Transcriptional Attenuation in Colon Carcinoma Cells in Response to Butyrate

The short-chain fatty acid butyrate is produced in the colon by fermentation of dietary fiber, which generates levels of ∼20 mmol/L in the colonic lumen, well above the 2 to 5 mmol/L concentration that induces cell cycle arrest and differentiation in culture (1–3). Colonic epithelial cells in vivo use these high levels as their principal energy source (4), and efficient metabolism of short-chain fatty acids is necessary for their induction of colon cell maturation (5, 6). Gene expression profiling showed that, in culture, butyrate altered the steady-state levels of ∼7% of the genes interrogated, either increasing or decreasing expression monotonically with time (7). This implied an orderly cascade of events, with successive stages in cell maturation driven by sequential recruitment of genes and pathways, a hallmark of an integrated response necessary for intestinal tissue homeostasis.

c-myc and cyclin D1 are genes downregulated by butyrate during colonic cell maturation. Cyclin D1 is a driving component of the cell division cycle, and c-myc stimulates cell cycling and is important in differentiation as cells migrate from the crypt toward the intestinal lumen (8). Moreover, both genes are direct targets of Wnt signaling (9, 10), a fundamental pathway deregulated in colon tumors, most commonly by mutation of Apc, which alters the transcription mediated by β-catenin-T-cell factor targeting the cyclin D1, c-myc, and other genes (11). Targeted inactivation of either c-myc or cyclin D1 reduces Apc-initiated intestinal tumor formation, showing that both are essential for tumorigenesis (12, 13). Thus, butyrate-mediated decreased expression of cyclin D1 and c-myc may be fundamental to its chemopreventive activity (14, 15).

Transcriptional attenuation or pausing, first recognized for c-fos, c-myb, adenosine deaminase, tubulin, and cyclooxygenase-2 (16–20), is now understood to be fundamental for the regulation of many inducible as well as constitutively expressed genes (21, 22). In colorectal carcinoma cells, a transcriptional block in response to butyrate was shown by Heruth and coworkers (14) for c-myc, which was further explored by us (23). Cyclin D1 transcriptional downregulation, however, was reported to be associated with transcription factors acting at the promoter in response to histone deacetylase (HDAC) inhibitors and other agents (15, 24). We used imaging of growing transcripts at their site of synthesis to show that cyclin D1 was also transcriptionally paused by butyrate downstream of the transcription start site (TSS; ref. 25). Thus, transcriptional pausing of c-myc and cyclin D1 by butyrate can play a major role in determining steady-state mRNA levels (23, 25).

Transcriptional regulation is associated with chromatin landmarks, such as specific histone modifications (reviewed in ref. 26). In this regard, butyrate is an inhibitor of HDAC activity, which leads to histone hyperacetylation and chromatin remodeling of induced target genes, whereas histone hypoacetylation has been associated with transcriptionally silent chromatin (27, 28). Here, we first determined the position near the 5′ end at which transcriptional pausing of c-myc and cyclin D1 is initiated by butyrate, as well as chromatin modification along the length of the gene-coding regions. We then extended the investigation to the transcriptome level using novel microarrays, identifying a subset of sequences for which butyrate induced transcriptional pausing. This regulation by butyrate was highly complex and gene specific, and the data suggest that the attenuation mechanism is independent of mechanisms that determine the frequency of transcriptional initiation.

Human colorectal adenocarcinoma SW837 cells (CCL-235) and Caco-2 cells (HTB-37) were grown as described (23, 29) and monitored for Mycoplasma contamination (Mycoplasma Plus PCR Primer Set, Agilent Technologies). Exponentially growing cells at ∼70% confluency were treated or not with 5 mmol/L (SW837 cells) or 2 mmol/L (Caco-2 cells) sodium butyrate (NaB; Sigma) for different time periods. A lower concentration of NaB was used in Caco-2 cells, as these cells are more sensitive to butyrate induction of cell death.

RNA was isolated from cells treated or not with NaB for different time periods. Double-stranded cDNA for microarrays was synthesized from RNA as described (NimbleGen) using SuperScript cDNA synthesis kit (Invitrogen). Details are given in Supplementary Methods.

cDNA gene expression arrays were designed and manufactured in collaboration with NimbleGen. Two oligonucleotide arrays were used: “5′ 3′ array” and “tiling array.” Details are provided in Supplementary Methods.

Western blot analysis and isolation of acid soluble histones were done as described (7, 30). Details are given in Supplementary Methods.

To study polymerase II (Pol II) occupancy and histone modification at a specific sequence, chromatin immunoprecipitation assay was done as described in detail in Supplementary Methods.

cDNA from butyrate-treated and untreated cells, as well as DNA isolated by chromatin immunoprecipitation, was analyzed by quantitative real-time PCR using SYBR Green (Applied Biosystems) following the manufacturer's protocol, as described in Supplementary Methods.

Each sequence potentially paused by butyrate based on the 5′ 3′ array data was classified according to its gene ontology by DAVID Bioinformatics Resources 2008 (NIAID/NIH), as described in Supplementary Methods.

The mean value of replicates was analyzed using Student's t test. One-way ANOVA and Bonferroni multiple comparison test were used to analyze the data from the tiling array. Results were considered significant when P < 0.05.

We first analyzed the effects of NaB on the transcription of cyclin D1 and c-myc in human colorectal adenocarcinoma SW837 cells using a custom-designed oligonucleotide array that tiled through the length of these sequences (Materials and Methods). These arrays were hybridized to Cy3-labeled (untreated, cycling cells) or Cy5-labeled (butyrate-treated cells) cDNA transcribed from total, nuclear, or cytoplasmic RNA. Cells were treated with 5 mmol/L NaB for 0, 6, or 8 hours, time points that preceded the G₁ arrest and apoptosis induced by NaB at ∼24 hours (not shown). Figure 1 focuses on the expression level across the first 355 nucleotides of each sequence, shown as the average ratio of three independent experiments, and the insets show the expression throughout the entire coding regions. As expected, the ratio of hybridization of probes from untreated cells labeled with either fluorochrome was approximately 1 along each sequence (Fig. 1). For cyclin D1, 6 or 8 hours of butyrate treatment decreased its expression by ∼50%, beginning at nucleotide 92 (Fig. 1A). There was a similar pattern for c-myc, with its expression decreased by 80% to 90% downstream of nucleotide 84 (Fig. 1B). The repressed levels of both genes generally persisted throughout the length of each sequence (Fig. 1, insets). A normalization step was done for each of the 628 and 347 probes for cyclin D1 and c-myc, respectively (see Materials and Methods). ANOVA for the NaB-treated data sets (after normalization) versus untreated cells (0 hours), followed by Bonferroni multiple comparison test, showed a highly significant difference at both 6 and 8 hours (P < 0.0001), but no significant difference between 0-hour Cy3 and 0-hour Cy5. We further investigated the effects of NaB on cyclin D1 and c-myc transcription in another human colorectal adenocarcinoma cell line, Caco-2, and similar results were obtained. For cyclin D1, the decrease in transcript expression was localized at nucleotide 107 (Supplementary Fig. S1A). For c-myc, the decrease was somewhat further downstream, at nucleotide 186 (Supplementary Fig. S1B), but still well within the 5′ untranslated region of the message, thus abrogating the generation of a full-length coding mRNA in Caco-2 cells, as it did in SW837 cells. In both cell lines, there were highly reproducible spikes in the generally consistent pattern of repression along both sequences, which could be related to differential hybridization of some probes on the array, despite precautions taken in designing the arrays [e.g., Fig. 1 (insets) and Fig. 5B]. By Western blot, we confirmed that cyclin D1 and c-myc expression was ultimately reduced by NaB at the protein level between 6 and 24 hours of treatment (Fig. 2A).

In untreated SW837 cells, analysis of Pol II by chromatin immunoprecipitation revealed a higher occupancy toward the 5′ end, relative to the rest of the sequence, in the cyclin D1 gene, whereas more similar levels were observed throughout the length of c-myc (Fig. 3A, black columns). Following butyrate treatment of the cells for 6 hours, Pol II occupancy for cyclin D1 decreased most substantially in exon 1 (nucleotides 1-407), as shown in Fig. 3A (gray columns). Whereas the butyrate-induced difference at the 5′ end of cyclin D1 was not observed downstream, for c-myc the pattern was different, with the butyrate-induced decrease being maintained at approximately the same level in exons 2 and 3 (Fig. 3A).

Based on the well-characterized activity of butyrate as a general inhibitor of HDACs (31), we determined the effects of butyrate on chromatin remodeling. An ∼8-fold increase in the overall level of histone H3 acetylation (AcH3) at 30 minutes following butyrate treatment was observed and persisted for ∼6 hours (Fig. 2B). Despite this overall induction of AcH3, it has been reported that butyrate induces histone hypoacetylation of target genes downregulated in expression (32, 33). For both genes, a similar pattern of distribution for AcH3 as for Pol II was found in uninduced cells (Fig. 3B). However, following butyrate treatment, AcH3 levels were differentially altered along cyclin D1, with modest decreases associated with decreased polymerase loading at the 5′ end, whereas for c-myc, AcH3 remained consistently decreased, similar to the persistent decrease of Pol II (Fig. 3B). The increases in AcH3 in exons 4 and 5 of cyclin D1 after butyrate treatment were not associated with increases in Pol II occupancy (Fig. 3A and B). This may be a technical issue because there is a human sequence (FLJ42258) located downstream of the cyclin D1 gene on chromosome 11q13.2 and transcribed from the complementary strand, whose transcript is induced ∼8-fold by butyrate (not shown). Hence, the elevation in AcH3 may be due to the overlap of this transcript that colocalizes with the 3′ end of cyclin D1.

We investigated additional histone methylation marks important in transcriptional regulation. These are trimethylated histone H3 on lysine 4 (H3me3K4), generally associated with transcription initiation; trimethylated histone H3 on lysine 36 (H3me3K36), generally associated with elongation; and trimethylated histone H3 on lysine 27 (H3me3K27), considered a mark of transcriptional silencing (32, 33). Figure 3C shows that in untreated cells, H3me3K4 was relatively higher at the 5′ end of cyclin D1, consistent with its role in transcription initiation, although the highest level was in exon 2, not in exon 1. Similarly, the highest level of this modification was found in exon 2 of c-myc (Fig. 3C). As shown in Fig. 3D, the pattern of H3me3K36 in uninduced cells was different, in each gene increasing toward the 3′ end, consistent with its association with elongation (33). Whereas there were decreases in these histone modifications along each gene following butyrate treatment (Fig. 3C and D), these were modest and generally further downstream of the marked transcriptional attenuation seen in Fig. 1A and B and, thus, may not account for this butyrate effect. Finally, the level of H3me3K27 was low along each gene (note scale of axis) and was increased by butyrate (Fig. 3E), in accordance with the role of this modification in transcriptional repression, although the data do not suggest a specific role in the attenuation of cyclin D1 and c-myc by butyrate.

Similar patterns and modest histone modification in response to butyrate were seen for Caco-2 cells (Supplementary Fig. S2), with the exception of Pol II distribution, which increased at exon 1 of c-myc, and H3me3K4, whose levels in response to butyrate were somewhat elevated in both sequences (Supplementary Fig. S2).

The above studies and our prior work focused on butyrate downregulation of cyclin D1 and c-myc because of the clear functional roles of these genes in colonic cell maturation and colon tumorigenesis. However, the effects of butyrate on colon carcinoma cells are highly pleiotropic, altering the expression of ∼7% of sequences interrogated in our prior report (7). We therefore investigated the extent to which the regulation of transcriptional elongation contributes to downregulation of gene expression. We first used a 5′ 3′ microarray, with five probes within 100 bp of the 5′ end and 3′ end, for each of the 17,378 canonical coding sequences in GenBank (Materials and Methods). In SW837 cells untreated or treated with 5 mmol/L NaB for 2 to 12 hours, data from two independent experiments identified an overlap of 367 sequences (∼2% of the sequences analyzed), the expression of which was downregulated at the 3′ end following butyrate treatment (Fig. 4A). Approximately 53% of these sequences were upregulated by butyrate at the 5′ end. For many of the remaining sequences, there were differences in the extent of downregulation at the 5′ and 3′ ends, with a large number not altered in expression at all at the 5′ end (∼43%). Each of these sequences was a candidate for transcriptional attenuation downstream of the TSS.

Because butyrate can induce chromatin modification, we investigated whether the 367 sequences downregulated at the 3′ end were clustered in regions in the human genome. Figure 4B shows that there was a linear distribution of these loci as a function of the frequency of representation of genes per chromosome on the 5′ 3′ array. Therefore, there was no overall physical clustering of these sequences. A further classification by chromosomal cytoband showed that only ∼12% were clustered in groups, with a maximum of four sequences per cytoband (not shown). Therefore, as for cyclin D1 and c-myc, butyrate-induced transcriptional attenuation may be targeted, rather than involving broad regions affected by chromatin modification.

Recent data show that transcriptional units, previously thought to have a straight-forward organization that generates a single canonical pre-mRNA transcript, are in fact much more complex. In particular, multiple short transcripts are generated at the 5′ end as well as the 3′ end of most genes (34). Thus, there is likely to be an underlying complexity in the structure and regulation of the individual transcripts identified in Fig. 4A.

We therefore analyzed the 367 sequences downregulated at the 3′ end in response to NaB using another custom-designed microarray with probes that tile through each of these sequences, as was done for cyclin D1 and c-myc (Fig. 1). Focusing on the 6- to 8-hour time period that showed the greatest changes in expression, there were 85 sequences whose 3′ end of the canonical transcript was expressed at a lower level than the 5′ end, based on the 5′3′ array analysis. Data from four independent experiments using the tiling array confirmed that 42 of these sequences were expressed to a significantly lower extent (>1.5-fold) toward the 3′ end than at the 5′ end and, thus, potentially paused by butyrate. Closer examination revealed that these 42 sequences fell into two general classes, which were represented equally, as shown in Table 1 and Fig. 5. Category I sequences were upregulated at the 5′ end by butyrate, but either downregulated or not significantly altered in expression further downstream, as illustrated by the TMOD1 gene (Fig. 5A). Twenty-eight of the 29 sequences in category I showed similar patterns of graded decrease in expression along the coding sequence, although they varied in the relative position at which expression was downregulated by butyrate. TMEM49 sequence (NM_030938) was the exception, showing increased expression by butyrate for the first half of the sequence but an abrupt decrease at the junction of exons 10 and 11 (Supplementary Fig. S3), which could suggest induction of alternative splicing by butyrate. However, by reverse transcription-PCR and quantitative real-time PCR, we have not detected any alternatively spliced form of this gene in SW837 cells in response to butyrate (data not shown). Category II genes (Table 1) were not altered by butyrate in expression at the 5′ end, as is the case for cyclin D1 and c-myc (Fig. 1), but then showed heterogeneous patterns among the sequences, which was highly reproducible across independent experiments. These patterns encompassed decreased expression, such as for TFAP4 (Fig. 5B), or increased expression before being downregulated by butyrate, as delineated in Table 1. These patterns may result from complex transcription, characterized by multiple sites of initiation and termination as well as pausing.

The distribution of these two categories with respect to their gene ontology is shown in Table 2, with their gene IDs given in Supplementary Table SIA. The repressed targets in category I were enriched in “cytoskeleton organization” (∼18%; Table 2A) and “apoptosis” (∼15%), possibly contributing to the induction by butyrate of cell cycle arrest and differentiation in the cells analyzed (2, 3). In contrast, category II sequences were associated with the terms “zinc ion binding,” a characteristic of many transcriptional factors involved also in transcriptional elongation (35, 36); indeed, the term “regulation of transcription” was also linked to this category (Table 2B). However, for category II, the input number of genes analyzed (Materials and Methods) was below the threshold required for assigning a P value. Nevertheless, the data suggest a differential pattern of transcriptional downregulation by butyrate, which may be related to different biological functions.

Butyrate, a natural dietary product and an HDAC inhibitor fundamental to normal colonic cell maturation and with potential activity in chemoprevention and chemotherapy, has extensive effects on gene expression. Butyrate alters gene expression as well as the underlying epigenetic mechanisms of transcriptional regulation (reviewed in ref. 37). We first focused on the effects on two specific genes, cyclin D1 and c-myc, which are fundamental in intestinal cell maturation and tumorigenesis, and then expanded the analysis to the transcriptome level, which revealed a complexity of response to butyrate not previously appreciated.

Transcriptional pausing was previously reported at the exon 1/intron 1 boundary of c-myc in hematopoietic tumor cells carrying translocations involving this gene (38, 39). Here, by using high-resolution custom-designed tiling arrays, we determined that in colon carcinoma cells, after 6 to 8 hours of butyrate treatment, there is almost a complete decrease in c-myc transcript expression and a marked downregulation of cyclin D1 expression (∼50%) due to a decreased steady-state level of the transcripts that begin at ∼100 nucleotides downstream of the TSS (i.e., <1% and ∼2% of cyclin D1 and c-myc coding sequences, respectively; Fig. 1), in agreement with previous reports that transcriptional attenuation of the genes occurs close to the TSS (40). The more complete decrease in the steady-state level of c-myc than of cyclin D1 transcripts may be related to the fundamental role that alterations in c-myc play in colon tumor development (12, 41), as well as the critical role of c-myc in reprogramming normal intestinal cell maturation (8). It should also be noted that deletion of only a single allele of cyclin D1 was sufficient to decrease intestinal tumor formation in Apc^Min/+ mice, suggesting that the effects of this locus may be particularly sensitive to more modest changes in gene expression (13).

Transcriptional attenuation in the c-myc gene has been attributed to paused RNA Pol II and/or premature termination in the promoter-proximal region (21, 39). Whereas in Caco-2 cells, Pol II distribution increased with butyrate treatment at the very 5′ end of c-myc, in accord with the induction of transcriptional attenuation by butyrate in this locus, in SW837 cells butyrate inhibited Pol II occupancy toward the 5′ end of cyclin D1 and throughout c-myc. In SW837 cells, butyrate-induced attenuation could be associated with a decrease in elongating polymerase complexes, as described before for other cell types (21, 42, 43).

Butyrate, a well-characterized inhibitor of HDAC activity, induced a large increase in AcH3, a mark of transcriptional activity, which we determined to occur, for example, on CDKN1A (not shown), confirming published data (44). Butyrate also induces histone hypoacetylation close to the TSS in downregulated genes (45). The butyrate-induced differences in the levels of AcH3 that we detected across the exons of cyclin D1 and c-myc were generally related to Pol II occupancy. However, although they decreased toward the TSS, the AcH3 changes were modest. Similarly, basal levels of H3me3K4 were higher nearer the 5′ end of each gene, whereas H3me3K36 increased toward the 3′ end, consistent with reports that these modifications associate with transcriptional initiation and elongation, respectively, in actively transcribed genes (46, 47). However, butyrate again only modestly decreased the levels of these two modifications in both genes analyzed. On the other hand, there was a moderate increase in the levels of H3me3K27 by butyrate throughout cyclin D1 and c-myc, consistent with its role in transcriptional silencing (33). Thus, although the alterations we detected in these histone marks were consistent with gene silencing by butyrate, direct correspondence with the region of attenuation was not definitive. This is consistent with the proposal that these histone modifications are not rigidly associated with the transcriptional state (26) and that the functional epigenetic landscape is more complex than previously thought (48).

Our previous data have shown that butyrate is highly pleiotropic in its effects on gene expression, with a cascade of changes that increases as a function of time following exposure, a hallmark of sequential recruitment of genes and pathways with cell maturation (7). Therefore, we pursued these widespread effects of butyrate using a whole transcriptome analysis, identifying a subset of 42 genes potentially attenuated by butyrate. Transcription of these genes, as well as of cyclin D1 and c-myc, was decreased by NaB downstream of the TSS, suggesting that the initiation step of the transcriptional process was not the principal target, consistent with previous work showing transcriptional elongation as a rate-limiting step of transcription (49). Moreover, a recent report showed that in developmental systems, many genes that do not produce full-length mRNAs are nonetheless transcriptionally initiated (47).

Our transcriptome analysis revealed that whereas genes responding to butyrate fell into general categories based on the location and nature of altered expression along the coding sequence (Table 1; Fig. 5), the pattern of transcription was distinct and highly reproducible for each gene, reflecting the high complexity of genome transcription recently elucidated (50). High-throughput parallel sequencing of transcripts should be informative in understanding this heterogeneity along these specific loci.

There are two important points regarding the complexity of transcriptional regulation by butyrate. First, butyrate is a short-chain fatty acid, consumed as a nutrient and generated by fermentation of dietary fiber, and the primary energy source for colonic epithelial cells (4). Therefore, the intestine evolved in the presence of this compound, and the complex patterns of gene expression triggered by butyrate may be key to generating the highly coordinated response that establishes and maintains the homeostasis of the colon. Second, butyrate may also exhibit antitumor properties, and the control of expression of proto-oncogenes such as c-myc, as well as of genes that directly promote selective advantage of tumor cells (e.g., cell growth, survival, vascularization), by transcriptional attenuation is broader than originally thought.

In conclusion, transcriptional attenuation of the cyclin D1 and c-myc genes in response to butyrate begins at about 100 nucleotides downstream of the TSS. The role of chromatin remodeling and histone modification in response to butyrate is complex, may be independent of the HDAC inhibitory activity of butyrate, and may follow a gene-specific pattern. It has been proposed that control of transcriptional elongation could be a widespread mechanism of gene regulation in eukaryotes (38). Our results show that this is the case for the effects of the short-chain fatty acid butyrate and identify gene targets regulated by this mechanism.

No potential conflicts of interest were disclosed.

We thank Paolo Norio, John Mariadason, Anna Velcich, and Georgia Corner for help in conducting these experiments.

Grant Support: American Institute for Cancer Research (M.C. Daroqui); National Cancer Institute grants U54CA100926, CA123473, and CA135561 (L.H. Augenlicht); and Cancer Center Support Grant P30CA13330.

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