that they appear as clouds in the interphase nuclei (42). However, KDM4A depletion significantly reduced the size of EGFR-amplified clouds in the HCC827 cells (Fig. 3A and B; Supplementary Fig. S3A and S3B). Furthermore, KDM4 family inhibition significantly reduced the extent of EGFR amplification in these cells as well (Fig. 3C and D). On the basis of these observations and the fact that EGFR DNA amplification has been shown to correlate with better EGFR inhibitor responses in patients (24–29), we hypothesized that KDM4 inhibitor treatment would reduce the sensitivity of HCC827 cells to the EGFR inhibitor Gefitinib. As anticipated, a reduction in EGFR DNA amplification directly correlated with a reduced response to Gefitinib (Fig. 3E, right). Taken together, these data highlight a functional and significant role for KDM4A in modulating EGFR amplifications, expression, and response to both EGFR inhibitors and growth factors in both diploid, nontransformed cells and EGFR-amplified cancer cells.

Because K9 methylation interference and catalytically active KDM4A overexpression promoted EGFR amplification, we tested whether each of the H3K9 lysine methyltransferases (K9 KMT) were equally capable of inhibiting EGFR copy gains or whether there was enzyme specificity. Specifically, RPE cells were siRNA depleted with at least two independent siRNAs for each K9 KMT. The knockdowns were confirmed and cell-cycle profiles were generated for each siRNA, which ensures no overt arrests occurred and, in turn, interfered with EGFR copy gains (Supplementary Fig. S4A–S4F). Samples were then analyzed by EGFR DNA FISH. With the exception of G9a/KMT1C, depletion of all other K9 KMTs resulted in a significant increase in EGFR DNA copy number (Fig. 4A). These data coupled with the methylation interference and demethylase activity requirement strongly suggest that maintaining the degree of H3K9me1/2/3 methylation at this locus is critical for preventing amplification. Consistent with this notion, H3K9me1/2/3 ChIP-seq illustrated that the higher-order organized domains that contained the EGFR locus were enriched for H3K9me1/2 between the H3K9me3 flanking blocks (Fig. 4B). Taken together, these data imply the need to balance H3K9 methylation to prevent or promote EGFR copy gains.

In addition to H3K9 methylation interference causing EGFR amplification, we also observed that H3K27 methylation interference increased EGFR copy number (Fig. 1B and C). Therefore, we sought to further explore the importance of K27 methylation in modulating EGFR amplification. First, we compared a publicly available EZH2 ChIP-seq dataset (44) to our H3K27me3 profiles. We observed an overlap between EZH2 occupancy and the H3K27me3 domains that flank the EGFR locus (Fig. 4B). These binding profiles and the fact that H3K27 methylation interference promoted EGFR amplification (Fig. 1B and C) and reduced H3K27me3 flanking the EGFR locus (Fig. 1M) prompted us to then test whether EZH2 siRNA–mediated depletion or chemical inhibition would phenocopy H3K27 methylation interference–induced EGFR copy gains.

Using two independent siRNAs against EZH2 in RPE cells, we observed a significant increase in EGFR copy number while not changing copy number of other chr7 regions (Fig. 4C; Supplementary Fig. S4G–S4K). We further demonstrated that a specific EZH2 pharmacologic inhibitor [EZH2i (C24); ref. 45] was able to promote EGFR amplification in a dose-dependent manner while not altering copy number of other loci in RPE cells (Fig. 4D and E; Supplementary Fig. S4L and S4M). EGFR copy number was also increased in colorectal cancer cell lines that have either low (HCT-15) or high (HT-29) EGFR copy number when treated with the EZH2i (Supplementary Fig. S4N–S4Q). Furthermore, removing EZH2i from RPE cells (referred to as washout) allowed EGFR copy gains to return to baseline, which highlights their transient nature (Fig. 4F; Supplementary Fig. S4M). At the higher doses of EZH2 inhibition, EZH2i also resulted in a higher number of copies within some of the copy-gained nuclei and increased EGFR transcript (Fig. 4G), which phenocopies H3K27M-transduced cells (Fig. 1F–H). Furthermore, ChIP-seq analyses of PC9 cells (lung adenocarcinoma) treated with an EZH2 inhibitor (44) also showed a reduction in the H3K27me3 domains flanking EGFR (Fig. 4H). To further explore the importance of H3K27me3 in preventing the EGFR locus from amplifying, we overexpressed each of the H3K27 demethylases KDM6A and KDM6B individually and then conducted EGFR FISH (Fig. 4I; Supplementary Fig. S4R and S4S). In contrast to the specificity observed with KDM4 family members, both KDM6 members promoted EGFR copy gains (Fig. 4I). Taken together, these results demonstrate a critical role for H3K27 methylation and the associated enzymes in preventing the EGFR locus from undergoing amplification and expression.

Because EZH2 inhibition promoted EGFR amplification and increased expression levels, we tested whether cells treated with the EZH2i or EZH2 siRNA depletion would respond differently to EGF supplementation. Specifically, cells were pretreated with EZH2 inhibitor for 24 hours or depleted with two independent EZH2 siRNAs to promote increases in EGFR copy number before being supplemented with DMSO or EGF. Both EZH2 inhibitor– and EZH2 siRNA–treated cells demonstrated a significantly increased proliferation in response to exogenous supplementation with EGF compared with DMSO-treated cells (Fig. 4J and K). Moreover, EGFR/EZH2 siRNA codepletion completely blocked the increased response to EGF (Fig. 4K; Supplementary Fig. S4T). These data are consistent with a prior report noting that combined inhibition of EZH2 and EGFR was able to increase the sensitivity of colorectal cancer cells when compared with either drug alone (46). Our data suggest that this observation could be in part through EZH2i-induced EGFR DNA amplification. In support of this hypothesis, EZH2i treatment in two different colorectal cancer cell lines (HCT-15 and HT-29) resulted in increases in EGFR DNA copy number in these cells (Supplementary Fig. S4N–S4Q), which paralleled their sensitivity (46). Taken together, these data illustrate a critical role for K27 methylation, EZH2, and KDM6 in modulating EGFR locus amplification, expression, and, in turn, cellular response to either EGFRi or growth factors.

In a recent study, we demonstrated that H3K4 methylation enrichment at specific genomic loci was sufficient to recruit KDM4 family members to chromatin and, in turn, promote copy-number gains on chromosome 1 (e.g., 1q12h and 1p32.3; ref. 10). Upon evaluating H3K4me1/2/3 across the EGFR locus, we observed an inverse relationship between H3K4 methylation states and H3K27me3, which appears to flank the regions containing H3K4 methylation (Fig. 4B). Therefore, we hypothesized that inhibition of EZH2 promoted the KDM4A-driven EGFR copy gains through H3K4 methylation. Consistent with this hypothesis, RPE cells transduced with H3.3 K4M did not generate EGFR copy gains upon EZH2 inhibition (Fig. 5A; Supplementary Fig. S5A and S5B), which highlights the importance of this amino acid in promoting EGFR copy gains downstream of EZH2 depletion or inhibition. Furthermore, KDM4A was required for EZH2 depletion to generate EGFR amplification (Fig. 4C).

Because H3K4 methylation is key to KDM4A binding and copy gain generation at other specific sites in the genome (10, 18), we also tested whether H3K4M would block KDM4A-promoted EGFR copy gains. Indeed, H3K4M blocked KDM4A-driven EGFR amplification (Fig. 5B; Supplementary Fig. S5C and S5D). Consistent with these observations, HCC827 cells containing the EGFR amplification clouds that were reduced in size upon KDM4A depletion and inhibition (Fig. 3) also had a significant reduction EGFR cloud size with H3K4M transduction (Fig. 5C and D). A similar reduction was also observed with H3K4M transduction in the HT-29 cells that contain EGFR copy gains (Supplementary Fig. S5E and S5F). Furthermore, the Tudor domains that recognize H3K4 methylation within KDM4A (39) were required to generate EGFR copy gains (Fig. 2B). Taken together, these data collectively emphasize the importance of H3K4 methylation for generating the EGFR amplifications and highlight the need to balance K4/27 methylation states to modulate EGFR copy gains through KDM4A (Fig. 5E).

H3K4 KMTs were recently shown to be important in controlling the predilection of regions to amplify downstream of KDM4 members. For example, overexpression of each KMT2 family member promoted copy gains of specific chromosome 1 TSSG loci (10). Therefore, we overexpressed each H3K4 KMT (MLL1/KMT2A, MLL2/KMT2B, MLL4/KMT2D, SETD1A, and SETD1B; ref. 47) to determine whether all or select KMTs promote EGFR amplification (Supplementary Fig. S5G–S5I). Only KMT2A, SETD1A, and SETD1B were able to significantly promote EGFR copy gains (Fig. 5F). To date, the EGFR loci was the only target identified for SETD1A. Because EZH2 and the MLL family of KMTs oppose one another and their associated methylation states (48), we tested whether EGFR amplifications that occur downstream of EZH2 depletion are dependent on these H3K4 KMTs. When EZH2 and each KMT2 member (KMT2A, SETD1A, SETD1B) were codepleted, the EGFR copy gains were blocked, which highlights the importance of H3K4 methylation upon EZH2 depletion (Fig. 5G; Supplementary Fig. S5J and S5K).

Because specific H3K4 KMTs controlled EGFR amplification, we tested whether the same was true for the H3K4 KDMs. The KDM5 enzymes are H3K4 tridemethylases and have been shown to affect other TSSG sites (10). Depletion of the H3K4 tridemethylases will increase H3K4 methylation, thereby promoting copy gains of EGFR if copy gains of this locus are indeed dependent on H3K4 methylation. Therefore, we depleted each KDM5 member with at least two independent siRNAs before conducting cell-cycle profiles and EGFR FISH (Supplementary Fig. S5L and S5M). Only KDM5A depletion generated EGFR copy gains, whereas other genomic loci were unaffected (Fig. 5H; Supplementary Fig. S5N–S5P; ref. 10). Consistent with these genetic experiments, we observed enrichment for H3K4me3 upon ChIP-seq at an intergenic region within the EGFR structural domain that is 87 Kb upstream of the EGFR transcription start site (TSS; Fig. 5I, blue shaded region). Furthermore, we observed a loss of this H3K4me3 peak upon shMLL1 treatment in a published dataset (Fig. 5J; ref. 49) and enrichment upon shEZH2 treatment in a public dataset that was accompanied with reduced H3K27me3 (Fig. 5K; ref. 50). Moreover, RPE cells that overexpress KDM4A exhibited increases in H3K4me3 as well as a reduction in H3K9me3 at this same region (Fig. 5L and M, respectively). Taken together, these data suggest that KMT2 enzymes KDM5A and EZH2 modulate the balance of H3K4/27 methylation at the EGFR locus and, in turn, EGFR copy gains (Fig. 5N).

Our studies suggest that promoting H3K4 methylation through KDM5A depletion or inhibition would promote EGFR amplification through KDM4A. First, we evaluated the impact of KDM5 inhibitor (KDM5i) treatment on EGFR amplification [KDM5i (C70); refs. 51, 52]. KDM5i treatment promoted EGFR copy gains; however, they returned to baseline upon KDM5i washout (Fig. 5O; Supplementary Fig. S5Q), which emphasizes the reversibility of the amplifications as observed with EZH2i washout. We then used both genetic and chemical inhibition of KDM4 enzymes (KDM4i; ref. 43) to determine whether this would block EGFR amplifications generated upon either KDM5A siRNA depletion or KDM5i treatment (Fig. 5P–R; Supplementary Fig. S5R–S5U). In fact, KDM5 and KDM4 inhibitors can be used to generate and prevent EGFR copy gains (Fig. 5Q and R), which highlights the plasticity of the copy gains and the ability to therapeutically control DNA amplification with drugs targeting epigenetic factors.

Having therapeutic control of key growth factor receptors could have a profound impact on the ability to control cell proliferation and drug response when delivering inhibitors. Therefore, we tested whether KDM5i would alter cell proliferation when treated with supplemental EGF. Consistent with the previous experiments demonstrating that factors promoting EGFR copy gains increased EGF-associated proliferation, KDM5i increased cell proliferation with supplemented EGF when compared with KDM5i alone (Fig. 5S), which was also observed with KDM5A siRNA depletion and, in turn, was blocked with EGFR siRNA depletion (Fig. 5T; Supplementary Fig. S5V). Finally, we were also able to demonstrate that KDM5i increased sensitivity to Gefitinib when compared with KDM5i alone (Fig. 5U), which was consistent with a previous report (53). Taken together, these data highlight the ability to modulate H3K4 methylation to harness control of EGFR DNA copy levels and, in turn, growth factor and drug response.

Previous work from our laboratory demonstrated that hypoxia directly promoted TSSG formation of chromosome 1–associated loci (e.g., 1q12h) via stabilization of the KDM4A protein, through reduced association with the SKP1–CUL1–F-box (SCF) ubiquitin ligase complex (17). Consistent with this previous observation, 24 hours of exposure to hypoxia stabilized KDM4A in RPE cells and promoted EGFR copy gains that required KDM4A (Fig. 6A–D; Supplementary Fig. S6A–S6D). The hypoxia-induced copy gains of EGFR were transient in nature, in a similar manner to those observed with KDM5i and EZH2i. Indeed, moving cells back to normoxia following 24 hours of hypoxia exposure completely restored EGFR copy number to baseline levels (Fig. 6E and F). In agreement with the observation that KDM4A is required for hypoxia-induced EGFR DNA copy gains, pretreatment of cells with KDM4 family inhibitor prior to hypoxic exposure also blocks hypoxia-induced EGFR copy gains (Fig. 6G; Supplementary Fig. S6E). Furthermore, H3.3 K4M mutant–expressing cells did not generate the hypoxia-induced EGFR amplifications (Fig. 6H; Supplementary Fig. S6F and S6G). Taken together, these data demonstrate that hypoxia promotes EGFR copy gains in a similar fashion to KDM4A overexpression in that the amplifications require increased KDM4A levels and proper targeting via H3K4 methylation (Supplementary Fig. S6H).

While investigating hypoxia modulation of EGFR amplification, we also tested whether EGF supplementation would affect EGFR copy gains. A recent report demonstrated that cancer cells with EGFR amplification required EGF supplementation in the media to propagate the copy gains, which raised the possibility that EGF could directly promote EGFR copy gains (54). Consistent with this possibility, we demonstrated that treating cells for 24 hours with 50 ng/mL EGF, the preferred ligand of EGFR, results in significant copy-number gains of the EGFR locus (Fig. 6I and J). Furthermore, these gains were entirely dependent on KDM4A, as both siRNA-mediated depletion and pharmacologic inhibition of the KDM4 family was able to completely suppress these gains (Fig. 6I and J; Supplementary Fig. S6I–S6K). We also observed increased EGFR transcripts that were reduced upon KDM4 inhibition (Fig. 6K). Even though KDM4A was required for both the copy gains and increased expression, we did not observe increased KDM4A protein levels (Supplementary Fig. S6L), which suggests another pathway could be promoting KDM4A-dependent copy gains.

Given the importance of H3K4 methylation in targeting KDM4A so that TSSGs occur, we hypothesized that EGF was promoting the TSSGs via the H3K4 KMTs and, in turn, H3K4 methylation. Therefore, we first tested whether H3K4M would block EGF-induced EGFR copy gains. We observed that expression of the methyl-deficient H3K4M mutant was sufficient to block these growth factor–induced copy gains (Fig. 6L; Supplementary Fig. S6M and S6N). Therefore, we then depleted with two independent siRNAs each of the H3K4 methyltransferases that generated amplification of the EGFR locus (KMT2A, SETD1A, SETD1B; Fig. 5F; Supplementary Fig. S6O and S6P). Following depletion, cells were treated with EGF (50 ng/mL) for 24 hours and their EGFR copy number was assessed by DNA FISH. Depletion of KMT2A (MLL1) and SETD1A was able to completely inhibit EGF-induced copy gains of EGFR, whereas depletion of SETD1B had no impact on the amplifications (Fig. 6M). Thus, EGF treatment appears to promote EGFR copy gains through KMT2A, SETD1A, and H3K4 methylation, which illustrates how extrinsic cues can promote selective copy gains via specific methyltransferases. Taken together, these data suggest that two different extrinsic cues (hypoxia and EGF) are promoting EGFR copy gains through similar but distinct epigenetic mechanisms (Supplementary Fig. S6H).

In this study, we have identified a network of chromatin regulators and physiologic signals that influence EGFR copy gains. In the case of hypoxia, KDM4A stabilization mirrors overexpression and promotes EGFR amplification, whereas EGF appears to promote EGFR amplification through KMT2A/SETD1A and targeting KDM4A (Supplementary Fig. S6H). Therefore, multiple extracellular cues could promote parallel triggers for the amplifications, which raises the question of whether the combined signals elicit stronger amplification events. To test this possibility, cells were exposed to hypoxia for an initial 24 hours, followed by supplementation with EGF (50 ng/mL) under hypoxia conditions for an additional 24 hours. At this point, cells were harvested and their EGFR copy number assessed by DNA FISH. Hypoxia, in combination with EGF, resulted in a modest but additive increase in EGFR DNA copy number (Fig. 7A; Supplementary Fig. S7A). Moreover, an increased percentage of these cells demonstrated higher EGFR DNA copies per nucleus (≥5 copies; Fig. 7B and C). This observation suggests that hypoxia-induced KDM4A stabilization in concert with EGF-stimulated increases in H3K4 methylation promotes EGFR locus plasticity so higher DNA copy numbers are achieved. To strengthen this model, RPE cells that stably overexpress KDM4A, thus mimicking the phenotype observed upon hypoxic exposure, were also treated with EGF (Supplementary Fig. S7B). EGF treatment of KDM4A-overexpressing cells phenocopied the increase in EGFR amplification levels and increased DNA copies per nucleus that were observed in cells treated with hypoxia and EGF (Fig. 7A, D–F; Supplementary Fig. S7C). Moreover, in further support of this model, inhibition of KDM5 or EZH2, both of which increase H3K4me3 at the EGFR locus (Fig. 5I and K), enhances EGFR amplification when combined with hypoxic exposure (Supplementary Fig. S7D–S7I). Taken together, our data support a model by which physiologic triggers such as increased EGF concentration and/or hypoxia function in combination with epigenetic perturbation to directly modulate chromatin states and determine whether site-specific low or high copy DNA amplifications occur (Supplementary Fig. S7C).

To date, little knowledge exists about the molecular mechanisms that promote specific oncogene amplifications. In this study, we uncover epigenetic regulators and physiologic cues that facilitate amplification of the oncogene EGFR. Moreover, we provide some of the first evidence of the ability to rheostat an oncogenic amplification through therapeutic intervention. These data illustrate a molecular basis for EGFR amplification and establish that the extracellular microenvironment can directly contribute to DNA amplification heterogeneity in both normal and tumor cells. Furthermore, we demonstrate that these copy gains are transient and that combined cues and/or epigenetic factor manipulation are sufficient to promote higher copy-number amplifications. Overall, this study describes a series of key observations that demonstrate that oncogenic amplification is hardwired into cells, providing a definable basis for cellular plasticity for EGFR copy number in both normal and cancer cells, which has significant clinical implications.

Prior studies have illustrated that both somatic and tumor cells have ecDNA (11, 55), with key oncogenes such as MYC and EGFR occurring as ecDNA in as many as 50% of tumors (11). Early studies on extrachromosomal MYC demonstrated that the ecDNA harbored epigenetic states associated with active gene expression (56). Consistent with these observations, a recent study in somatic cells illustrated that ecDNAs were observed in gene-rich chromosomal regions (55), which suggested a relationship between their generation and actively marked loci. Our study provides additional evidence to these prior correlative observations by being able to directly promote or block such modifications through manipulation of histones, histone-modifying enzymes, and their upstream regulators such as hypoxia. The data presented within this article illustrate a critical role for KMT–KDMs in balancing the methylation states controlling the repressive state (H3K9/27 methylation) and more accessible, active states (H3K4 methylation) so that EGFR amplification is either blocked or promoted.

Methylation states appear to control the predilection of a region to amplify; however, not all enzymes controlling those states are responsible for generating the EGFR amplifications. For example, KDM4A and KDM5A were the only members within their lysine demethylase enzyme families to promote EGFR. However, the KDM6 family members KDM6A and KDM6B were both sufficient to generate EGFR amplification upon overexpression, which suggests that these enzymes could have functional redundancy at this locus in controlling H3K27me3 balance. These data highlight that enzyme families could have unique targets and, in certain cases, overlapping specificity.

Similar observations were also true for KMTs targeting H3K4/9 methylation. For example, KMT2A/MLL1, SETD1A, and SETD1B promoted EGFR amplification (Fig. 5). SETD1A was not responsible for previously identified TSSGs (10), highlighting the specificity for KMT2 family members in controlling genomic regions undergoing TSSG. Similarly, all H3K9 KMTs except KMT1C/G9a/EHMT2 promoted EGFR copy gains when depleted from cells. The EGFR locus has a clear pattern for the H3K9me1/2/3 distribution across the locus, which suggests an important arrangement for these methylation states and their associated KMTs. In fact, H3K9M introduction promoted EGFR low and higher level copy gains (Fig. 1). Therefore, future studies need to interrogate the dependencies of regions on the various members and establish regions that have unique targets or overlapping control.

Data presented here imply that KDM4A utilizes the same mechanism to generate EGFR amplification as other previously mapped regions undergoing copy gains (e.g., 1q21.3- CKS1B; refs. 10, 17, 18). KDM4A and KDM4B were shown to recruit the replication machinery and facilitate rereplication (10, 18). Our previous studies also demonstrated that H3K4 methylation was key to the recruitment of KDM4 family members and the modulation of chromosome 1 targets, which was driven by select H3K4 KMTs (10). These previous studies did not observe a role for H3K27 methylation in controlling the TSSG formation (10, 18). However, H3K27me3 was a key modulator of EGFR amplification. Indeed, EZH2 depletion and chemical inhibition promoted EGFR copy gains. EZH2 occupies the blocks of H3K27me3 that flank H3K4 methylation within the EGFR locus. Furthermore, H3K4 methylation interference or depletion of the KMT2 enzymes that control EGFR amplification completely blocked the EGFR copy gains generated by EZH2 suppression or inhibition. These data are consistent with the collection of studies illustrating an antagonistic relationship between these methylation states and the associated enzymes promoting H3K4/27 methylation balance (48). Our data are also consistent with a recent report demonstrating that cross-talk between EZH2 and KMT2A disrupts H3K27 methylation balance, resulting in resistance to EZH2i monotherapy (57). Therefore, common principles assigned to gene regulation could also be true for TSSG regulation, which has direct clinical implications.

Future studies need to evaluate the impact that combinations of histone modifications have on enzyme targeting and TSSG regulation. Studies should also consider the contribution of other genomic features (e.g., insulator elements, enhancers, CpG islands, etc.) in controlling these events. On the basis of HiC data and larger H3K9me3 and H3K27me3 blocks, we suspect that in addition to localized affects, higher-order structure is also key to regulating site-specific copy gains. Future studies need to explore the impact of altering higher-order genome organization at regions undergoing site-specific amplification. The EGFR locus would serve as an ideal region to evaluate these future questions.

Previous studies have illustrated EGFR amplification plasticity (12). EGFR copy gains can range from few in number to large clouds in the nuclei of cancer cells (see Fig. 3; ref. 42). Patients with glioblastoma who received EGFRi have been shown to develop resistance because ecDNA copies of EGFR disappear. However, upon drug removal, EGFR ecDNA recurs, reestablishing sensitivity to targeted therapy (12). This plasticity has also been illustrated in cell culture as EGFR copy gains disappear in some cell lines (e.g., glioblastoma cells). A recent study demonstrated, however, that supplementing EGF enables propagation of the EGFR amplifications (54). Therefore, there is a critical need to understand the mechanisms that promote or suppress this amplification event.

Consistent with the prior body of work suggesting certain extracellular cues (hypoxia and EGF) associate with tumor cells harboring EGFR amplification, we have now demonstrated that these stimuli directly control EGFR DNA copy number (Figs. 6 and 7). We observe that hypoxia-induced stabilization (17) and the catalytic activity of KDM4A is a central component in generating EGFR amplification. In fact, KDM4 chemical inhibition was sufficient to reduce this oncogenic amplification in hypoxia-stimulated cells as well as lung cancer cells harboring EGFR ecDNA. This observation is critical because it suggests that targeting KDM4A with small molecules could serve as a novel approach to control EGFR amplification and heterogeneity observed in hypoxic tumors. In addition, amplification of wild-type EGFR has been shown to drive acquired resistance to mutation-selective EGFR tyrosine kinase inhibitors, identifying an additional therapeutic arena for potentially deploying KDM4i therapy (58).

EGF treatment does not affect KDM4A levels but rather triggers EGFR amplification through two specific H3K4 KMTs, KMT2A/MLL1 and SETD1A. Although both KMT2A and SETD1A controlled EGF-induced copy gains, SETD1B was dispensable, which highlights that enzymes can be selectively required under certain physiologic conditions to generate DNA copy gains. Future studies should address whether other cellular signals or stresses could serve as important triggers to selectively activate or repress the enzymes required to generate amplification at EGFR and other TSSG sites. Understanding the triggers for amplifications will provide insights into tumor heterogeneity and uncover novel biomarkers and drug targets in controlling amplification.

Consistent with these two pathways (hypoxia and EGF) working in parallel, when hypoxia is combined with KDM4A, there is little change in the low copy number; however, when combined with increased H3K4 methylation there are higher copy-number gains of EGFR per nucleus. The same is true when EGF is combined with KDM4A overexpression. These data illustrate that combining extracellular cues and epigenetic factor alterations promotes EGFR copy-number generation and the degree of amplification. Given the range of EGFR amplification across tumors (Fig. 1A), these observations could be important when considering variables affecting DNA copy-number plasticity in tumors and when considering ways to therapeutically intervene.

Data presented within this article are suggestive of two populations of EGFR DNA amplifications. For example, in cancer cell lines which exhibit significant DNA amplification of EGFR (e.g,. HCC827 or HT-29 cells), these amplifications are reduced by modulation of K4 methylation via the introduction of a H3.3 K4M methyl–deficient mutant, or inhibition/depletion of KDM4A. However, despite a reduction in EGFR DNA amplification under these conditions, significant levels of EGFR amplification remain. Future studies will need to investigate this in more detail, but these results could indicate a balance between integrated DNA copy-number amplification and transient extrachromosomal amplifications of EGFR, the latter of which appears targetable with compounds directed toward epigenetic modifiers.

In closing, we have uncovered both chromatin modifiers and extracellular signals that control EGFR amplification and demonstrate that epigenetic therapies could hold a key to modulating EGFR copy-number heterogeneity in cancer and associated diseases, which could have significant clinical implications in the future.

RPE and 293T cells were cultured in DMEM-high glucose (Sigma) media supplemented with 10% heat-inactivated FBS, 100 U/mL penicillin, 100 μg/mL streptomycin, and 2 mM l-glutamine. HCC827 (lung adenocarcinoma), HCT-15 (colorectal adenocarcinoma), and HT-29 (colorectal adenocarcinoma) cells were cultured in RPMI-1640 Glutamax media (Gibco) supplemented with 10% heat-inactivated FBS, 100 U/mL penicillin, 100 μg/mL streptomycin, 1% sodium pyruvate, and 1% glucose. Cell line identities were authenticated by short tandem repeat analysis and Mycoplasma tested using the MycoAlert Detection Kit (Lonza, LT07-218).

Cells were plated in cell culture dishes and allowed to adhere for 16 to 20 hours in normoxia (5% CO₂, 21% O₂, 74% N₂). For hypoxic treatment, cells were maintained in a HERA Cell 150 incubator (Thermo Fisher Scientific) flushed with 5% CO₂ and 1% O₂ and balanced with N₂ for the duration of the experiment. Cells were cultured in hypoxia for between 24 and 48 hours, prior to harvesting and downstream experimental processing. Hypoxic culture conditions were validated by immunoblot detection of carbonic anhydrase IX (CA-IX) expression, a well-established hypoxic biomarker signature (17).

Cells were plated in cell culture dishes and allowed to adhere for 16 to 20 hours. Cell culture medium was removed and cells rinsed with PBS prior to siRNA transfections (5 nM–10 nM/transfection) being performed using Lipofectamine 3000 transfection reagent (Life Technologies) in Opti-MEM medium (Life Technologies). Transfections were changed to complete cell culture media after 4 hours of transfection, and cells were collected 72 hours post-transfection. For cotransfection experiments, both siRNA sequences were transfected at the same time. For experiments involving hypoxia or drug treatment, cells were exposed to these conditions at least 48 hours post-siRNA transfection.

Transient overexpression transfections were performed using Lipofectamine 3000 transfection reagent and P3000 reagent (Life Technologies) in OPTI-MEM medium for 4 hours, followed by changing to complete media. Silencer select negative controls and siRNAs were purchased from Life Technologies. Their sequences and unique identification numbers are tabulated in Supplementary Table S1. At least two different siRNAs against every gene were used for each experiment.

Plasmids for H3.3 wild-type (WT), K4M, K9M, K27M, and K36M mutants were provided by Peter Lewis (University of Wisconsin). Virus was generated by cotransfection of specific plasmids along with the packaging plasmids (AmphoPAK and VSVG) in 293T cells, using Lipofectamine 3000 transfection reagent (Life Technologies). DNA transfection was performed in Opti-MEM medium (Life Technologies) overnight. The virus-containing supernatant was collected after 24 hours. RPE cells were infected in the presence of 8 mg/mL polybrene for 12 hours with the viral supernatant (18). Cells were washed two times with DMEM prior to being cultured in complete medium for 48 hours before harvesting. For hypoxia, drug or growth factor treatment and overexpression experiments involving H3.3 K4M, RPE cells were infected as described above and cultured for 24 hours postinfection before additional treatment (hypoxia, drug, or growth factor treatment, or KDM4A transient overexpression). Cells were harvested 24 hours later. For all experiments involving Histone H3.3 variants, RPE, HT-29, and HCC827 cells were harvested 48 hours after viral infection for Western blot, cell-cycle profiles, and DNA FISH analysis.

Cells were washed and collected by trypsinization, followed by washing in PBS two times. Cell pellet was resuspended in Qiazol reagent (QIAGEN) and stored at −80°C before further processing. Total RNA was extracted using miRNeasy Mini Kit (QIAGEN) with an on-column DNase digestion according to the manufacturer's instructions. RNA was quantified using NanoDrop 2000 (Thermo Fisher Scientific). Single-strand cDNA was prepared using Super Script IV First Strand Synthesis Kit (Invitrogen) using random hexamers. Expression levels were analyzed using FastStart Universal SYBR Green Master (ROX; Roche) according to the manufacturer's instructions on a LightCycler 480 PCR machine (Roche). Samples were normalized to β-actin. Primer sequences are provided in Supplementary Table S2.

Cells were trypsinized and washed two times with PBS before resuspending in RIPA lysis buffer (50 mM Tris pH 7.4, 150 mM NaCl, 0.25% sodium deoxycholate, 1% NP40, 1 mM EDTA, 10% Glycerol) freshly supplemented with protease inhibitor and PhosSTOP phosphatase inhibitor cocktails (Roche). Cells were lysed on ice for 15 minutes and stored at −80°C until further processing. Lysates were sonicated for 15 minutes (30 seconds ON and 30 seconds OFF cycle) at 70% amplitude in QSonica Q700 sonicator (Qsonica) followed by centrifugation at 12,000 rpm for 15 minutes. Cell lysate was transferred to a fresh tube and protein estimations were performed with Pierce BCA reagent (Thermo Fisher Scientific). Equal amounts of proteins were separated by SDS gel electrophoresis and transferred on nitrocellulose membrane (BioTrace NT, Pall Life Sciences) for at least 3 hours at a constant current. The membranes were blocked for at least 1 hour in 5% BSA-PBST (1× PBS with 0.5% Tween-20) or 5% milk-PBST and probed overnight with specific antibodies as follows at the following dilutions: anti-GFP (Neuro mAb; 1:100); anti-β-actin (1:10,000); anti-KDM4A (Neuro mAb; 1:100); anti-FLAG (Sigma Aldrich; 1:500); anti-KDM5A (Abcam; 1:500); anti-HALO (Promega; 1:1,000); anti-Carbonic Anhydrase IX (Abcam; 1:1,000). Catalog numbers for all antibodies used in this study can be found in Supplementary Table S3. Membranes were washed three times in PBST the next day, incubated with goat anti-mouse IgG peroxidase–conjugated secondary antibody (170-6516, Bio-Rad) or goat anti-rabbit peroxidase–conjugated secondary antibody (A00167, GenScript) at 1:2,500 in 5% milk-PBST for at least 1 hour at room temperature, washed three times with PBST and incubated in Lumi-Light Western blotting substrate (12015200001, Roche) or SuperSignal West Pico PLUS Chemiluminescent substrate (34577, Thermo Fisher Scientific) for 2 to 4 minutes. Membranes were developed with Lumi-Film Chemiluminescent detection film (11666657001, Roche). The Western blot images displayed in the figures have been cropped and autocontrasted.

Samples were washed with PBS, centrifuged at 1,400 rpm for 5 minutes, and permeabilized with 500 mL PBS containing 0.5% Triton X-100 for 30 minutes. After this incubation, cells were washed with PBS and centrifuged at 1,400 rpm for 5 minutes. Samples were then stained with 1:100 dilutions of 1 mg/mL PI solution and 0.5 M EDTA with 100 mg RNase A, overnight at 4°C. Cell-cycle distribution was analyzed by flow cytometry using the LSRII Flow Cytometry System (BD Biosciences).

The FISH protocol was performed as described previously in Black and colleagues (2013; ref. 18). Briefly, cell suspensions were fixed in ice-cold methanol:glacial acetic acid (3:1) solution for a minimum of 4 hours, before being spun onto 8 Chamber Polystyrene vessel tissue culture–treated glass slides (BD Falcon, Thermo Fisher Scientific), using a centrifuge at 900 rpm. The slides were airdried and incubated in 2× SSC buffer for 2 minutes, followed by serial ethanol dilution (70%, 85%, and 100%) incubations for 2 minutes each, for a total of 6 minutes. Air-dried slides were hybridized with probes that were diluted in appropriate buffer overnight at 37°C, following a 4-minute incubation on a heat block at 78°C. The slides were washed the next day for 3 to 4 minutes in appropriate wash buffers at 69°C with 0.4× SSC for Cytocell probes or commercially available Agilent wash buffer 1 followed by washing in 2× SSC with 0.05% Tween-20 (Cytocell probes) or commercially available Agilent wash buffer 2 (Agilent probes). The slides were incubated in 1 mg/mL DAPI solution made in 1% BSA-PBS, followed by a final 1× PBS wash. After the wash, the slides were mounted with ProLong Gold Antifade Reagent (Invitrogen).

FISH images were acquired using an Olympus IX81 or Olympus IX83 spinning disc microscope at 40× magnification and analyzed using Slidebook 6.0 software. A minimum of 25 z-planes with 0.5-μm step size were acquired for each field. Copy-number gains for EGFR, 7 centromere (7C), 7p telomere (7p22.3), IKZF1 (7p12.2), and 8 centromere (8C) were scored in RPE cells as three or more foci. A minimum of 100 nuclei are scored for each independent experiment.

HCC827 cells have many EGFR amplifications that present as large EGFR amplification clouds (42). Therefore, the length of the EGFR DNA amplification cloud(s) was measured at its longest point, using the measuring tool within the Slidebook software. If multiple amplification clouds were present in a single nucleus, each cloud was measured. Each measurement was plotted and comparisons made between the overall size of the amplification cloud (μm) in cells treated with siRNA to KDM4A, an inhibitor to the KDM4 family or transduced with either H3.3 WT or K4M. The analysis represents data from more than 400 nuclei from two independent experiments with two different siRNAs to KDM4A, or across three independent experiments for the KDM4 inhibitor treatment and H3.3 transduction experiments. Each treatment condition is compared with either a nontargeting siRNA control or a DMSO vehicle control.

Control or stable KDM4A overexpression RPE cells were plated in 24-well tissue culture plates at a density of 5 × 10³. Cells were allowed to adhere for approximately 16 hours before Lapatinib (Abcam; dissolved in DMSO) was supplemented to media to a final concentration of 1 μM. Cells were cultured in Lapatinib for a total of 48 hours before harvesting. Cells were stained with Trypan blue solution (Sigma Aldrich) to assess cell viability and counted using a hemocytometer. Each condition was plated in triplicate wells and each well was counted in duplicate. An average was taken of all triplicates and used as a representative total. Data are displayed in Fig. 2G, in a comparable manner to Nathanson and colleagues (2014; ref. 12).

Control or stable KDM4A overexpression RPE cells were plated in 24-well tissue culture plates at a density of 8 × 10³. Cells were allowed to adhere for approximately 16 hours before Gefitinib (Abcam; dissolved in DMSO) was supplemented to media to a final concentration of 1, 2.5, or 5 μM. Cells were cultured in Gefitinib for a total of 48 hours before harvesting. Cells were stained with Trypan blue solution (Sigma Aldrich) to assess cell viability and counted using a hemocytometer. Each condition was plated in triplicate wells and each well was counted in duplicate. An average was taken of all triplicates and used as a representative total. Data are displayed in Fig. 2H, in a comparable manner to Nathanson and colleagues (2014; ref. 12).

Control, KDM4A overexpression, or WT parental RPE cells were plated at a density of 5 × 10³ per well in a 6-well tissue culture plate. Each condition was plated in triplicate for each independent experiment. Cells were allowed to adhere for 16 hours before fresh complete media was added containing a final concentration of 50 ng/mL human recombinant EGF (Abcam). For control and KDM4A overexpression RPE cells, cells were harvested after 48 hours of EGF treatment and cell number calculated using a hemocytometer.

For combinatorial drug experiments using parental RPE cells, EGF was added for 24 hours, following a 24-hour drug treatment with either KDM5i (1 μM) or EZH2i (3 μM).

For siRNA conditions, parental RPE cells were plated in 10-cm tissue culture plates at a density of 3 × 10⁵. Cells were allowed to adhere for approximately 16 hours, prior to siRNA transfection (as described previously). Twenty-four hours post siRNA transfection, cells were replated in triplicate into 24-well tissue culture plates at a density of 8 × 10³. The remaining cells were replated and harvested 24 hours later for RNA extraction and qPCR transcript analysis. Cells were allowed to adhere in the 24-well plates for 24 hours before media was supplemented with human recombinant EGF to a final concentration of 50 ng/mL EGF. Cells were cultured in EGF for 24 hours before harvesting and counting using a hemocytometer, as described previously.

Control or stable KDM4A overexpression RPE cells were plated in 6-well tissue culture plates at a density of 2 × 10⁵. Cells were allowed to adhere to plates for 24 hours. After 24 hours in culture, a p200 pipette tip was used to introduce a scratch wound in the center of the well from the 12 o'clock to 6 o'clock position. Following the induction of the scratch wound, media were removed from each well and 1 mL of DMEM was used to rinse the wells, removing any cellular debris. After this wash, 3 mL of DMEM supplemented with vehicle or human recombinant EGF (Abcam) to a final concentration of 50 ng/mL was added to each well. Cells were imaged at 0 hours, 12 hours, and 24 hours, using the EVOS imaging platform at 4× magnification. Scratch wound measurements were performed using the EVOS software with a minimum of 5 measurements taken at various locations, per scratch wound. All measurements were averaged. Each condition was performed in triplicate for each independent experiment.

Human recombinant EGF was added to histone H3.3 WT- or K4M-expressing RPE cells, 24 hours after viral transduction. Cells were cultured in EGF for 24 hours before harvesting for RNA, protein, cell cycle, and DNA FISH analysis.

HCC827 cells were treated at approximately 80% confluency with KDM4 inhibitor at a final concentration of 5 nM for 48 hours before being harvested for RNA, protein, and DNA FISH analysis.

For hypoxia, EGF and KDM5i combinatorial experiments, RPE cells were pretreated with 1 nM of KDM4 inhibitor exactly 24 hours prior to hypoxic exposure or EGF treatment. Immediately before transferring cells to hypoxia, or before drug or growth factor treatment, KDM4 inhibitor was supplemented to cells again at a concentration of 1 nM (double spike). Cells were then cultured in the respective conditions for 24 hours prior to harvesting for RNA, protein, cell cycle, and DNA FISH analysis.

RPE cells were treated with KDM5 inhibitor (C70) at a final concentration of 1 μM for a total treatment time of 48 hours.

For experiments involving combination treatment with KDM4i, cells were pretreated with KDM4i (1 nM) for 24 hours. After this 24-hour treatment, KDM5 inhibitor was supplemented along with an additional dose of KDM4 inhibitor at doses of 1 μM and 1 nM, respectively. Cells were harvested 48 hours after combination treatment.

For experiments involving combinations of KDM5 inhibitor and hypoxic exposure, RPE cells were treated with 1 μM KDM5 inhibitor for 24 hours. After this 24-hour treatment, cells were transferred to hypoxic conditions (1% O₂) for an additional 24 hours prior to harvesting for RNA, protein, cell cycle, and DNA FISH analysis.

Parental RPE cells were plated in 24-well tissue culture plates in triplicate at a density of 5 × 10³. Cells were allowed to adhere for approximately 16 hours before KDM5 inhibitor and Gefitinib alone or in combination were supplemented to each well at a final concentration of 1 μM and 2.5 μM, respectively. Cells were cultured for a total of 72 hours under drug treatment conditions before being harvested, stained with trypan blue, and counted using a hemocytometer.

Parental RPE cells were treated with 1, 3, or 5 μM of EZH2 inhibitor for a total treatment duration of 72 hours before being harvested for RNA, protein, cell cycle, and DNA FISH analysis. HT-29 and HCT-15 cells were treated with 3 μM EZH2i for 48 hours before being harvested for RNA, protein, cell cycle, and DNA FISH analysis.

Parental RPE cells were plated in 24-well tissue culture plates in triplicate at a density of 5 × 10³. Cells were allowed to adhere for approximately 16 hours before media was supplemented with EZH2 inhibitor (C24) at a final concentration of 3 μM. After an initial 24-hour treatment, human recombinant EGF was supplemented to each well at a final concentration of 50 ng/mL. After 48 hours of EGF treatment, cells were harvested and counted using a hemocytometer.

Parental RPE cells were virally transduced with either histone H3.3 WT or K4M constructs, as described previously. Twenty-four hours after viral transduction, EZH2 inhibitor was supplemented to cells at a final concentration of 3 μM. Cells were treated under these conditions for 24 hours, followed by harvesting for RNA, protein, cell cycle, and DNA FISH analysis.

Parental RPE cells were pretreated with EZH2 inhibitor at a final concentration of 3 μM for 24 hours, prior to being transferred to hypoxic conditions for an additional 24 hours. Cells were then harvested and processed for RNA, protein, cell cycle, and DNA FISH analysis.

Cells were incubated with Hoechst 33342 (Thermo Fisher Scientific H3570) at 1/1,000 directly into the media for 1 hour at 37°C. Cells were then trypsinized and resuspended in media containing Hoechst at 1/1,000. Cells were sorted with a BD FACS Fusion using the laser BV421-A into Qiazol, based on DNA content. Late S-phase RNA was purified from cells using the Qiagen miRNeasy Kit including a DNAse treatment. Total RNA-sequencing libraries were prepped using the TruSeq Stranded Total RNA Sample Preparation with Ribo-Zero Kit (Illumina). Libraries were paired-end sequenced (150 cycles each way) using a NextSeq500 (Illumina). STAR (59) aligner was used to map sequencing reads to transcripts in human hg19 reference genome. Read counts for individual transcripts were produced with HTseq-count (60), followed by the estimation of expression values and detection of differentially expressed transcripts using edgeR (61).

Cells were incubated with Hoechst 33342 (Thermo Fisher Scientific H3570) at 1/1,000 directly into the media for 1 hour at 37°C. Cells were then trypsinized and resuspended in media containing Hoechst at 1/1,000 before cross-linking with 1% formaldehyde for 13 minutes at 37°C and quenching with 0.125 M glycine. Cells were washed with 1× PBS and resuspended in media containing Hoechst (1/1,000). Cells were sorted with a BD FACS Fusion using the laser BV421-A based on DNA content. For siKDM5A ChIP, cells were harvested as described previously (10, 18). Sonication of chromatin was done with the Qsonica Q800R2 system (Qsonica). Chromatin in a range of 0.5–10 μg was used based on DNA content (nanodrop concentrations) with the following antibodies: H3K4me1 (Abcam ab8895), H3K4me2 (Abcam ab32356), H3K4me3 (Millipore 07-473), H3K9me1 (Abcam ab8896-100), H3K9me2 (Abcam ab1220), H3K9me3 (Abcam ab8898). ChIP-seq libraries were prepped using the TruSeq ChIP Sample Preparation Kit (Illumina). Libraries were single-end sequenced (75 cycles) using a NextSeq500 (Illumina).

ChIP-seq data for the cells with KDM4A overexpression and the corresponding controls were based on merged samples for multiple points of the cell cycle. Datasets for MLL knockdown samples (GSE81795; ref. 49) and EZH2 knockdown samples (50) with their respective controls were retrieved from the Gene Expression Omnibus (GEO). Sequencing reads were aligned against the human hg19 reference genome using BWA (62). Alignments were filtered for uniquely mapped reads and duplicates were removed. Input-normalized ratio coverage tracks were generated using Deeptools (63).

The ENCODE ChIP-seq data on histone modification enrichment were downloaded from the ENCODE website (www.encodeproject.org). These data were normalized by the ENCODE pipeline (64). Other public ChIP-seq datasets were downloaded as fastq files from GEO (GSE64243, GSE118954, GSE81795; ref. 65), followed by mapping to the hg19 reference genome using BWA (66) and the generation of input-normalized coverage tracks using Deeptools (63).

Data from TCGA were obtained from the Broad Institute's Genomic Data Analysis Center Firehose (http://gdac.broadinstitute.org). A set of 7,069 tumors spanning 21 tumor types was analyzed. Expression values for EGFR were extracted from the database in units of transcripts per million (TPM). Copy–number values for EGFR were also extracted from the database. Figure 1A shows these EGFR expression and copy–number values for each tumor analyzed.

All pairwise comparisons were done using two-tailed Student t test unless otherwise stated. Significance was determined if the P value was ≤ 0.05. All FISH experiments were carried out with at least two independent siRNAs unless otherwise stated, and at least 100 nuclei per replicate per experiment were counted for all the FISH studies conducted. All error bars represent the SEM.

J.R. Whetstine has received funding from AstraZeneca, is a consultant at Qsonica, previously consulted for Celgene, and served on the Salarius Pharmaceuticals advisory board. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T.L. Clarke, J.R. Whetstine

Development of methodology: T.L. Clarke, C. Van Rechem, J.R. Whetstine

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T.L. Clarke, R. Tang, D. Chakraborty, C. Van Rechem, F. Ji, S. Mishra, J. Jin

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T.L. Clarke, F. Ji, S. Mishra, M.S. Lawrence, R.I. Sadreyev, J.R. Whetstine

Writing, review, and/or revision of the manuscript: T.L. Clarke, D. Chakraborty, C. Van Rechem, F. Ji, S. Mishra, H.U. Kaniskan, R.I. Sadreyev, J.R. Whetstine

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T.L. Clarke, A. Ma, J.R. Whetstine

Study supervision: T.L. Clarke, R.I. Sadreyev, J.R. Whetstine

Other (providing small-molecule inhibitors used in the studies): A. Ma

Other (synthesis and characterization of the inhibitors used in the study): H.U. Kaniskan

We are grateful to Ravi Mylvaganam and the Massachusetts General Hospital (MGH) Flow Cytometry Core for assistance with flow cytometry. We thank Peter Lewis (University of Wisconsin) for the H3K–M constructs. We also thank Reuben Duttweiler for technical contributions. We would like to thank Drs. Margie Clapper, Jon Chernoff, and Hossein Borghaei for comments on the manuscript. Work related to this study is supported by the R01GM097360 (to J.R. Whetstine) and the NIH/NCI Cancer Center Support Grant P30 CA006927. J.R. Whetstine was a Tepper Family Massachusetts General Hospital Scholar and Leukemia and Lymphoma Scholar, as well as a recipient of an American Lung Association Lung Cancer Discovery Award. R.I. Sadreyev was supported by NIH P30 DK040561. J. Jin acknowledges support by grant R01GM122749 from the NIH. T.L. Clarke is supported by an EMBO long-term fellowship (EMBO ALTF 449-2017). S. Mishra was supported by a Senior Research Training Fellowship from the American Lung Association and the MGH ECOR Tosteson Postdoctoral Training Fellowship. D. Chakraborty is a recipient of an award from the Ovarian Cancer Research Fund Alliance (Ann and Sol Schreiber Mentored Investigator Award-543667).