bZIP Dimers CREB1, ATF2, Zta, ATF3|cJun, and cFos|cJun Prefer to Bind to Some Double-Stranded DNA Sequences Containing 5-Formylcytosine and 5-Carboxylcytosine

Abstract

In mammalian cells, 5-methylcytosine (5mC) occurs in genomic double-stranded DNA (dsDNA) and is enzymatically oxidized to 5-hydroxymethylcytosine (5hmC), then to 5-formylcytosine (5fC), and finally to 5-carboxylcytosine (5caC). These cytosine modifications are enriched in regulatory regions of the genome. The effect of these oxidative products on five bZIP dimers (CREB1, ATF2, Zta, ATF3|cJun, and cFos|cJun) binding to five types of dsDNA was measured using protein binding microarrays. The five dsDNAs contain either cytosine in both DNA strands or cytosine in one strand and either 5mC, 5hmC, 5fC, or 5caC in the second strand. Some sequences containing the CEBP half-site GCAA are bound more strongly by all five bZIP domains when dsDNA contains 5mC, 5hmC, or 5fC. dsDNA containing 5caC in some TRE (AP-1)-like sequences, e.g., TGACTAA, is better bound by Zta, ATF3|cJun, and cFos|cJun.

Graphical Abstract

Transcription factors (TFs) bind sequence-specific dsDNA and regulate gene expression.^1,2 The question of dsDNA binding specificity has become more complex in mammals with the observation that 5mC that occurs in the genome is iteratively oxidized by the ten-eleven-translocation (TET) family of dioxygenases to 5-hydroxymethylcytosine (5hmC),³ 5-formylcytosine (5fC), and 5-carboxylcytosine (5caC).⁴ 5fC and 5caC are replaced with cytosine by thymine DNA glycosylase, completing the cytosine demethylation cycle.^5,6

These cytosine modifications occur in regulatory regions of the genome^7–11 and change through development and pathology.¹² 5mC is particularly enriched at CG dinucleotides but is known to occur outside of CG dinucleotides in stem cells and brain.^13,14 5hmC enrichment occurs in embryonic stem cells (ESCs),³ neural cells,¹⁵ primordial germ cells,¹⁶ fertilized oocytes,¹⁷ and tumor cells.¹⁸ 5fC is present at levels similar to those of 5mC and 5hmC at specific genomic sites, e.g., poised or active enhancers⁷ and some exons.^19,20 5caC is enriched at some major satellite repeats.²¹ In some pathogeneses and cancers, like myeloid malignancies, this demethylation pathway is misregulated.²² Several groups have used various methods to identify proteins that preferentially bind modified dsDNA,^23–29 but the biological importance of this preferential binding is difficult to determine.

5mC can both inhibit and enhance the binding of TFs to specific dsDNA sequences.^23,30–34 Our previous work using PBMs has shown that cytosine methylation outside of CG dinucleotides can dramatically alter sequence-specific DNA binding of Tcf3|Ascl1,³⁵ CREB1,³⁶ and Zta.³⁷ Examination of the effect of the oxidative products of 5mC on sequence-specific TF binding to dsDNA is a new endeavor. The three 5mC oxidative products can change sequence-specific binding _{of TFs.}^28,35,38,39

bZIP proteins are eukaryotic TFs that bind sequence-specifically to dsDNA as homodimers or heterodimers.^40,41 The bZIP domain is a long bipartite α-helix.^42,43 The C-terminus contains an amphipathic α-helix that can homodimerize and/or heterodimerize to form a leucine zipper coiled coil.^40,43,44 The N-terminal basic region lies in the major groove of dsDNA and binds sequence-specifically with each monomer binding one half-site of the often-palindromic motif.⁴³ The dsDNA binding specificity of each bZIP dimer is determined by specific residues in the α-helical basic region⁴⁵ that differ between subclasses of bZIP proteins.^43,46

Five amino acid residues of the bZIP protein directly contact DNA nucleotides.^47,48 These include conserved consecutive alanines that interact with the methyl groups of two thymines at T² and T^{−4 49} in the canonical CRE (T⁻⁴G⁻³A⁻²C⁻¹G¹T²C³A⁴) and the related TRE (T⁻⁴G⁻³A^−2C/_G⁰T²C³A⁴) that differ by one nucleotide at the center of the motif.⁵⁰ We used a numbering strategy explaining the nucleotide positions as described previously.³⁷ Our previous work examined how 5mC and 5hmC affected dsDNA binding of Zta, CREB1, CEBPB, and cFos|cJun to specific sequences.³⁶ Methylation of cytosine at C⁻⁴ increases the level of binding by 10-fold for the CEBPB|ATF4 heterodimer to C⁻⁴GATGC²AA.²³ Zta also binds methylated C⁻⁴.⁴⁹ Methylation of cytosine, mimicking the methyl group of thymine, at C² in GC²AA, the half-site of the canonical CEBP motif, stabilizes binding of many bZIP dimers, including CREB³⁶ and Zta.³⁷ At a structural level, this work highlighted an alanine residue in CREB1, cJun, and cFos, a serine residue in Zta, and a valine in CEBPB explaining the altered DNA binding specificity.³⁷ Mutagenesis experiments that mutated serine 189 in Zta⁵¹ and a conserved valine in CEBPB⁵² highlighted the important roles of these residues in determining the specificity of binding to 5mC and 5hmC. In this study, we examined the effects of 5mC, 5hmC, 5fC, and 5caC in one strand of dsDNA on sequence-specific binding of five bZIP domains (CREB1, ATF2, Zta, ATF3|cJun, and cFos| cJun).

EXPERIMENTAL PROCEDURES

Cloning and Expression of bZIP DNA Binding Domains.

The bZIP DNA binding domains (DBDs) of mouse CREB1, ATF2, cJun, and Epstein-Barr viral protein Zta were obtained from T. Hughes (University of Toronto, Toronto, ON) as a chimeric GST construct. CREB1, ATF2, and cJun were cloned into the pETGEXCT (C-terminal GST) vector.⁵³ Zta was cloned into a modified pDEST15 MAGIC vector (N-terminal GST).⁵³ The DBDs of mouse ATF3 and cFos were obtained as a construct cloned into a pT5 expression plasmid.⁵⁰ Proteins were expressed using the PURExpress in vitro protein synthesis kit (NEB) according to the manufacturer’s protocol²³ in a 25 μL reaction volume containing 180 ng of plasmid. The α-helical dsDNA binding regions of the bZIP domains are aligned. Conserved bZIP domain residues are highlighted in yellow, and the variations in red (Chart 1).

Chart 1.

Design of the 40K Feature Array and PBM Experiments.

The design of the Agilent 40K array has been described previously.^23,54–56 Specifically, we used the “HK” array design available on the NCBI Gene Expression Omnibus platform GPL11260 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GPL11260). HK microarrays with 40330 single-stranded DNA 60-mers were double-stranded to generate five types of dsDNA using either cytosine (C, NEB), 5-methylcytosine (5mC, NEB), 5-hydroxymethylcytosine (5hmC, Zymo Research), 5-formylcytosine (5fC, TriLink Biotechnologies), or 5-carboxylcytosine (5caC, TriLink Biotechnologies).^28,35,38 The double-stranding efficiency was monitored using either 4% Cy3-dCTP (GE Healthcare)⁵² or 4× SYBR-Gold (LifeTechnologies Corp.) DNA intercalating dye. These microarrays were used for protein binding microarray (PBM) experiments in which a GST-tagged bZIP DNA binding domain binds to the DNA on the microarray slide. Protein binding reactions were performed as described previously.^23,36,51,52 Briefly, 180 ng of plasmid containing DNA binding domains was used to express proteins using the PureExpress in vitro transcription translation kit (NEB) in a 25 μL reaction volume following the manufacturer’s instructions. The double-stranded arrays and in vitro-expressed proteins were blocked with 4% milk and BSA before the binding reaction. The protein binding reactions were carried out in a hydration chamber for 1 h followed by incubation with the Alexa Fluor 647-conjugated anti-GST antibody for 1 h. Finally, array slides were washed, dried, and scanned using an Agilent Sure Scan II microarray scanner to measure the fluorescence intensities of bound proteins.

Image Quantification and Calculation of 8-mer Z Scores.

For each PBM, scanned image quantification and Z score calculation were performed as described previously.³⁸ Microarray images were analyzed using ImaGene (BioDiscovery Inc.), and the extracted data (probe fluorescence intensity values) were used for further analysis. The probe median intensities were used to calculate the Z score for all 65536 8-mers for all other PBM experiments except dsDNA with cytosine in both strands, as complementary 8-mers are different due to the asymmetric nature of the double-stranding protocol for all four cytosine modifications. For these 8-mers, we compute the Z scores of the reverse complement of the 8-mer extracted from the array probe design. Thus, all 8-mers shown and analyzed for these experiments are taken from the strand that can contain 5mC, 5hmC, 5fC, and 5caC. We also note that the PBM design we used contains features in which all 8-mers occur 32 times on the array when double-stranded. For Z score calculations in which complements are not identical, certain 8-mers are rare. For these experiments, we show only Z scores for 8-mers that appear on at least 10 features, eliminating 1148 8-mers. All of the TFs were assayed with at least two replicates, with good agreement (R > 0.8) (Figures S1 and S7). Data (raw probe intensities and 8-mer Z scores) are available at the NCBI GEO database under accession number GSE141050.

Analysis of 8-mers in ChIP-seq Data Sets.

Human ChIP-seq peaks from the ENCODE Web site^57,58 (Table S6) were downloaded, and Zta ChIP-seq peak coordinates in Akata cells were obtained.⁵⁹ For the analysis of bZIP heterodimers, we generated a set of overlapping ATF3|cJun and cFos|cJun ChIP-seq peaks. To determine the enrichment of each 8-mer in a given ChIP-seq data set, we calculated an “enrichment score”, $E=\frac{{\text{OCC}}_{\text{obs}}}{{\text{OCC}}_{\text{exp }}}$, for each 8-mer across the human genome (UCSC build hg19). OCC_obs is the number of observed occurrences in each set of peaks, and OCC_exp is the number of expected occurrences of the motif. OCC_exp was calculated with the equation ${\text{OCC}}_{\exp }=N\frac{L_{\mathrm{r}}}{L_{\mathrm{g}}}$, where N is the total number of motifs in the whole genome, L_r is the total length (in base pairs) in the set of peaks, and L_g is the total length (in base pairs) of the human genome.

RESULTS

Protein Binding Microarray with Five Types of dsDNA.

We exploited the property of T7 DNA polymerase to incorporate biologically occurring cytosine modifications into dsDNA.^60,61 Cytosine and the four modified cytosines (5mC, 5hmC, 5fC, and 5caC) were used to generate five types of dsDNA on the PBM. The five dsDNAs contain either cytosine in both DNA strands [DNA(C|C)] or cytosine in one strand and either 5mC, 5hmC, 5fC, or 5caC in the second strand [DNA(5mC|C), DNA(5hmC|C), DNA(5fC|C), and DNA(5caC|C)].^36,37,56

The progress of elongation on the glass slide of T7 DNA polymerase producing a new ssDNA templated by the ssDNA on the glass slide to form dsDNA was monitored in two ways. One is to incorporate Cy3-dCTP into the newly synthesized ssDNA strand.²³ The second method is to use SYBR-Gold, which intercalates into dsDNA.^62,63 The histogram of fluorescence intensities at 570 nm for Cy3-dCTP incorporation after the five double-stranding DNA reactions is plotted in Figure 1A, and the average intensities are shown in Figure 1C. Cytosine, 5hmC, and 5fC have similar amounts of Cy3-dCTP fluorescence, with that of 5mC being 2-fold higher. 5caC showed an almost 10-fold higher fluorescence. We divided the fluorescence intensities of each feature by the number of cytosines in the newly synthesized 35-mer variable sequence that incorporates Cy3-dCTP. This analysis indicates that probes with more cytosines have more Cy3-dCTP signal (orange lines in Figure 1A).

Figure 1.

DNA double-stranding efficiency on the PBMs with cytosine, 5mC, 5hmC, 5fC, and 5caC measured by (A and C) Cy3-dCTP incorporation or (B and D) SYBR-Gold binding. In panels A and B, 40300 features are ordered by fluorescence intensity (blue). The orange histogram in panel A shows the fluorescence normalized by cytosine number in the newly made DNA strand (number of guanines in the ssDNA sequence on the microarray). In panels C and D are comparisons of the average fluorescence intensities of 40330 features of ssDNA and five dsDNA with the standard error between two replicates in panel D.

Next, we added SYBR-Gold, which intercalates into dsDNA.⁶³ Histograms of fluorescence intensities for ssDNA and five types of dsDNA were measured at 550 nm (Figure 1B,D). All of the dsDNAs have 4–6 times more fluorescence than ssDNA, suggesting that the modified cytosines were incorporated into dsDNA (Figure 1D). Comparing the Cy3-dCTP and SYBR-Gold data suggests that the T7 DNA polymerase prefers Cy3-dCTP compared to 5caC. Though T7 polymerase did not prefer 5caC, it produced similar amounts of dsDNA as cytosine and the other modified cytosines.

Binding Data Analysis.

dsDNA microarrays were used for protein binding experiments using in vitro-synthesized GST-tagged bZIP domains. Binding is detected by the fluorescence intensity of a Cy5-labeled anti-GST antibody. These fluorescence data were analyzed by three methods. First, we compared the fluorescence intensities of the 40330 features on the microarray for the five types of dsDNA. Each feature on the array contains a different dsDNA 35-mer. Essentially, we are examining sequence-specific dsDNA binding in a nonspecific dsDNA background. Second, we computed a standardized score (Z score) for all possible 8 bp DNA sequences (8-mers) to examine sequence-specific dsDNA binding.⁵⁵ We focused on features and Z scores that contained either the CEBPB half-site GC²AA or the CRE half-site TGAC^‑1 to examine the C² and C⁻¹ positions in an 8 bp motif. Third, single-nucleotide variant (SNV) tables were generated for several 8-mers, including a non-cytosine-containing 8-mer, to examine the contribution of modified cytosines throughout the motif to sequence-specific dsDNA binding.

CREB1 Binding to Five Types of dsDNA.

The binding of CREB1 to three types of dsDNA [DNA(C|C), DNA(5mC|C), and DNA(5hmC|C)] has been examined.³⁶ Binding to the five types of dsDNA was carried out in parallel and yielded results that matched previously published results (Figure S2). We compared binding to normal dsDNA [DNA(C|C)] and each of the four modified dsDNAs to (1) each feature on the PBM (Figure S3A–D), (2) Z scores for all 8-mers (Figure S3E–H), and (3) Z scores for 8-mers with one cytosine to determine the contribution of a single cytosine modification to dsDNA binding (Figure 2A–D). We also examined how each step in the methylation cycle changed binding (Figure S3I–K and Figure 2E–G). Strong binding to some features is observed with all five types of dsDNA. Average and median binding values are presented (Table S1). The greater the difference between average and median binding, the more sequence-specific dsDNA binding is occurring. For example, the difference is greater for CREB1 than for ATF2, suggesting more sequence-specific binding for CREB1.

Figure 2.

CREB1 binding to five types of dsDNA. Z scores for 8-mers with one cytosine for CREB1 binding: (A) dsDNA containing cytosine in both strands DNA(C|C) (x-axis) and DNA(5mC|C) (y-axis), (B) DNA(C|C) (x-axis) and DNA(5hmC|C) (y-axis), (C) DNA(C|C) (x-axis) and DNA(5fC|C) (y-axis), (D) DNA(C|C) (x-axis) and DNA(5caC|C) (y-axis), (E) DNA(5mC|C) (x-axis) and DNA(5hmC|C) (y-axis), (F) DNA(5hmC|C) (x-axis) and DNA(5fC| C) (y-axis), and (G) DNA(C5f|C) (x-axis) and DNA(5caC|C) (y-axis). 8-mers are divided into three groups for panels A–D [8-mers containing the C/EBP half-site GC²AA (green), the CRE half-site TGAC (red), and the rest of the 8-mers (black)] and in five groups for panels E–H [8-mers containing the C/EBP half-site GC²AA (green), the CRE half-site TGAC (red), C⁻⁴GAT (blue), the rest of 8-mers with cytosine (black), and 8-mers with no cytosines (gray)].

As observed previously, 5mC and 5hmC strengthen binding to 8-mers with GC²AA and weaken binding to 8-mers with TGAC⁻¹³⁶ (Figure 2A,B and Figure S3A,B,E,F). With 5fC, the same trend is observed (Figure 2C). Several 8-mers with multiple cytosines and GC²AA like CCACGC²AA are strongly bound with 5fC (Figure S3C,G). The best bound 8-mer with 5mC, 5hmC or 5fC is TGATGC²AA. CREB1 binds dsDNA with 5caC in one strand similarly or slightly more weakly to dsDNA with cytosine in both strands (Figure 2D and Figure S3D,H). 8-mers containing TGAC⁻¹ and 5caC are better bound compared to 8-mers containing 5mC, 5hmC, and 5fC (Figure 2E–G and Figure S3I–K) but still weaker than cytosine (Figure 2D and Figure S3H).

We next generated single-nucleotide variant (SNV) tables for binding of CREB1 to dsDNA and dsDNA containing 5mC, 5hmC, 5fC, and 5caC in one strand (Table 1). The 8-mers examined include the canonical CRE motif TGAC⁻¹GTC³A containing two cytosines. We also examined 8-mers containing one cytosine to determine the contribution of modified cytosine at different parts of the motif to CREB1 binding strength. We focused on cytosine and its modifications at positions C⁻¹, C², and C³. These 8-mers are TGATGTC³A, TTAC⁻¹GTAA, TGAC⁻¹GTTA, and TGATGC²AA. We also examined an 8-mer with no cytosines, replacing both cytosines in the canonical CRE with thymine (TGATGTTA). These SNV table results reinforce what we learned from examining Z scores for 8-mers. 5mC, 5hmC, and 5fC facilitate binding at C² and inhibit binding at C⁻¹. 5caC slightly favors C⁻¹ but not C².

Table 1.

SNV Tables of Z Scores for the Binding of CREB1 to (A) the Canonical CRE Motif, TGAC⁻¹GTC³A, (B) Variants of CRE Motif TGATGTC³A, (C) PAR Consensus Motif TTAC⁻¹GTAA, (D) Variants of PAR Consensus Motif TGAC⁻¹GTTA, (E) Best Bound Motif TGATGC²AA, and (F) Variants 8-mers TGATGTT³A^a

^aShaded cells highlight specific substitutions mentioned in the text.

ATF2 Binding to Five Types of dsDNA.

The ATF2 binding fluorescence is approximately half of that observed for CREB1 (Figure S4A–D and Table 1). Figure 3A–D shows ATF2 Z scores for 8-mers containing one cytosine. The best bound 8-mer is the CRE like ATGAC⁻¹GTA. As observed with CREB1, 5mC, 5hmC, and 5fC strengthen the binding to 8-mers containing GC²AA and weaken the binding to 8-mers containing TGAC⁻¹. Binding to 8-mers containing 5caC is similar to that with cytosine. When we examined the change in binding with oxidation, the most dramatic changes are with methylation of cytosine and the change from 5fC to 5caC (Figure 3A,G and Figure S4E,K). 5fC containing 8-mers like TACCGC²AA with multiple cytosines and GC²AA is strongly bound (Figure S4E,G). SNV tables for binding of ATF2 to these five dsDNAs are similar to that of CREB1 (Table S2).

Figure 3.

ATF2 binding to five types of dsDNA. Z scores for 8-mers with one cytosine for ATF2 binding to dsDNA as described in Figure 2.

Zta Binding to Five Types of dsDNA.

The binding of Zta to three types of dsDNA [DNA(C|C), DNA(5mC|C), and DNA(5hmC|C)] has been described.³⁷ We extended those studies by examining 5fC and 5caC in one strand of dsDNA (Figure S5A–H and Figure 4A–D). The preference for 5mC, 5hmC, and 5fC at C² is stronger than observed for CREB1 and ATF2. 5caC inhibits binding to 8-mers containing GC²AA. Only a few 8-mers with CRE half-site TGAC⁻¹ are bound strongly in the presence of 5caC. Upon examination of the change in Z score with each step in the methylation cycle, 5fC weakens binding to 8-mers containing GC²AA compared to 5hmC (Figure 4E–G). Examination of SNV tables confirms that 5mC, 5hmC, and 5fC are preferred in C² but not in C⁻¹ (Table S3).

Figure 4.

Zta binding to five types of dsDNA. Z scores for 8-mers with one cytosine for Zta binding to dsDNA as described in Figure 2.

ATF3|cJun Heterodimer Binding to Five Types of dsDNA.

ATF3|cJun binding is not dramatically changed with the cytosine modifications with a few exceptions, particularly for 8-mers with multiple cytosines (Figure 5A–D and Figure S6A–H). Preferential binding to TRE and not the CRE is observed. Several 8-mers with multiple cytosines and TGAC⁻¹ like CATGAC⁻¹GT are strongly bound with 5caC unlike CREB1, ATF2, and Zta (Figure S6H). Next, we examined SNVs of the six 8-mers examined with CREB1 (Table S4). Though Z scores are low, the trends observed at C⁻¹ and C² for CREB1, ATF2, and Zta are observed. 5mC, 5hmC, and 5fC is favored at C², and 5caC is not. 5caC is favored at C⁻¹.

Figure 5.

ATF3|cJun binding to five types of dsDNA. Z scores for 8-mers with one cytosine for ATF3|cJun binding to dsDNA as described in Figure 2.

cFos|cJun Binding to Five Types of DNA.

cFos|cJun also preferentially binds the TRE (Figure 6A–D and Figure S7A–H). When Z scores are compared as cytosine becomes methylated and subsequently oxidized (Figure 6A,E–G and Figure S7E,I–K), dsDNA with 5mC showed a moderate binding preference with 8-mers containing GC²AA half-sites with single or multiple cytosines, but only 5fC has a binding preference with 8-mers containing multiple cytosines and GC²AA. SNVs of the six 8-mers examined with CREB1 (Table S5) reinforce our previous observations, similar to ATF3|cJun with strengthened binding for 8-mers containing 5mC, 5hmC, and 5fC at the C² position, but weakened binding at C⁻¹. 5caC at C⁻¹ is favored, but not at C².

Figure 6.

cFos|cJun binding to five types of dsDNA. Z scores for 8-mers with one cytosine for cFos|cJun binding to dsDNA as described in Figure 2.

Comparison of bZIP TFs Binding to Five Types of dsDNA.

We summarized the effects of each of the five types of dsDNAs on binding for each of the bZIP proteins examined. Correlation plots of the five bZIP domains binding the five types of dsDNA reveal both similarities and differences between bZIP proteins (Figure 7). Binding to 8-mers containing 5mC, 5hmC, and 5fC is similar for CREB1, Zta, and ATF2 (Figure 7A–C). Correlations between 8-mers containing these modifications are moderate to high (R = 0.48–0.7) and approach correlations obtained for replicate experiments [R = 0.88–1 (Figure 7 and ref 37)]. 8-mers with cytosine or 5caC are bound more similarly by CREB1, Zta, and ATF2 (Figure 7A–C) but less so by ATF3|cJun and cFos|cJun (Figure 7D,E). When we examine binding by dsDNA type, new differences are revealed. Zta does not form groups with other bZIPs for all five dsDNAs (Figure 8), highlighting its distinct properties for binding to DNAs containing modified cytosines as shown previously.³⁷

Figure 7.

Summary of correlations of each bZIP protein binding five types of dsDNA. Summary of Pearson correlations computed using the top 500 8-mer Z scores for (A) CREB1, (B) ATF2, (C) Zta, (D) ATF3|cJun, and (E) cFos|cJun.

Figure 8.

Correlations between TF binding to five types of dsDNA. Summary of Pearson correlations computed for the top 500 8-mers Z scores for CREB1, ATF2, Zta, ATF3|cJun, and cFos|cJun binding to dsDNA (A) (DNA(C|C), (B) DNA(5mC|C), (C) DNA(5hmC|C), (D) DNA(5fC|C), and (E) DNA(5caC|C).

We next generated sequence logos of well-bound 8-mers and examined comparisons of the similarities and differences in the preferences of binding of each bZIP protein to modified DNAs (Figures S8–S11). CREB1 and ATF2 prefer binding the CRE (TGAC⁻¹GTC³A), while Zta, ATF3|cJun, and cFos|cJun prefer the TRE (TGA^C/_G⁰TC³A) (Figures S8 and S9). CREB1 more specifically prefers the five types of dsDNA than the four other bZIPs. When binding DNA(C|C), CREB1 binds the canonical CRE sequence more specifically than the all of the other bZIPs while Zta, ATF3|cJun, and cFos|cJun prefer the TRE [TGA^C/_G⁰TC³A (Figures S8 and S9B–D)].³⁷ With DNA(5mC|C), all of the bZIPs bind 8-mers with GC²AA, particularly Zta³⁷ (Figures S8 and S9E–H). With DNA(5hmC|C), many new 8-mers are strongly bound. Zta again binds strongly to some 8-mers with GC²AA (Figures S8 and S9I–L). With DNA(5fC|C), the greatest difference is between CREB1 and Zta, which bind strongly to different 8-mers with GC²AA (Figures S8 and S9M–P). With DNA(5caC|C), the bZIP domains strongly bind different 8-mers with TGAC⁻¹ (Figures S8 and S9Q–T). Upon comparison of the binding specificities between DNA(5fC|C) and DNA(5caC|C), correlation plots distinctly segregate the bindings of all five bZIPs TFs into two groups (Figure S1). TFs bind specifically with 8-mers containing GC²AA for 5fC and 8-mers with TGAC⁻¹ for 5caC (Figures S8 and S10).

We compared ATF3|cJun and cFos|cJun binding each type of dsDNA (Figure S11). With DNA(C|C), ATF3|cJun prefers the canonical CRE motif while the cFos|cJun heterodimer prefers TRE (Figure S11A). With DNA(5hmC|C), the heterodimers prefer different types of new 8-mers (Figure S11C), while with DNA(5fC|C), they prefer different 8-mers with GC²AA (Figure S11D). For DNA(5caC|C), both heterodimers prefer similar 8-mers with TGAC⁻¹ with only a few exceptions. For example, ATF3|cJun has a slight binding preference for 8-mers like ACCATGAC⁻¹ or CCATGAC⁻¹A, while cFos|cJun prefers ATGAC⁻¹TAA (Figure S11E).

bZIP-Bound Modified 8-mers Are Enriched in Bound Genomic Locations In Vivo.

We examined publicly available ChIP-seq data to determine whether 8-mers that we identify as being strongly bound when they contain modified cytosines are enriched at bound genomic regions in vivo. For this analysis, we examined a set of strongly bound 8-mers for each TF that are well-bound when they contain any modified cytosine (Table S7, Figure S12, and Experimental Procedures). We find that 8-mers that are well-bound as modified DNAs on our PBMs are enriched in ChIP-seq peaks, suggesting that these 8-mers may function in vivo. In particular, the TGATGC²AA 8-mer, which is well bound by CREB1 and ATF3|cJun when it contains modified cytosine, is >2-fold enriched in all CREB1 and ATF3|cJun peaks examined but is not enriched in Zta ChIP-seq peaks (Figure S12). In contrast, the 8-mer ATGAGC²AA, which is well-bound by Zta as modified DNAs on PBMs, is most highly enriched in Zta ChIP-seq peaks compared to the other bZIP ChIP-seq data sets examined (Figure S12).

DISCUSSION

In this study, we examined the effect of modified cytosines (5mC, 5hmC, 5fC, and 5caC) in one strand of dsDNA on sequence-specific binding of five bZIP dimers (CREB1, Zta, and ATF2 homodimers and ATF3|cJun and cFos|cJun heterodimers). For all five bZIPs, 5mC, 5hmC, and 5fC but not 5caC strengthened binding to 8-mers containing CEBP half-site GC²AA (Figures 2–6 and Figure S2–S6 and S8). At C⁻¹, 5mC, 5hmC, and 5fC inhibit binding while 5caC is bound similarly to cytosine for all five bZIP proteins. We also find several differences. dsDNA containing 5caC in some TRE (AP1)-like sequences, e.g., TGAC⁻¹TAA, is better bound by Zta, ATF3|cJun, and cFos|cJun (Figures 5 and 6 and Figures S5 and S6). Finally, examination of ChIP-seq data indicates that many 8-mers that are well-bound by bZIP proteins when they contain modified cytosines are likely bound in vivo.

A possible structural explanation for similar binding to 5mC, 5hmC, and 5fC containing dsDNA at C² by the different bZIP domains is that in all cases the methyl group of the conserved alanine can form a stabilizing hydrophobic interaction with the added carbon of these modifications. 5hmC and 5fC do not clash with the dsDNA. Some of these strongly bound 8-mers with 5hmC, 5fC, and 5caC are interesting structural targets for gaining more insight into sequence-specific dsDNA binding.

Previously, we discussed how 5mC and 5hmC change binding to dsDNA for CREB1³⁶ and Zta.³⁷ 5mC mimics the methyl group of thymine, particularly for binding CEBP half-site GC²AA. 5fC can be accommodated like 5hmC, while 5caC, which has a second bulky oxygen atom, clashes with both the protein and/or dsDNA depending on the rotamer.

An interesting aspect of our experimental approach is we can examine binding to 8-mers containing multiple modified cytosines. The presence of multiple modified cytosines in the same dsDNA sequence is less studied, and many of these sequences are novel. Modified cytosines are known to change the structure of dsDNA. ^64–66 5fC alters dsDNA structure more than 5mC or 5hmC.²⁷ These novel 8-mers may be interesting targets for further structural studies.

5caC strengthens the binding of Zta to the TRE more than the binding of CREB1 to the CRE.³⁷ A structural difference between binding the CRE (T⁻⁴G⁻³A⁻²C⁻¹G¹T²C³A⁴) and the related TRE (T^–4G^–3A^–2C/_G⁰T²C³A⁴) is that a conserved arginine from each monomer binds a guanine at the center of the CRE dyad. In the case of the TRE, which is a deletion of a nucleotide at the center of the dyad, has one arginine in the major groove interacting with guanine while the arginine from the second monomer is out of the major groove and in the water.⁶⁷ For the binding of CREB1 to a carboxylated CRE, there are steric conflicts with the added carboxylate groups at C⁻¹. However, with rearrangement of the central arginines, one could argue that the structure is stabilized by a stabilizing salt bridge between the carboxylate and one of the arginines or both.³⁶ With Zta binding the TRE (2C9N), there is more room in the center; however, the added carboxylate makes a serious conflict with the central arginine. Again, with some rearrangement of the side chain, this could become a stabilizing salt bridge. This may contribute to the different effects 5caC has on binding the TRE and CRE.

It will be interesting to expand this approach to additional TF families, as has been done previously for 5mC.³¹ A more complete analysis of additional TFs binding to modified dsDNA will help evaluate how widespread binding specificity changes occur.

ABBREVIATIONS

5mC

5-methylcytosine

5hmC

5-hydroxymethylcytosine

5fC

5-formylcytosine

5caC

5-carboxylcytosine

CREB1

cyclic AMP response element binding 1

ATF3

activating transcription factor 3

cJun and cFos

proto-oncogene

dsDNA

double-stranded DNA

PBM

protein binding microarray

SNVs

single-nucleotide variants