Assessment of Histone Tail Modifications and Transcriptional Profiling During Colon Cancer Progression Reveals a Global Decrease in H3K4me3 Activity

Abstract

During colon cancer, epigenetic alterations contribute to the dysregulation of major cellular functions and signaling pathways. Modifications in chromatin signatures such as H3K4me3 and H3K9ac, which are associated with transcriptionally active genes, can lead to genomic instability and perturb the expression of gene sets associated with oncogenic processes. In order to further elucidate early pre-tumorigenic epigenetic molecular events driving CRC, we integrated diverse, genome-wide, epigenetic inputs (by high throughput sequencing of RNA, H3K4me3, and H3K9ac) and compared differentially expressed transcripts (DE) and enriched regions (DER) in an in-vivo rat colon cancer progression model. Carcinogen (AOM) effects were detected genome-wide at the RNA (116 DE genes), K9ac (49 DERs including 24 genes) and K4me3 (7678 DERs including 3792 genes) level. RNA-seq differential expression and pathway analysis indicated that interferon-associated innate immune responses were impacted by AOM exposure. Despite extensive associations between K4me3 DERs and colon tumorigenesis (1210 genes were linked to colorectal carcinoma) including FOXO3, GNAI2, H2AFX, MSH2, NR3C1, PDCD4 and VEGFA, these changes were not reflected at the RNA gene expression level during early cancer progression. Collectively, our results indicate that carcinogen-induced changes in gene K4me3 DERs are harbingers of future transcriptional events, which drive malignant transformation of the colon.

1. Introduction

Colorectal cancer (CRC) ranks as the third leading cause of cancer death among adults. Every year in the United States, more than 150,000 cases of colorectal cancer are diagnosed and approximately 57,000 patients die of the disease (1). During carcinogenesis, major cellular functions and pathways, including drug metabolism, cell cycle regulation, potential to repair DNA damage or to induce apoptosis, response to inflammatory stimuli, cell signaling, and cell growth control and differentiation, are dysregulated (1). Epigenetic alterations contribute to these cellular defects. For example, the epigenetic modulation of master transcription factors by promoter methylation and modification of histones and non-histone proteins lead to genomic instability and perturb the expression of gene sets associated with cell adhesion and apoptosis (2–5). From a mechanistic perspective, chromatin signatures are tightly linked to epigenetic regulation. For instance, transcriptionally active genes are characterized by active chromatin marks, such as trimethylated histone H3 lysine 4 (K4me3) and acetylated histone H3 lysine 9 (K9ac) (2,5). Alterations in these histone modifications can drive oncogenic processes, such as proliferation, invasion, angiogenesis, and dedifferentiation, by perturbing normal gene expression patterns (5). This is particularly relevant to CRC, because altered K4me3 levels are associated with the onset of colorectal cancer (6).

Current evidence indicates that genetic mutations, epigenetic changes, and aberrant immunologic signaling pathways are major contributors to CRC (7). However, until recently, appreciation of epigenetic alterations has lagged behind genetic mutations with regard to their contributions to human cancer development. To date, preclinical animal models as well as human studies that elucidate the pretumorigenic epigenetic molecular events driving CRC are limited. Colon cancer profiles for H3K4me3 and H3K9ac enrichment and other histone modifications have been previously generated, however, in most cases cancer cell lines and/or genome-limited methodology were used (6,8). In addition, few studies to date have integrated diverse epigenetic inputs in an in vivo model of CRC. This type of analysis would improve the statistical and interpretative power of the changes in transcription and chromatin state during cancer progression. Since distinct genomic and epigenetic events drive the initiation, promotion and progression of colon cancer (1), in this study we integrated global chromatin immunoprecipitation sequencing (ChIP-seq) and RNA-seq data in order to explore the progression of colon cancer in a rat model at a genome-wide epigenetic level. The azoxymethane (AOM) chemical rodent carcinogenesis model was utilized because it serves as one of the most definitive means of assessing mechanisms related to human colon cancer risk (9,10). We have previously demonstrated that at 10 weeks post AOM injection, the colonic mucosa is precancerous, equivalent to the progression stage of colon cancer (11).

2. Methods

2.1. Animals

Sixty-seven weanling male Sprague Dawley rats (Harlan, Houston, TX) were individually housed and acclimated for one week in the same room, maintained in a temperature and humidity-controlled animal facility with a daily 15 h light/9 h dark photoperiod. The animal use protocol was approved by the University Animal Care Committee of Texas A&M University and conformed to NIH guidelines. The study examined the effects of two treatments (AOM or saline control). Animals were stratified by body weight after the acclimation period so that mean initial body weights did not differ. After 1-week acclimation on standard pelleted diet, rats were assigned to one of four diet groups which differed in type of fat plus fiber as previously described, (12) see Supplementary Methods for additional details. Body weight and food intake were monitored during the study.

2.2. Carcinogen Treatment

After two weeks of feeding, 24 rats were injected with saline (control), and 43 rats were AOM (Sigma, St. Louis, MO) injected s.c. at 15 mg/kg body weight. Equal numbers of animals from each diet group were randomly selected (n=3) from saline and (n=10–11) AOM injected rats. Each rat subsequently received a second AOM or saline injection one week later and animals were terminated 10 weeks after the first AOM injection.

2.3. Aberrant Crypt Foci Scoring

Immediately after removal, colons (11 per diet group with AOM injection and two per diet with saline injection) were flattened between Whatman one filter paper and fixed in 70% ethanol for 24 h. The number of high multiplicity ACF (more than three aberrant crypts per foci) was scored as previously described (11). Half of the colon was used for ACF scoring while isolation of colonic crypts as described below was performed on the remaining half. Phenotypic data were analyzed using ANOVA to determine the effect of carcinogen on HM-ACF.

2.4. Isolation of colonic crypts

The large intestine was resected from the junction between the cecum and the rectum, and was opened longitudinally and washed in 1× PBS. Subsequently, the visible “herringbone” folds were used to identify the proximal colon. Colonic crypts were extracted from the distal region (distal colon) as previously described (13), and an aliquot of the isolated crypts was subsequently used to generate mRNA expression profile libraries. The remaining crypts were immediately crosslinked for ChIP analysis (see Supplementary Methods for details).

2.5. Western Blot

Colonic crypt nuclear protein was analyzed by immunoblot as previously described (13) (see Supplementary Methods for details). Primary antibodies were used to detect histone H3 tri methyl K4 (Active Motif 39160), H3 acetyl K9 (ab10812), and histone H3 (ab1791) levels. Peroxidase conjugated goat anti-rabbit IgG was purchased from Kirkegaard and Perry Laboratories (Gaithersburg, MD).

2.6. Chromatin immunoprecipitation

ChIP-seq analyses were performed in order to determine global histone mapping in crypts isolated from the distal colon. The ChIP protocol described by Triff et al. was utilized (13) with one modification, cells were cross-linked by adding freshly prepared formaldehyde at a 1% concentration for 15 min at room temperature (see Supplementary Methods for details). ChIP antibodies included: ChIP Grade (Active Motif 39160) anti-histone H3 (tri methyl K4) antibody, ChIP Grade anti-histone H3 acetyl K9 antibody (ab10812). The specificities of all antibodies were tested by Western blot and ChIP-qPCR (data not shown). Equal amounts of (200–500 bp) ChIPed DNA from 2–3 AOM rats with the same dietary treatment (biological replicates) were pooled into 16 barcoded groups (representing 43 individual rats), and the saline biological replicates were similarly pooled into 12 barcoded groups (representing 24 individual rats) prior to high throughput sequencing. Due to limiting cell numbers it was not possible to use cells exclusively isolated from ACF, therefore all crypts from each distal colon were ChIPed.

2.7. ChIP sequencing

BioScientific NETflex (ChIPseq kit 5143-01, Barcodes kit 514120) multiplex libraries from ChIPed DNA (10 nM) were sequenced using an Illumina HiSeq 2000 DNA Sequencer. Sequence reads with poor quality bases and with adaptors or other contaminants were filtered. The remaining reads (>290 million total per sample) were mapped to the reference rat genome (rn4) with commonly used Burrows-WheAligner (BWA) for Illumina (version 1.2.3) settings and only non-identical uniquely mapped reads were retained. The peak caller program MACS (version 1.4.1) (14) was used to identify peaks/islands. Islands (enriched regions) were defined as the genomic areas enriched with the ChIPed protein (peaks aka enriched regions) in at least one sequenced sample (using merge function of BEDTools (15)), and reads were quantified using coverageBed function of BEDTools (15). The UCSC Genome Browser was used to visualize bigwig data tracks. The nearest gene to each island, i.e., within five kb of the island was identified using closestBed from the BEDTools software suite (15) and the refGene table downloaded from the UCSC Genome Browser for the Baylor 3.4/rn4 assembly files.

Regions showing differences in histone modification were identified using the edgeR package (16,17) for the R software environment (14,16). In order to increase the statistical power of our analysis (higher number of samples per treatment) and detect key AOM effects associated with cancer progression, rats were pooled across the various diet groups described in the methods section “Chromatin Immunoprecipitation”. Using this approach, we focused on the main effects of colon cancer progression on histone tail modifications and transcriptional profiling. Read counts per gene were normalized using the scaling factor method of Anders and Huber (18). Differential expression testing of genes was performed using likelihood ratio tests on the negative binomial GLMs estimated by edgeR, accounting for biological variability (16,17). By formulating the linear predictor of the GLM in either cell means or factor effects form (19), we tested for significance of both main and overall effects of the different factors in addition to testing for significant treatment combinations. Regions with FDR < 0.1 and minimal threshold of one count per million mapped reads in at least four samples were selected as differentially enriched regions (DERs).

2.8. RNA isolation

For total RNA isolation, colonic crypts were homogenized on ice in lysis buffer (RNAqueous Isolation kit, Ambion) and frozen at −80°C until RNA was isolated. Subsequently, total RNA was isolated using the RNAqueous kit, followed by DNase treatment. RNA integrity was analyzed on an Agilent Bioanalyzer to assess RNA integrity.

2.9. RNA sequencing

Total RNA (1000 ng) was used to generate multiplex libraries for whole-transcriptome analysis following Illumina’s TruSeq RNA v2 sample preparation protocol. Libraries from 24 individual rats per treatment were sequenced on an Illumina HiSeq 2000. At least 151 million, 50 bp single-end reads per treatment were obtained for each sample. Reads were mapped with the STAR aligner using the default parameters and genome assembly (20). More than 85% of reads aligned uniquely to the rat genome. Genes that did not have at least one read count per million mapped reads in at least four samples were removed. Read counts per gene were normalized using the scaling factor method of Anders and Huber (18). The read counts were modeled directly using negative binomial distribution and generalized linear models accounting for the differences in diet and subsequently fit with the R programming language (14) and Bioconductor package edgeR (16,17). Differential expression was then tested using likelihood ratio tests involving the fitted models, accounting for biological variability (17). Genes with false discovery rate adjusted p-value (FDR) less than 0.1 were selected as differentially expressed transcripts (DE).

2.10.Non-coding RNA protocol

Non-coding RNAs were identified using the lncRNApipe (https://github.com/biocoder/Perl-for-Bioinformatics/releases) software. Reads were first trimmed using Trimmomatic. At least 90% of the reads were retained after trimming and filtering, which were then aligned to rn4 genome assembly using tophat (21). Over 50% of the reads were uniquely mapped to the genome. Replicate data were merged and transcripts assembled using cufflinks (21) were then examined to identify and catalog putative long and middle non-coding RNAs into five categories; 214 long intergenic lncRNAs (LincRNAs), 36 intronic contained lncRNAs (Incs), 0 partially overlapping lncRNAs (Poncs), 0 completely overlapping lncRNAs (Concs) and 0 exonic overlaps when compared to known RefSeq gene annotation of rat (rn4) which contains both protein-coding and non protein-coding genes. FASTA sequences were then created for any non-overlapping catalogued putative middle and long non-coding RNA with intron regions spliced out to run CPC software (22). For any transcripts identified as “noncoding” by CPC, BLAST homology searches were performed against hg38, mm10 and rn6 RefSeq genes from UCSC to eliminate the possibility of reporting already known ncRNAs in other closely related species as putative novel ncRNAs. Finally, RNAfold (23) and INFERNAL 1.1 (24) were used to predict minimum free energy structure of the putative novel ncRNAs and to assign any possible ncRNA families (see Supplementary Methods for details and references).

2.11. Ingenuity pathway analyses

“Functional enrichment” analysis was performed using Ingenuity Pathway Analysis (IPA) version 2.0 software (Ingenuity Systems Inc., Redwood City, CA) as we have previously described (13) (see Supplementary Methods for details).

Data access

Sequencing data have been deposited in the GEO database under the accession number GSE87525.

3. Results

In an effort to identify carcinogen-induced genes that contribute to CRC progression, transcriptomic and histone tail H3K4me3 and H3K9ac alterations in early precancerous colonic epithelial cells were examined. Gene expression profiling by Next Generation Sequencing of RNA and ChIPed DNA (Supplementary Figure 1) was performed. The majority of expressed genes (88%) contained H3K4me3 and H3K9ac enrichment within 5kb of the TSS (Supplementary Table 1), and over 90% of gene histone enrichment was observed at the TSS as expected (6,13). Differentially expressed transcripts (DE) and differentially enriched chromatin regions (DERs) were determined by comparing rats 10 weeks post AOM injection (cancer progression stage) to saline (control).

Consistent with carcinogen exposure, an average of 46 high multiplicity aberrant crypt foci (HM-ACF) were detected in the rats treated with AOM, while there were no detectable HM-ACF in rats injected with saline (Figure 1.A). Since HM-ACF formation precedes carcinoma and shares many of the histopathological characteristics of human CRC (25), it is a useful biomarker to determine the extent of colon cancer progression. A comparison of total H3K4me3 and H3K9ac nuclear levels in AOM and saline treated rat colonic epithelial cells by Western blotting showed similar levels of histone tail modifications between the two treatments (Figure 1.B).

Figure 1. AOM induced pathophysiology and epigenetic changes during cancer progression.

A, High multiplicity aberrant crypt foci (HM-ACF) number, a precursor to tumorigenesis is shown. Data represent 43 rats at 10 weeks post AOM injection. No HM-ACF were observed in the 24 saline-injected animals. ***, p-val < 0.001 (one-way ANOVA). B, Colonic epithelial cell protein levels of trimethylated H3K4 and acetylated H3K9 are not affected by AOM exposure. Representative Western blots of colonic crypt nuclear protein extracts from AOM and saline injected animals are shown. Blue bars represent AOM treated and red bars saline injected colon. Band volumes were quantified using Quantity One software. Values are means ± SE (n = 7). At least two independent assays were conducted. n.s., p-val > 0.05. C, H3K4me3 is highly sensitive to AOM exposure. MAplots indicate the differential expression of all transcribed genes or histone tail enriched regions (y-axis, log-ratio of difference in intensity of histone tail modifications enriched regions) vs their overall intensity of expression (x-axis, log-average of read counts) following AOM vs saline treatment. Pink represents differentially expressed (DE) transcripts and differentially enriched regions (DERs) with a p-value < 0.05, and genes FDR <0.1 in red, all other detected genes are highlighted in blue.

3.1. Global effects of AOM on transcription and histone tail modifications

To identify DE transcripts and DERs with histone tail H3K9ac (K9ac) and H3K4me3 (K4me3) modifications, gene lists from treatment samples were filtered according to multiple criteria as described in the “Methods”. Briefly, DE genes and DER peaks with an FDR<0.1 were compared across AOM and saline injected (control) rats. Figure 1.C shows the distribution of expression strength relative to the log-ratio of DE and DERs. Included above each MAplot are the total number of DE genes (including different isoforms) and the total number of DERs (annotated and un-annotated). Supplementary Table 2 contains a list of the transcripts and DERs with an FDR<0.1 in AOM vs saline treated animals. Enriched regions (peaks) further than 5 kb from the transcription start site were considered “un-annotated” and classified with an ID representing the rn4 genomic location of that island.

The number of DER genes affected by AOM carcinogen treatment was significantly greater in K4me3 (3792), with 3200 of the DERs found at the gene transcription start site (TSS), 41 of which were located 1-1k bp and 71 were located 1k-5k bp from the gene TSS. Additionally, more genes with K4me3 DERs were downregulated (3171) vs upregulated (621) in AOM vs saline. This contrasted with the total number of differentially expressed genes, RNA (116) and K9ac DERs (24) which were less than 5% as abundant as K4me3 DERs (Figure 1.C). These data indicate that H3K4me3 exhibited greater sensitivity to AOM exposure as compared to K9ac DERs and RNA DEs. This pattern of enhanced sensitivity was also observed with respect to fold changes, e.g., K4me3 log2(fold change) DERs ranged from −7.02 to 5.61 as compared to K9ac DERs (−2.96 to 4.22) and RNA DEs (−4.03 to 5.20).

3.2. Lack of correlation between AOM affected genes at the transcription and histone tail modification levels

Based on previous work indicating that histone tail modifications regulate gene expression (5), we expected that K4me3 and K9ac DER genes would correlate with RNA DE genes. Generation of a global plot of all the K4me3 and K9ac DER fold changes against RNA DE genes revealed poor correlation between these histone marks and RNA, regardless of p-values (the axes include number of genes with an FDR<0.1 (Figure 2.A–C). Similarly, a poor correlation was observed between annotated K9ac and K4me3 DER (Figure 2.A–C).

Figure 2. Correlation between transcriptional DE and histone tail DERs.

A, Poor correlation is observed between DE transcripts and histone tail modifications with DERs. Scatterplots reveal low correlation between DE transcripts and histone tail modifications with DERs by comparing the log2(fold changes) from AOM vs saline treatments. Specific contrasts include RNA vs. K4me3, RNA vs. K9ac, and K4me3 vs. K9ac. B, Chromatin signature H3K4me3 occupancy following AOM exposure was associated with downregulation of FOXO3, NR3C1, PDCD4 and VEGFA, without altering H3K9ac occupancy and transcription. C, Venn diagram illustrating the number of genes modulated by AOM in common between the various epigenetic stages (H3K4me3, H3K9ac and RNA). OASL2, RTP4, and TPM2 were affected across all measured epigenetic states. D, Chromatin signature of H3K9ac and H3K4me3 occupancy following AOM exposure is associated with upregulation of OASL2 and RTP4 and downregulation of TMP2. Representative UCSC genome browser images of all DNA reads mapped and grouped by genomic location in highly differentially enriched H3K4me3-regulated genes. H3K4me3 data from AOM treated animals are shown as red peaks, and saline control as blue peaks. H3K9ac data from AOM treated animals are shown as purple peaks, and saline control as brown peaks.

Genes that were altered by AOM at multiple epigenetic stages and/or transcription are listed in Table 1. Only three genes exhibited a change in transcription and K4me3 and K9ac enrichment following AOM exposure (Figure 2.D, Table 1). This included the upregulated novel oncogene OASL2, an interferon-induced antiviral enzyme which plays a critical role in cellular antiviral response by degrading dsDNA in the cytosol (26). RTP4 was also upregulated by AOM exposure. This is noteworthy, because RTP4 is a receptor transporter protein chaperone that escorts GPCRs to the plasma membrane, and has been identified as a breast cancer biomarker correlated with poor patient survival (27). In addition, TPM2, a cytoskeleton-regulating protein necessary for cancer cell survival, was upregulated transcriptionally but downregulated across K4me3 and K9ac DERs.

Table 1.

List of AOM induced differentially expressed (DEs) and differentially enriched regions (DERs) detected across multiple epigenetic states.

		RNA	RNA	K4me3	K4me3	K9ac	K9ac
Symbol	Entrez Gene Name	FDR	log2(FC)	FDR	log2(FC)	FDR	log2(FC)	Location	Type(s)
Anp32a	acidic (leucine-rich) nuclear phosphoprotein 32 family, member A	*	*	2.09E-02	−0.95	5.39E-02	0.50	Nucleus	other
APOB	apolipoprotein B	6.11E-03	1.40	5.91E-02	0.60	*	*	Extracellular Space	transporter
B4GALT1	UDP-Gal:betaGlcNAc beta 1,4-galactosyltransferase, polypeptide 1	1.01E-02	1.86	2.08E-04	−1.63	*	*	Cytoplasm	enzyme
Casp12	caspase 12	4.78E-02	2.07	1.51E-03	2.94	*	*	Cytoplasm	peptidase
CDH11	Cadherin 11, type 2, OB-cadherin (osteoblast)	7.18E-08	3.04	1.16E-03	0.78	*	*	Plasma Membrane	other
Cxcl9	chemokine (C-X-C motif) ligand 9	1.05E-02	1.70	4.09E-02	1.17	*	*	Extracellular Space	cytokine
FIGN	fidgetin	9.74E-03	−1.31	*	*	3.38E-02	−0.74	Nucleus	other
Ifi27	interferon, alpha-inducible protein 27	9.15E-02	1.68	2.62E-05	1.80	*	*	Cytoplasm	other
Ifi47	interferon gamma inducible protein 47	8.06E-02	0.66	3.84E-05	1.14	*	*	Cytoplasm	other
IFIT2	interferon-induced protein with tetratricopeptide repeats 2	2.84E-03	1.58	4.10E-03	0.98	*	*	Cytoplasm	other
IFIT3	interferon-induced protein with tetratricopeptide repeats 3	8.85E-05	3.11	7.20E-02	0.74	*	*	Cytoplasm	other
IRF9	interferon regulatory factor 9	1.26E-05	0.91	6.79E-04	1.06	*	*	Nucleus	transcription reg
ISG15	ISG15 ubiquitin-like modifier	1.17E-03	1.87	9.02E-04	−1.17	*	*	Extracellular Space	other
NEGR1	neuronal growth regulator 1	3.81E-02	−1.36	*	*	3.85E-02	−0.68	Plasma Membrane	other
NFIA	nuclear factor I/A	*	*	4.57E-02	0.58	2.54E-03	0.47	Nucleus	transcription reg
OAS1	2′–5′-oligoadenylate synthetase 1, 40/46kDa	4.29E-04	1.97	3.97E-02	1.02	*	*	Cytoplasm	enzyme
Oasl2	2′–5′ oligoadenylate synthetase-like 2	1.27E-07	2.07	2.39E-02	0.95	4.96E-04	0.78	Other	enzyme
PHLDA3	pleckstrin homology-like domain, family A, member 3	1.27E-03	2.66	9.09E-02	−0.86	*	*	Plasma Membrane	other
PRDM1	PR domain containing 1, with ZNF domain	8.86E-02	0.92	2.01E-02	0.55	*	*	Nucleus	transcription reg
RET	ret proto-oncogene	*	*	1.58E-02	1.16	7.05E-02	−0.93	Plasma Membrane	kinase
RRAGD	Ras-related GTP binding D	7.72E-03	−0.58	5.03E-04	−3.01	*	*	Cytoplasm	enzyme
RTP4	receptor (chemosensory) transporter protein 4	8.33E-03	1.08	4.08E-02	0.63	2.54E-02	0.45	Plasma Membrane	other
SRD5A2	steroid-5-alpha-reductase, alpha polypeptide 2 (3-oxo-5 alpha-steroid delta 4-dehydrogenase alpha 2)	6.58E-02	−1.20	4.84E-02	−0.54	*	*	Cytoplasm	enzyme
ST3GAL5	ST3 beta-galactoside alpha-2,3-sialyltransferase 5	5.44E-02	1.17	4.59E-02	−0.87	*	*	Cytoplasm	enzyme
TMEM50B	transmembrane protein 50B	2.80E-02	0.46	1.62E-02	−0.73	*	*	Plasma Membrane	other
Tpm2	tropomyosin 2, beta	4.68E-02	1.65	1.00E-01	−0.41	3.38E-02	−0.81	Cytoplasm	other
TTC7A	tetratricopeptide repeat domain 7A	*	*	8.56E-03	−0.85	3.65E-02	0.51	Plasma Membrane	other
VAMP7	vesicle-associated membrane protein 7	2.14E-02	−0.42	9.20E-02	1.04	*	*	Cytoplasm	transporter
ZBP1	Z-DNA binding protein 1	1.59E-03	2.02	4.32E-04	1.24	*	*	Cytoplasm	other

^*FDR>0.1

3.3. Identification of upstream regulators perturbed by AOM

To identify the cascade of transcriptional regulators linked to AOM induced DE and DER genes, we performed an IPA Upstream Regulator (URs) analysis. This in silico analysis is based on prior knowledge of expected effects between transcriptional regulators and their target genes stored in the Ingenuity Knowledge Base. Initially, we quantified known targets of transcriptional regulators present in our dataset and compared their direction of change (over-or under-expression) to predict likely relevant regulators including transcription factors, nuclear receptors and enzymes. Supplementary Table 3 provides a summary Ingenuity Pathways Analysis of annotated RNA DEs and K4me3 and K9ac DER genes.

The number of AOM modulated URs in common between K4me3, RNA, and K9ac (Figure 3.A) was greater than the number of DER and DE genes in common between those same epigenetic levels (Figure 2.C). The top ranked AOM modulated URs affected at all levels (K4me3, RNA and K9ac) included the ion channel-ANXA7, insulin receptor-INSR, along with transcriptional regulators-SOX1/3 and TRIM24 (Figures 3.A–D). In addition to p-values, activation z-scores were used to infer likely activation states of URs based on comparison with a model that assigned random regulation directions (Supplementary Table 3 and Figures 3.A–D). Using this approach, the inhibition of ANXA7 activity was linked to 17 K4me3 DERs, two RNA DE genes (upregulated CXCL9 and PHLDA1) and K9ac upregulated NFIA (Figure 3.B and Supplementary Table 3). The activation of INSR was associated with 90 K4me3 DERs, 10 RNA DEs and two K9ac DEs (Supplementary Table 3). Activated SOX1/3 was linked to 19 K4me3 DERs and three RNA DEs including OASL2, a gene affected at all measured epigenetic stages (RNA, K4me3 and K9ac).

Figure 3. Top Upstream Regulators (URs) affected by AOM across multiple epigenetic states.

A, Venn diagram illustrating the total number of URs in common across epigenetic states (H3K4me3, H3K9ac and RNA). The five URs predicted to be affected by AOM are listed. P-values and activity (z-scores) of URs were determined by Ingenuity Pathway Analysis of genes with an FDR<0.1. B–D, Representative networks of genes regulated by top ranked upstream regulators (URs). Yellow fill represents the projected increase in UR activity, green fill represents projected decrease in UR activity. Blue fill indicates decreased gene activity (DE and/or DERs) and orange fill indicates increased gene activity (DE and/or DERs), deeper color hue indicates genes with greater |log2(fold change)|. Solid lines represent direct and dashed lines represent indirect gene interaction. B, K4me3 DERs associated with activation of SOX1/3 development associated transcriptional regulators and inhibition of ANXA7, an ion channel with tumor suppressor function. C, Concurrent increase in mRNA DEs and K4me3 DERs predicted inhibition status of TRIM24 (Tif1α), a tumor suppressor transcriptional inhibitor of proto-oncogenes that activate transcription, immune response and cell proliferation. D, Decrease in K4me3 DERs of key genes (blue) indicating AOM induced activation of KDM5B (histone K4me3 demethylase) associated with inhibition of KMT2D (histone K4 methylase). E, Heatmap of top URs in common between multiple epigenetic stages (H3K4me3, H3K9ac and RNA). Data represent −log(Benjamini-Hochberg-adjusted p-value) with a score >1.29 indicating a B–H p-value<0.05. URs are grouped according to families, color scale ranges from yellow to brown, with decreasing p-value corresponding to deeper shades of brown.

At the transcriptome level, activity inhibition of the master regulator TRIM24 was linked to 23 mRNA DEs (8 of which were also categorized as URs, including OASL2), all of which were detected within the 29 upregulated K4me3 DERs and the one K9ac DER (RTP4) (Figure 3.C and Supplementary Table 3). At the H3K4 trimethylome level, only two of the many established K4me3 methylases and demethylases (URs) were detected as having predicted activity changes in response to AOM induced deregulation of K4me3 DERs. These histone tail modifiers were, activated H3K4 demethylase KDM5B (z-score 2.2) and inhibited H3K4 methylase KMT2D (z-score −5.6) (Figure 3.D). The expression (DE) of KDM5B and KMT2D was not affected by AOM. In addition, H3K4me3 or K9ac ChIPed sequences bound to MLL and JARID1C genomic regions (DER) were similarly unaffected, suggesting that no mutations were induced by AOM treatment (Supplementary Table 2). Their predicted activity and inhibition, respectively, were consistent with the increased number of downregulated versus upregulated K4me3 DERs.

Among the URs detected at multiple epigenetic stages, 34 have been previously associated with adenocarcinoma (Figure 3.E and Supplementary Table 3). The majority of these hits are transcriptional regulators and enzymes, especially kinases. We also cross-correlated top IPA predicted URs against all RNA DE genes to determine whether URs affected by AOM were being modulated at the early transcriptional stage (versus the translational or protein activity stages). Of the top URs predicted to affect the AOM induced transcription of DE genes, 13 were transcriptionally upregulated (APOB, CXCL10, CYBB, IRF7, IRF9, Irgm1, ISG15, SLC29A1, SPRY4, STAT1, TRAF3, USP18, ZBP1), and one was downregulated (EIF4B) (Supplementary Table 3). Interestingly, these genes seem to play a role in both ‘cause’ and ‘effect’ of the transcriptional deregulation. Among K4me3 and K9ac associated URs, only FOXC1 (a K4me3 UR) was differentially expressed (Supplementary Table 3).

3.4. Identification of pathways and networks perturbed by AOM

To understand the biological relevance of the AOM-induced dysregulated genes, a functional analysis using the IPA algorithm was performed to identify canonical pathways (CP). The score computed by IPA for each canonical pathway is derived from a p-value and indicates the likelihood of the genes being found together in that network by random chance (28). Output from this analysis revealed a minimal commonality between the canonical pathways affected at the K4me3 level (262 CPs) compared to RNA (28 CPs) and K9ac (2 CPs) (Supplementary Figure 2.A). As expected, based on the number of AOM affected genes at each epigenetic stage, we observed fewer RNA DE and K9ac DER genes associated with each pathway category compared to K4me3. In the case of K9ac DERs, very few candidate genes/pathways were detected, making the analysis less reliable (Supplementary Table 3).

At the transcriptome level (RNA-seq), Activation of IRF (interferon regulatory factor) by Cytosolic Pattern Recognition Receptors and Interferon Signaling were identified as the most significantly enriched pathways (Figure 4.A). In addition to their contribution to immunity, accumulating evidence indicates that IRFs also have critical functions in the regulation of oncogenesis and metabolism (29). Furthermore, 38 of the 64 molecules that make up the IRF Cytosolic Pattern Recognition Receptors pathway were included in the URs associated with RNA-seq differentially expressed genes (DEs) (Supplementary Figure 2.B). We subsequently compared the top two enriched pathways with disease networks relevant to these data and noted that altered molecules in both pathways, including IRF7, ISG15, OAS1, STAT1 and TRAF3, were also primarily correlated with the top DE enriched disease network, Infectious Disease, Antimicrobial Response, Inflammatory Response (Figure 4.B and Supplementary Table 3.

Figure 4. Top canonical pathways and networks affected by AOM.

A, List of the top three enriched pathways, followed by top activated and inhibited pathways grouped by category. The RNA-seq category includes the only two pathways that were activated/inactivated based on mRNA DE data analysis. All other canonical pathways listed were extrapolated from K4me3 DERs. Activation z-scores indicate whether an upstream transcription factor is significantly more “activated” or “inhibited” based on mRNA DE or K4me3 DER data. P-values are −log(Benjamini-Hochberg-adjusted p-value) with any score >1.29 indicating a B–H p-value<0.05. B–C, Blue fill indicates decreased gene activity (DE and/or DERs) and orange fill indicates increased gene activity (DE and/or DERs), deeper color hue indicates genes with greater fold change. Dashed lines indicate indirect interactions, solid lines indicate direct interactions. The arrow style indicates specific molecular relationships and the directionality of the interaction. B, Top network annotation of transcriptionally DE genes. C, Top network annotation of K4me3 DER genes.

For K4me3, the top most activated and inhibited canonical pathways are shown in Figure 4.A (p<0.05, z-score>|1.4|) and grouped by Signaling Pathway Categories to provide a broader overview of the effect AOM during the promotional (pre-adenomatous polyp) stage of CRC. The top canonical pathway (Molecular Mechanisms of Cancer) mapped the greatest number (144) of genes with K4me3 DERs. Interestingly, none of these genes were differentially transcribed (DE) (Supplementary Table 3). As expected, based on previous findings, RAR Activation and Protein Kinase A Signaling were also enriched (30,31). The top three pathways in Figure 4.A lack an activation score because they include findings without associated directional attributes necessary for activity prediction by IPA. The disease network most associated with K4me3 DERs was Cancer, Inflammatory Response, centering around the glucocorticoid receptor (NR3C1) and included genes that are also part of the top affected canonical pathway (Figure 4.C and Supplementary Table 3).

Examination of K4me3 DERs revealed that AOM downregulated multiple apoptotic pathways (Figure 4.A). However, a similar pattern was not observed at the transcriptome level. Interestingly, AOM also inhibited Cellular Stress associated pathways including the p38 MAPK signaling genes, MAPK12–14 and MAP2K4 (32) (Figure 4.A and Supplementary Table 3). Many second messenger signaling pathways were also inhibited by AOM induced cancer progression, including RhoA Signaling (Figure 4.A). Activated RhoGDI Signaling (Rho GDP-dissociation inhibitors) was the exception to this second messenger inhibitory pattern. Since RhoGDI negatively regulates Rho-family GTPases, the activation of this pathway corroborates the observed inhibition of RhoA signaling (Figure 4.A). Additionally, among the top activated diseases we found 1210 genes specifically associated with increased colorectal carcinoma, z-score=2.43 and p-value<0.0001 (Supplementary Table 3). The majority of these genes were enzymes (427 genes), including upregulated G3BP1, and downregulated B3GNT6, MSH2, SOD1, MMP9, FGFR2, TGFBR2, and FZR1; transcription regulators (199 genes) including downregulated H2AFX, NOTCH2, and SMAD3; and transporters (88 genes) including downregulated ABCB4, ATP2A2, and BAX (Supplementary Table 3).

3.5. Cataloging annotated middle and long ncRNAs and un-annotated H3K4me3 and H3K9ac enriched regions

For the purpose of assessing middle and long ncRNAs, we isolated annotated and un-annotated ncRNA reads. Based on technical constraints associated with standard RNA-seq, in which transcripts <200 bp are eliminated prior to sequencing, the ncRNA genes reported in this section are confined to middle and long ncRNAs (Supplementary Table 2). A total of 20 annotated middle and long ncRNAs were transcribed in the colonic epithelium, none of which were found to be differentially expressed. Out of the 46 annotated ncRNAs with K4me3 enriched regions, 19 contained DERs, all of which were lncRNAs except for Terc (Supplementary Table 4). Only one of the 50 K9ac annotated ncRNAs were differentially enriched (Supplementary Table 4). It is noteworthy that many of the ncRNA DERs detected have been previously reported to play a role in carcinogenesis (33). Additionally, there was no overlap between K4me3 and K9ac DERs of annotated ncRNAs induced by AOM, suggesting the same poor correlation between K9ac and K4me3 DERs and non coding RNA DEs as observed with protein coding genes.

In an effort to extract further knowledge from the genome-wide data we also examined un-annotated genomic regions for AOM induced epigenetic changes and patterns. An assessment of K4me3 and K9ac enriched regions >5000 bp away from an annotated gene (un-annotated enriched region) exposed 2514 DERs out of 18477 K4me3enriched un-annotated regions and 20 K9ac DERs out of 25,426 un-annotated enriched regions (Supplementary Table 2). Interestingly, there were similar numbers of upregulated (1392) and downregulated (1127) un-annotated K4me3 DERs, six of these DERs co-localized between K9ac and K4me3, five of which were upregulated by AOM. Thus, the originally described pattern of AOM induced K4me3 downregulation (Figure 1) was only observed in DERs corresponding to annotated genes. We also searched for putative middle and long noncoding RNA, de-novo generated transcripts >150bp long that did not align to any annotated rat gene. These middle and long ncRNAs were compared for homology against annotated mouse and human genes and identified as “noncoding” by CPC to eliminate any potential un-annotated protein coding genes. A total of 242 genomic regions were identified as putative middle and long ncRNA transcripts (Supplementary Table 5). However, none were differentially expressed following AOM treatment, and many of these transcripts were accompanied by K4me3 and K9ac occupancy (Figure 5). The majority of middle and long ncRNAs identified were categorized as small nucleolar RNA (snoRNA) by the RFAM covariance model (Supplementary Table 5 and Figure 5), implying that the regulation of rRNA splicing and translation is linked to un-annotated functional RNAs in the colon.

Figure 5. H3K4me3 and H3K9ac enriched regions co-localize with ncRNAs.

A, Putative ncRNA loci with K4me3 and K9ac enriched regions. (B–C) UCSC genome browser snapshots of ncRNA, as well as their conserved, thermodynamically stable secondary structures predicted by RNAfold (below); putative ncRNAs also include predicted annotation. The entropy color scale represents the values at the weakest spots of the structure where 0 entropy means no deviations, and entropy >0 indicates some deviations; the higher the entropy, the more likely for the folding structure to deviate. B, Genomic occupancy and structure of annotated and putative ncRNAs with unregulated K4me3 DERs (black box) and putative ncRNA homologous to a mouse annotated lncRNA.

4. Discussion

To our knowledge, this is the first in vivo study to globally assess the chromatin state and transcriptome profile of colonic crypt epithelial cells at a critical early stage of cancer progression. Our analyses were directed towards addressing three major biologically relevant questions: (1) are carcinogen-induced transcriptional profiles (DE) similar to DERs with respect to histone tail modifications associated with gene activity?; (2) which molecules play major epigenetic regulatory roles in DE and DERs?; and (3), what are the cellular functions of the DE genes and DERs?

4.1. AOM selectively modulates H3K4me3 genome association

Although carcinogen effects were identified at the RNA (116 DE genes), K9ac (49 DERs, including 24 genes) and K4me3 (7678 DERs, including 3792 genes) levels, the most notable overall changes were detected in the 3171 downregulated K4me3 genes. These novel findings indicate that AOM suppresses the enrichment of H3K4me3 in a very large number of genes at an early stage of colon cancer progression (Figure 1). We also noted that DE genes were poorly correlated with K4me3 (<28%) and K9ac (<24%) DERs (Figure 2). In order to identify proteins that may affect the expression of genes detected in our experimental dataset, IPA Upstream Regulator (UR) Analysis was employed. We noted that carcinogen-induced changes in gene transcription and H3K4 trimethylation were more correlated at the UR pathway level (76 URs vs 37 DE/DERs in common) (Figures 2.C & 3.A). This is consistent with previous findings indicating that translational alterations are more extensive relative to transcriptional alterations in mediating AOM induced malignant transformation in the colon (12). Collectively, these data suggest that the lack of correlation between DEs and DERs may be attributed to distinct cofactors already positioned at corresponding transcription start sites.

We were also interested in identifying which of the many H3K4 methylases and demethylases (such as KDM2A, KDM7 and KMT3C-E) that regulate gene expression could be contributing to the bias towards downregulation of K4me3 DERs in our data. Interestingly, specific key downregulated DERs were associated with H3K4me3 demethylase KDM5B and methylase KM2TD (Figure 3.D and Supplementary Table 3) both of which have been implicated in cancer development and proliferation (34,35). These findings suggest that increased KDM5B and decreased KM2TD activity are responsible for the AOM induced overall downregulation of K4me3 DERs.

4.2. AOM induced transcriptional changes are strongly correlated with antimicrobial responses

RNA-Seq results at the DE and pathway analysis levels revealed a surprisingly strong link to interferon associated innate immune responses (Figures 3.C & E and Figures 4.A–B and Supplementary Table 3). This included 58 interferon associated DE genes, the URs IFNG, IFNAR1, IRF-3,5,7, TRIM24, the top pathways Interferon Signaling and Activation of IRF by Cytosolic Receptors, and the Infectious Disease, Antimicrobial Response, Inflammatory Response network. Furthermore, recent evidence analyzing the effect of TRIM24 inhibition at various stages (week 5, week 14 and tumor) of hepatocarcinogenesis indicated the presence of increased transcription of STAT1 along with many immune response associated DE genes (36). It is noteworthy, that many of these genes were also upregulated in our study (BST2, CXCL9, IFIT2, IFIT3, IFIT47, IGTP, IRF7, IRF9, IRGM, ISG15, MYH10, OASL2, RTP4 and USP18) and were uniquely expressed at the earlier stages of carcinogenesis (week 5) (36). Taken together, our results are consistent with recent findings indicating that cancer related changes in gut microbial composition may induce interferon associated responses that contribute to the early progression of colon cancer in animal models as well as humans (26,37–40). Additionally, many key genes associated with microbial immune response were upregulated at the RNA plus K4me3 levels, hence, they may act as key modifiers of cancer progression (26,38,41).

One of the most prevalent URs detected from the analysis of both DE and DER genes was TRIM24. We propose a putative mechanism for future investigation (Supplementary Figure 2.C) in which decreased RXR/RAR activity (in a context specific manner) during colon cancer (42) leads to a decrease in KDM5B binding and therefore increased H3K4 trimethylation (43). Since TRIM24 cannot bind to trimethylated H3K4 (44), but can bind to KDM5B (45), an increase in K4me3 and decrease in KDM5B subsequently leads to a decrease in TRIM24 binding (36), thus inducing colon cancer (41).

4.3. Carcinogen induced changes in K4me3 DERs may predict for future transcriptional events

Despite the fact that extensive links between K4me3 DERs and colon tumorigenesis were observed (Supplementary Table 3), these changes were not reflected at the RNA gene expression level during early cancer progression (10 weeks post AOM exposure). For example, among the K4me3 DERs, 1210 have been linked to colorectal carcinoma (Supplementary Table 3). Biomarkers of colon tumors with K4me3 DERs corroborated by previous findings included: FOXO3, which regulates tumor suppressor genes associated with apoptosis (4), the G Protein GNAI2, NR3C1 (glucocorticoid receptor), PDCD4 (programmed cell death 4), VEGFA (an endothelial growth factor), and the DNA repair associated proteins H2AFX and MSH2 (9,46–48). Furthermore, we discovered a range of lncRNAs with K4me3 DERs (Supplementary Table 4), many of which have been previously associated with cancer, including colon cancer associated ATP2A2, CREBZF, PRRT2, H19 (49). Based on these findings, we propose that K4me3 DERs are harbingers of future transcriptional inducers of carcinogenesis.

Following AOM exposure, the observed decrease in Cellular Stress associated pathways and genes, including p38 MAPK Signaling associated genes (Figure 4.A) is consistent with previous findings indicating that MAPK activity may be down-regulated in colorectal cancer (32). Further supporting evidence of this finding includes the association of K4me3 DERs with the downregulation of multiple apoptotic pathways, a common phenotype during tumorigenesis (3,50). Reports of decreased apoptotic pathway activity (and transcriptional decrease of apoptosis associated genes) in tumors but not in earlier stages of colon cancer (11,51) are consistent with our data where a decrease in apoptosis pathways was not detected transcriptionally at the HM-ACF stage, but was detected in the K4me3 DERs, a marker of potential future changes in gene activity.

In summary, we have documented for the first time the chromatin structure associated with gene expression profiles in an in vivo murine colonic tumorigenesis model. Our high-throughput sequencing approach revealed many expected changes at various regulatory stages of gene expression, plus unexpected insight into gene regulation during colon cancer progression. Specifically, we were able to show that AOM induced transcriptional deregulation was primarily associated with interferon-associated immune response genes, while K4me3 deregulation was linked to genes associated with colon tumorigenesis, perhaps acting as a harbinger of changes in gene activity. These findings emphasize the value of genome-wide analyses and may have important clinical relevance for future therapeutic targeting of histone demethylases (35) and microbial composition alteration (37).

Highlights.

• Carcinogen induced K4me3 deregulation is linked to genes associated with CRC.

• AOM induced changes in colonic mRNA are strongly linked to antimicrobial immunity.

• Chromatin state and mRNA profiles are poorly correlated during CRC progression.

• H3K4me3 and H3K9ac enriched regions co-localize with non-coding RNAs.