The Myc-associated zinc finger protein (MAZ) works together with CTCF to control cohesin positioning and genome organization

Significance

The protein CTCF plays a major role in large-scale organization of the genome. Binding sites for the protein MAZ are found adjacent to many CTCF sites. We show that, at such double sites, MAZ stabilizes CTCF binding. MAZ, like CTCF, acts independently as an “insulator” element to block the effects of a distal enhancer on a promoter, and, like CTCF, it can block the advance of a transcribing RNA polymerase II, leading to alternative RNA splicing patterns. Depletion of MAZ causes loss of short-range interactions within the nucleus and disruption of some longer-range interactions. Thus, MAZ plays a complementary role to CTCF in the nucleus, enhancing the organizational properties of CTCF and displaying many functions related to genome organization.

Abstract

The Myc-associated zinc finger protein (MAZ) is often found at genomic binding sites adjacent to CTCF, a protein which affects large-scale genome organization through its interaction with cohesin. We show here that, like CTCF, MAZ physically interacts with a cohesin subunit and can arrest cohesin sliding independently of CTCF. It also shares with CTCF the ability to independently pause the elongating form of RNA polymerase II, and consequently affects RNA alternative splicing. CTCF/MAZ double sites are more effective at sequestering cohesin than sites occupied only by CTCF. Furthermore, depletion of CTCF results in preferential loss of CTCF from sites not occupied by MAZ. In an assay for insulation activity like that used for CTCF, binding of MAZ to sites between an enhancer and promoter results in down-regulation of reporter gene expression, supporting a role for MAZ as an insulator protein. Hi-C analysis of the effect of MAZ depletion on genome organization shows that local interactions within topologically associated domains (TADs) are disrupted, as well as contacts that establish the boundaries of individual TADs. We conclude that MAZ augments the action of CTCF in organizing the genome, but also shares properties with CTCF that allow it to act independently.

The Myc-associated zinc finger protein (MAZ) was identified and characterized as a regulatory protein associated with MYC gene expression (1), and independently as a regulatory factor called Pur-1 that can activate an insulin promoter in HeLa cells (2). It was also identified as a serum amyloid A-activating factor and named SAF-1 (3). Previous studies have shown that MAZ plays important roles in regulation of a wide variety of genes (4–6). MAZ is a six zinc finger protein with a G-rich binding motif. It shares a part of its motif with other factors, notably the three zinc finger protein SP1, but binding affinity of MAZ is greater to MAZ sites that have been tested (6).

Although MAZ acts as a transcription factor, our interest in MAZ initially related to the observation that, in K562 human erythroleukemia cells, there is an unusually high correlation of sites occupied by CTCF, the chromatin architectural factor, with adjacent bound MAZ (7). We confirm this correlation in K562 and other cell types. This raises the question of whether the properties of CTCF/MAZ double binding sites differ from those of CTCF-only sites. Most CTCF sites are associated with cohesin (8–12). CTCF is known to help form chromatin loop domains generated by the action of cohesin, which brings together pairs of distant sites on chromatin. Defined loop boundaries are created when a traveling cohesin encounters a bound and correctly oriented CTCF molecule (13, 14). We asked whether CTCF-associated MAZ could contribute to this process.

We show here that, in K562 cells, the cohesin component Rad21 is significantly more likely to be associated with a CTCF/MAZ double site than with a site where CTCF alone is bound. Consistent with the abundance of CTCF/MAZ/Rad21 sites, we find that MAZ and Rad21 interact with each other in nuclear extracts as well as in an in vitro pull-down assay, suggesting that the free energy of this interaction contributes to the effect of MAZ binding on CTCF affinity for DNA. There are a relatively small number of genomic sites where MAZ is bound independently of CTCF but is associated with Rad21. Depletion of MAZ results in loss of Rad21 from these sites.

These results suggest that MAZ plays a role in genome organization that supports and complements that of CTCF. In many genomic locations, it binds to sites adjacent to CTCF, stabilizes CTCF binding, and provides an additional contribution to the arrest of cohesin. Consistent with an ability to interfere with such processive mechanisms, MAZ can pause the RNA polymerase (Pol) II phosphorylated at serine 2 of the C-terminal domain (Pol II S2P), a form of Pol II involved in transcription elongation. Furthermore, MAZ protein also can serve as an insulator in assays like that used to demonstrate the insulating properties of CTCF. MAZ thus, in part, plays a complementary role to CTCF in genome organization. This is confirmed by Hi-C studies of K562 cells in which depletion of MAZ reduces contacts within topologically associatied domains (TADs) as well as at TAD borders.

Results

MAZ Colocalizes with Cohesin in the Nucleus.

Previous studies have reported that, in K562 cells, about 45% of CTCF sites are adjacent to MAZ binding sites (7, 15, 16). We confirmed this high level of proximity by analyzing a set of ENCODE chromatin immunoprecipitation sequencing (ChIP-seq) data generated from K562, HepG2, and HeLa cells (Fig. 1A). In these cell lines, 25% (HeLa) to 58% (K562) of CTCF sites are located next to sites occupied by MAZ. Since K562 cells show the greatest overlap between CTCF and MAZ, we focused our subsequent analysis on these cells. To characterize this overlap further, we analyzed the separation distances between CTCF motifs within known CTCF binding sites and MAZ motifs within MAZ binding sites. MAZ shows some preference for binding upstream of the CTCF N terminus (Fig. 1B). The peak separation distance is 40 to 50 bp. CTCF plays an important role in arresting the movement of cohesin and helping to establish chromatin loop domains (17–21). The presence of MAZ at a large fraction of CTCF sites raises the question whether MAZ contributes in any way to cohesin positioning. In K562 cells, the majority of MAZ sites that colocalize with the cohesin component Rad21 are at CTCF sites. About 70% of CTCF binding sites in K562 cells are also occupied by cohesin (Fig. 1C), so it is not surprising that cohesin is present at CTCF/MAZ double sites. However, we find that CTCF sites in K562 cells that are adjacent to MAZ are more likely to sequester Rad21 (88% of sites) than are CTCF sites that are not near MAZ (56%; Fig. 1D). Consistent with this observation, a double knockdown of MAZ and CTCF results in significantly greater loss of Rad21 than knockdown of CTCF alone (Fig. 1E).

Fig. 1.

CTCF sites associated with MAZ. (A) Genome-wide overlap of CTCF and MAZ binding sites in K562, HepG2, and HeLa cell lines. ChIP-seq data for these cell lines were downloaded from ENCODE and used to measure overlap between the CTCF and MAZ binding sites (SI Appendix, Methods and Table S6). (B) Distance distribution for the separation of MAZ motifs within measured MAZ binding sites from CTCF motifs within measured CTCF binding sites. (Left) Distance of MAZ motif from CTCF motif on the CTCF plus strand and (Right) distance of MAZ motif from CTCF motif on the CTCF minus strand. Horizontal axis is the distance to CTCF motif; zero marks the CTCF motif center; vertical axis is the number of CTCF binding sites. (C) Genome-wide overlap of CTCF and Rad21 binding sites in K562 cells. The ChIP-seq data used for overlapping analysis were obtained from ENCODE (SI Appendix, Table S6). (D) Percentage of CTCF sites associated with Rad21 in K562 cells in the absence (−) or presence (+) of adjacent bound MAZ. (E) Rad21 site occupancy in K562 cells after knockdown of CTCF (−CTCF) or both CTCF and MAZ (−MAZ-CTCF). Two independent data sets are shown. (F) Distribution of MAZ binding sites in the genome.

These results suggest that MAZ contributes to localization of cohesin in the genome. This could be because MAZ acts synergistically with adjacent CTCF and arrests cohesin only in cooperation with it. Alternatively, MAZ might be able to arrest cohesin sliding independently of CTCF. To investigate these possibilities, we calculated the binding changes compared to controls in the ChIP-seq data generated from the siMAZ and siCTCF knockdown K562 cells. Our analysis confirmed a significant decrease in CTCF and Rad21 binding at the CTCF and MAZ double sites upon MAZ depletion (Fig. 2 A and B). In addition, both CTCF and Rad21 binding were further reduced at the CTCF and MAZ double sites in the CTCF and MAZ double-knockdown cells (SI Appendix, Fig. S1 A–D). In contrast, CTCF and Rad21 binding were not significantly altered at CTCF-only sites upon MAZ depletion (Fig. 2 A–C). We note that the CTCF and Rad21 binding sites that exhibited reduced binding upon MAZ depletion showed 75% overlap in position and had similar distribution patterns in the genome (Fig. 2 D–F).

Fig. 2.

MAZ is required for stabilizing the binding of both CTCF and cohesin in chromatin. (A) Box plot of CTCF binding changes after MAZ KD (siMAZ) compared to control at CTCF and MAZ double binding sites (1), at CTCF-only binding sites (2), or at MAZ-only binding sites (3). P values for all comparisons were calculated by the two-sided Kruskal–Wallis H-test (***P < 0.001). (B) Box plot of Rad21 binding changes after MAZ KD compared to control at CTCF and MAZ double-binding sites (1), at CTCF-only binding sites (2), or at MAZ-only binding sites (3). P values for all comparisons were calculated by the two-sided Kruskal–Wallis H-test (***P < 0.001). (C) Representative plots showing that both CTCF and Rad21 binding were reduced at CTCF and MAZ double-binding sites after MAZ knockdown. Identity of each ChIP-seq is listed on the left, and two replicates are shown in blue for CTCF IP and in red for Rad21 IP. (D) Overlap between the reduced numbers of CTCF binding sites and Rad21 binding sites that were analyzed in A and B. After MAZ knockdown, 18,945 CTCF binding sites showed reduced binding, and 18,570 Rad21 binding sites showed reduced binding compared to that in the control. About 74% of these sites overlapped. (E) Genome-wide distribution of the CTCF binding sites that showed reduced binding after MAZ knockdown in A. (F) Genome-wide distribution of the Rad21 binding sites that showed reduced binding after MAZ knockdown in B.

Not surprisingly, genome-wide annotation reveals that MAZ sites have a genomic distribution similar to those of CTCF (Fig. 1F). Among these sites, a number of MAZ/Rad21 sites (∼3,000) in K562 cells that were not associated with CTCF lost Rad21 upon MAZ knockdown, indicating that MAZ does not require CTCF to arrest cohesin movement (Fig. 3 A and B). Thus, MAZ and cohesin can be associated in the genome in the absence of CTCF, and, at such sites, loss of MAZ results in loss of cohesin (Fig. 3B).

Fig. 3.

Interactions of MAZ with cohesin. (A) Rad21 lost from binding sites after MAZ knockdown, showing that most such sites are not associated with CTCF (CTCF,-2973). Sites shown are common to three separate ChIP-seq experiments for CTCF and for Rad21 before and after MAZ knockdown. MAZ binding sites are restricted to those sites detected from the ENCODE MAZ ChIP-seq data. (B) MAZ-dependent binding of Rad21 at a site that has no CTCF. Examples of ChIP-seq data showing that Rad21 binding is reduced after MAZ depletion at MAZ sites lacking CTCF binding. (C) Purification of MBP-tagged MAZ protein. MBP-tagged MAZ was expressed in bacteria and purified with MBP affinity beads. (D) Pull-down assay. Purified MBP-MAZ was added to nuclear extracts from K562 cells and immunoprecipitated with anti-MBP beads. Western blot shows that Rad21 coprecipitates with MBP-MAZ. (E) Co-IP. Flag-MAZ was expressed in HEK293T cells, then purified with anti-Flag beads. Western blot shows that Rad21 coprecipitates with Flag-MAZ. (F) Pull-down assay. MBP-tagged MAZ, expressed in bacteria and purified with MBP affinity beads, was used to immunoprecipitate Rad21 expressed in vitro in the TnT Coupled Wheat Germ Extract Systems (SI Appendix, Table S5). Western blot shows that Rad21 interacts with MBP-MAZ. (G) CTCF is more stable when MAZ is nearby (+). Shown are results of two independent experiments in which CTCF protein was partially depleted with siRNA. ChIP-seq data were used to measure CTCF binding in K562 cells. After CTCF depletion, more CTCF binding sites are retained when MAZ is nearby than in the absence of MAZ (−).

Interaction of MAZ and Rad21.

The frequent genome-wide association of MAZ and Rad21 raised the question whether the two proteins interact physically. To test this possibility, we purified bacterially expressed MBP-tagged MAZ and used it in coimmunoprecipitation experiments to incubate with nuclear extracts from K562 cells (SI Appendix, Methods). The MBP pull-down assay showed that Rad21 was present in the anti-MBP immunoprecipitated fraction (Fig. 3 C and D). In an alternative experiment, we expressed Flag-tagged MAZ in HEK293T cells. Anti-Flag immunoprecipitation showed that Rad21 coprecipitates with Flag-tagged MAZ (Fig. 3E). These results were confirmed by mass spectrometry analysis of similar anti-Flag immunoprecipitates. As shown in SI Appendix, Table S1, among the proteins that coprecipitate with MAZ are many of the components of the cohesin complex, including Rad21, SMC1A, and SMC3.

Although this establishes a MAZ–cohesin interaction, it does not distinguish between direct interaction and interactions mediated by a cofactor in the extract. We therefore repeated pull-down experiments using purified MBP-tagged MAZ expressed in bacteria and Rad21 expressed in the TnT Coupled Wheat Germ Extract Systems (Methods). As shown in Fig. 3F, the two proteins coprecipitate, confirming that Rad21 and MAZ interact directly. However, cohesin components SMC1A and SMC3 do not interact directly with MAZ (SI Appendix, Fig. S2B).

Given the proximity of MAZ binding sites to many CTCF sites, we asked whether there was any physical interaction between MAZ and CTCF. Coimmunoprecipitation and mass spectrometry studies described above to identify potential MAZ cofactors did not reveal direct or indirect MAZ–CTCF interaction (SI Appendix, Table S1). However, it seemed possible that CTCF and MAZ might interact in vivo when bound to adjacent DNA sites. To investigate this, we knocked down CTCF in K562 cells (Methods) and mapped the vacant CTCF sites that were created. In these experiments, CTCF binding sites were only partially depleted (∼60%). About 58% of sites that retained CTCF were adjacent to bound MAZ; in contrast, only ∼30% of sites not next to MAZ retained CTCF (Fig. 3G), suggesting that MAZ binding helps stabilize CTCF binding (Discussion).

MAZ at Regulatory Sites: Partnership with CTCF.

In K562 cells, 78% of MAZ binding sites are marked by H3K4 trimethylation (H3K4Me3), a modification typically associated with active promoters (22) (Fig. 4A), suggesting that MAZ may play a regulatory role in expression of a wide variety of genes (Discussion). We asked whether CTCF was associated with MAZ at these MAZ-H3K4Me3 regions. We find that 41% (12,501 of 30,371) of the CTCF sites at these regions are adjacent to bound MAZ. In contrast, very few CTCF sites not associated with MAZ are marked by H3K4 trimethylation (Fig. 4B).

Fig. 4.

MAZ and gene transcription. (A) Frequency of MAZ binding at sites associated with histone H3K4 trimethylation in K562 cells. (B) H3K4Me3 methylation at CTCF sites associated with MAZ (CTCF/MAZ+) and those not associated with MAZ (CTCF/MAZ−), showing that H3K4Me3 methylation is strongly associated with CTCF sites that are adjacent to MAZ. (C) RNA-seq data analysis results. Of the genes that are changed in expression twofold or greater following MAZ knockdown, 162 up-regulated and 204 down-regulated genes are associated with both MAZ binding and H3K4 trimethylation. The fraction of those sites associated with CTCF is shown compared to the fraction of all MAZ/H3K4Me3 sites in the whole genome. Dark green indicates sites with bound CTCF; light green indicates sites without bound CTCF. (D) Volcano plot showing distribution of changes in gene expression following MAZ knockdown. Up, up-regulated genes (P < 0.01); down, down-regulated genes (P < 0.01); NoSignifi, genes without significant changes. (E) GO analysis for the most affected genes following MAZ knockdown.

To explore further the regulatory role of MAZ, we analyzed RNA-sequencing (RNA-seq) data from ENCODE for the effects of MAZ knockdown on gene expression in K562 cells (SI Appendix, Table S2). These data show that 290 genes are up-regulated and 277 are down-regulated twofold or more when MAZ is depleted. Of these, 162 of the up-regulated and 204 of the down-regulated genes are associated with MAZ/H3K4Me3 sites (Fig. 4C). Surprisingly, 75 to 80% of these are adjacent to CTCF sites, whereas only 27% of all MAZ-H3K4Me3 sites are shared with CTCF in the whole genome. Thus, genes with expression most sensitive to MAZ depletion appear to be enriched in CTCF binding at sites associated with promoters (Fig. 4C). A summary of the effects of MAZ knockdown on expression of the entire gene population is given in a volcano plot in Fig. 4D, and a list of the affected genes is shown in SI Appendix, Table S2. The results of a Gene Ontology (GO) analysis for the most affected genes are presented in Fig. 4E. We note that MAZ appears to be associated with regulation of the mitotic cell cycle, which may be connected to its interaction with the cohesin complex (23).

MAZ as an Insulator Protein.

CTCF is well known to function as an “enhancer blocking” factor when placed between an enhancer and promoter (24, 25). To determine whether MAZ has similar properties, we made use of a construct similar to those used in many studies of CTCF, in which a sequence that binds the insulator protein to be tested is placed between an enhancer and a promoter that drives expression of luciferase (25). As shown in Fig. 5 A and B, the construct with a MAZ binding motif, when transfected into HEK293T or HeLaS3 cells, expressed luciferase at lower levels than were observed with a construct in which the MAZ binding motif was mutated, suggesting that MAZ acts as an insulator to block the interaction between enhancer and promoter.

Fig. 5.

MAZ acts as an insulator protein. (A, Top) Schematic diagrams of the reporter constructs pIHLMAZwtE and pIHLMAZmuE that were used in the insulation assay (wild type, pIHLMAZwtE; mutation, pIHLMAZmuE). A luciferase reporter is surrounded on each side either by sequences containing MAZ binding sites (see text) or mutated sites that do not bind MAZ. Each of the constructs was transfected into HEK293T or HeLaS3 cells with a Renilla luciferase control plasmid. Cell lysates were prepared 48 h after transfection with plasmids and used to measure luciferase activities. The relative luciferase activity for each construct was normalized to that of Renilla luciferase. (A, Bottom) Insulation assay results in HEK293T and HeLaS3 cells transfected with pIHLMAZwtE and pIHLMAZmuE. Wild type MAZ binding sites inserted between promoter and enhancer reduced luciferase expression, whereas mutated MAZ binding sites did not. (B) DNA sequences of wild type and mutated MAZ binding sites in constructs used in A. (C) Western blot showing MAZ knockdown efficiency. (D) Quantitative analysis of protein expression in C. (E) MAZ mRNA expression level after MAZ knockdown. (F) Effect of MAZ depletion on luciferase expression. Knockdown of MAZ results in loss of enhancer-blocking function. (G) Overexpression of Flag-MAZ in HEK293T cells. Western blot showing Flag-MAZ expression. (H) Effect on luciferase activity in cells transfected with either Flag-MAZ or Flag-only plasmid. Overexpression of Flag-MAZ increases enhancing blocking function (mean of n ≥ 3, error bar indicates ±SD).

To confirm this property of MAZ, we repeated the experiments with the constructs containing the MAZ binding sites, but used RNA interference (RNAi) to knock down MAZ expression (Fig. 5 C–E). This resulted in loss of the enhancer-blocking effect of the MAZ element: luciferase expression from the wild type MAZ construct (pIHLMAZwtE) was almost the same as that from the mutant construct (pIHLMAZmuE) lacking MAZ binding (Fig. 5F). In complementary experiments, we introduced a plasmid that resulted in overexpression of MAZ (Fig. 5G). Consistent with a role for MAZ as an enhancer-blocking insulator element, luciferase expression decreased when MAZ concentration increased (Fig. 5H). To test whether MAZ binds to sites in the constructs that have two MAZ binding sites but no CTCF binding site, we transfected the constructs into HEK293T cells and performed ChIP-qPCR analysis. As shown in SI Appendix, Fig. S3 A and B, MAZ binding was approximately twofold higher in the wild type construct pIHLMAZwtE than in the mutated construct pIHLMAZmuE, suggesting that mutated MAZ binding sites prevented MAZ binding.

Thus, MAZ, like CTCF, can block the action of a distal enhancer. This may be related to the shared ability of these two proteins to interfere with progression of RNA Pol II (see below).

MAZ, Pausing of RNA Pol II, and Alternative Splicing of RNA.

The ability of MAZ to arrest cohesin suggested that it might also impede the progress of transcribing RNA Pol II. Previous studies that addressed this question with small designed templates gave ambiguous results, because, under some circumstances, DNA containing MAZ binding motifs can form G-quadruplex structures (26) (Discussion). To address this problem, we analyzed ChIP-seq data in K562 cells for the elongating form of RNA Pol II (Pol II S2P) and identified the Pol II S2P sites that were adjacent to bound MAZ that was not also associated with CTCF. We then used RNAi to knock down MAZ (SI Appendix, Methods) and determined the effect on Pol II positioning. Of the 4,735 Pol II S2P sites, 1,707 are associated with MAZ. Of these, 809 were no longer present after MAZ was depleted (Fig. 6A and SI Appendix, Fig. S7A). Many of the sites that lost Pol II S2P were near promoters, but 274 were not. Those Pol II S2P sites not initially at promoters were fairly evenly distributed among exons, introns, and intergenic regions (Fig. 6B and SI Appendix, Fig. S7B). Typical examples of sites at which MAZ and Pol II S2P colocalize, and which lose Pol II S2P when MAZ is depleted, are shown in Fig. 6C. These results indicate that bound MAZ, in the absence of CTCF, can pause the advance of transcribing Pol II.

Fig. 6.

MAZ and Pol II S2P pausing in K562 cells. (A) Distribution of Pol II S2P in TSS regions in control and MAZ knockdown cells (P < 0.001). (B) Distribution of Pol II S2P in exons before and after MAZ knockdown (P < 0.001). P value for Pol II S2P overlap in the promoter region (TSS) and gene body region between the control and MAZ knockdown was calculated using the permTest function with ntimes = 1,000 in the regioneR package (48). (C) Examples of Pol II S2P ChIP-seq data showing loss of Pol II S2P at MAZ sites in the gene body after MAZ knockdown. (D) Global effects of MAZ binding sites on mRNA splicing. (Top) Schematic diagram showing the location of the alternative exons relative to the MAZ binding sites. An upstream MAZ binding site is defined as within −1,000 bp from the exon, and a downstream binding site as within +1,000 bp from the exon. To rule out the impact of CTCF, the MAZ binding sites that overlap with CTCF binding sites were removed in this calculation. (Bottom) Difference in the mean exon inclusion level between K562 control and MAZ knockdown (calculated by using data in SI Appendix, Table S7). The mean ± SEM for each class of exon is plotted against Bayes factor thresholds. Black bars indicate inclusion exons do not overlap with MAZ binding sites or do not have MAZ binding sites nearby (1,000 bp); red bars indicate inclusion exons have MAZ binding sites upstream; blue bars indicate that inclusion exons have MAZ binding sites downstream (*P < 0.05, ** P < 0.01, Wilcoxon rank-sum test for difference in exon inclusion at the different thresholds). (E) Pol II S2P reads centered on the corresponding downstream MAZ binding sites or the upstream alternative exons. (F) An example showing that depletion of MAZ affects alternative splicing in K562 cells. This example shows a reduction in the relative inclusion of exon 2 of the gene CMTR2 after MAZ is depleted. Red arrow indicates the position of the second exon. MAZ ChIP-seq data shows that MAZ is present in the exon 2 region of CMTR2.

Pausing of Pol2 has been shown previously to be an important property of bound CTCF and to lead to increased choice of alternate exons for splicing (27, 28). We asked whether sites occupied by MAZ (but not CTCF) had similar properties. The effect of MAZ knockdown on inclusion of alternatively spliced exons was measured for exons within 1 Kb downstream or upstream of a MAZ+/CTCF− site, as well as for exons not near such sites. As shown in Fig. 6D, presence of such a site downstream of the exon had a strong positive effect on inclusion of the exon, while an upstream location had relatively little effect, as did complete absence of a MAZ site. Consistent with the observation that MAZ can pause the elongating form of Pol II, we found that Pol II S2P was enriched at downstream MAZ binding sites relative to upstream exons (Fig. 6E). An example of MAZ effects on splicing is shown in Fig. 6F. Thus, MAZ strongly promotes Pol II pausing-associated splicing in the absence of CTCF.

Location of MAZ Between Closely Spaced Genes.

The early studies of MAZ binding-site effects on RNA Pol II noted that a genomic MAZ site located between a pair of closely spaced genes can contribute to transcription termination and prevent interference with the downstream gene (29, 30). We examined the genomic population of closely spaced genes and divided it into pairs oriented in the same direction, those facing (transcribing) toward each other and those facing away from each other, in all cases with a separation from 2 kb to 5 kb. We then asked in each case what fraction had an occupied MAZ binding site in K562 cells. Over 40% of closely spaced genes that are transcribed in the same direction have a MAZ site between them (SI Appendix, Fig. S7C). This probably reflects the high incidence of MAZ sites near promoters (Fig. 4A), which also explains the observation that over 60% of divergent genes (transcribed away from each other) are separated by MAZ, since the presence of two promoters in the intergenic space increases the probability of finding at least one MAZ site. Similarly, the absence of a promoter explains why only 20% of convergent genes (transcribed toward each other) have a MAZ site between them. It therefore does not appear that MAZ binding sites are present in unexpectedly high abundance between closely spaced genes. However, our RNA Pol II pausing data support the idea that a MAZ site between closely spaced genes transcribed in the same direction will impede the incursion into the downstream gene of a polymerase transcribing the upstream gene, as previously suggested (29, 30).

Effects of MAZ Depletion on Large-Scale Genome Organization.

Given the association of MAZ with cohesin, we asked what effect depletion of MAZ might have on large-scale genome organization. We carried out Hi-C analysis comparing genomic interactions in control K562 cells with interactions in cells in which MAZ had been knocked down. After evaluating the quality of the Hi-C libraries, we pooled the data and detected about 150 million normal paired reads for K562 control and K562 siMAZ knockdown cells (SI Appendix, Table S3). The whole-genome chromosomal interaction heat map is similar between the control and MAZ-depleted cells. The A and B compartments also are not significantly changed after depletion of MAZ (SI Appendix, Fig. S4).

Differential TAD analysis (Fig. 7 A and B) shows that cells depleted of MAZ have reduced interactions within a TAD compared to those in control cells, whereas there is little or no effect on inter-TAD contacts. Consistent with this conclusion, an aggregate TAD analysis (ATA) shows a marked decrease in intra-TAD contact frequency, corresponding to the strong blue diagonal. The blue points at the 5′ and 3′ borders of the ATA analysis indicate that loss of MAZ also decreases contact frequency in the 5′ and 3′ boundaries of individual TADs (Fig. 7B). In agreement with this, the genome-wide insulation scores at 10-kb resolution aligned to the TAD borders are lower in cells in which MAZ is depleted compared with those in control cells, indicating that MAZ knockdown reduced the border strength of TADs (Fig. 7C and SI Appendix, Fig.S5 A and B). Aggregate peak analysis reveals that most of these changes involve short-range contacts (Fig. 7D), but some loop domains encompassing more than one TAD are also down-regulated when MAZ is depleted (Fig. 7E).

Fig. 7.

MAZ and genome organization in K562 cells. (A) Hi-C data analysis was conducted using the GENOVA R package (18). Dot plot of differential TAD analysis results. K562 cells depleted of MAZ have fewer interactions within a TAD (dark blue; TAD distance N+0) compared to those in K562 control cells. Log₂ ratio of the contact frequency in siMAZ compared to control is plotted. (B) Aggregate TAD analysis results. The differential signal between K562 control and K562 siMAZ was calculated at 10-kb resolution. Blue indicates reduced signal, and red indicates increased signal. There is a decrease in contacts within the TADs and at TAD borders in the K562 siMAZ. (C) Alignment of border strength of TADs. The border strength of TADs is reduced in K562 siMAZ compared to that in K562Ctr. Genome-wide insulation scores were calculated in 10-kb resolution using the findTADsAndLoops.pl script in Homer. Heat map was generated with deepTools. Circles mark borders of TADs. (D) Aggregate peak analysis at 2.5-kb resolution. (E) Example of Hi-C contact matrices for a zoomed-in region on chromosome 11 (chr11:1,500,000–2,500,000). This region covers the imprinting control region of the h19 locus. Heat map was generated at 5-kb resolution using the Juicer program. The upper right half contact matrices are generated from K562Ctr, while the lower left half contact matrices are generated from K562siMAZ. The locations showing changes of contact frequency are marked in red circles. Loops shown below the heat map were generated with the hichipper program and visualized with WashU Epigenome Browser. Virtual 4C maps were generated using the coordinate chr11:2,015,001–2,020,000, which corresponds to the location of h19 gene, as viewpoint. ChIP-seq and gene tracks were generated with the pyGenomeTracks program (Methods).

These changes show that, although MAZ does not have the large-scale effects on genome organization associated with CTCF, it nonetheless appears to play an important role in the pattern of genomic contacts, principally within TADs, and it contributes to the integrity of TADs.

Discussion

We have shown that MAZ, often bound to DNA in association with CTCF, plays an important role in stabilizing CTCF binding and directly or indirectly contributes to the arrest of the cohesin complex at CTCF sites. To a lesser extent, it also contributes independently of CTCF to the organization of the genome through its interaction with cohesin. MAZ interacts directly with the Rad21 component of the cohesin complex and indirectly with other members of the complex. Although the majority of MAZ/cohesin (Rad21) sites in the genome are adjacent to CTCF sites, we identify in K562 cells about 3,000 MAZ/Rad21 sites that are not associated with CTCF, and find that many MAZ/Rad21 sites lose Rad21 when MAZ is knocked down (Fig. 3 A and B).

MAZ also makes an indirect contribution to cohesin binding when it occupies a site adjacent to bound CTCF: almost 90% of CTCF/MAZ sites in K562 cells are associated with cohesin, compared to 56% of CTCF-only sites (Fig. 1D). Depletion of both MAZ and CTCF results in loss of cohesin from some of those sites that is significantly greater than that resulting from depletion of CTCF alone (Fig. 1E). The presence of MAZ also helps to stabilize CTCF binding at such double sites. Knocking down CTCF results in preferential loss of CTCF from sites that are not adjacent to MAZ (Fig. 3G). It seems possible that cohesin, bound to both MAZ and CTCF at these double sites, may be responsible for the observed contribution of MAZ to CTCF binding affinity. The bound MAZ–CTCF–cohesin cluster would then constitute a mutually stabilizing complex (Fig. 8A).

Fig. 8.

Model of MAZ action. (A) Proposed model for the effect of bound MAZ on adjacent CTCF, in which the shared interaction with cohesin (red ring) provides increased stability of CTCF and cohesin. Loss of bound MAZ from these sites reduces CTCF binding affinity. (B) Pol II S2P sliding arrested by MAZ. (C) Loss of bound CTCF may allow cohesin to slide to a nearby MAZ site.

MAZ shares other properties with CTCF. In an experiment similar to one used frequently (25, 31) to demonstrate the enhancer-blocking properties of CTCF, we incorporated MAZ sites between an enhancer and the promoter of a reporter gene (Fig. 5). When the wild type construct is transfected into HEK293T or HeLa S3 cells, expression of the reporter gene is inhibited relative to that of the construct with mutated MAZ sites. Depletion of MAZ protein increases expression of the reporter gene, corresponding to decreased insulation. On the contrary, overexpression of MAZ decreases expression of the reporter gene, as expected if bound MAZ is blocking the effect of the upstream enhancer.

MAZ Binding, Pausing of RNA Pol II, and G Quadruplex Formation.

Yonaha and Proudfoot first observed (30) that an array of MAZ binding sites could pause transcribing RNA Pol II, and, subsequently, they found that this property was not necessarily dependent on binding of the MAZ protein (32, 33). These results showed that the MAZ binding motif alone, when present in multiple copies, was sufficient to impede RNA Pol II transcription. The canonical MAZ binding site has the central sequence G3AG3, though other related sequences also function as binding sites (34). Similarly, it has been shown that a series of four MAZ sites inserted in a minigene construct can interfere with transcription elongation and affect splicing choices (5). As noted above, MAZ sites not associated with CTCF are frequently located at promoters, and a genome-wide survey shows that, at many promoters, there are multiple MAZ binding motifs. As has been pointed out in many papers (26, 35), when such G-rich sequences are located close to one another and present in a transcribed region, the G-rich strand as well as a G-rich RNA transcript may be able to adopt the G quadruplex structure. It seems probable that G quadruplex formation is at least partially responsible for the behavior reported by Yonaha and Proudfoot (30). More recently, sequences with the potential to form G quadruplex have been identified upstream of a variety of genes, and G quadruplex-specific antibodies have revealed the presence of such sequences at TAD domain boundaries (36, 37). It should be remembered, however, that antibodies specific for a non-B form DNA will stabilize that form (38, 39). Thus, antibodies reveal only the presence of potential G quadruplex-forming sequences.

This raises the question of the relationship of MAZ binding to quadruplex structure and the role of each in regulation of gene expression. In the case of single binding motifs, or where multiple motifs are not close together, G quadruplex will not form. Most importantly, although there is evidence that MAZ can bind to the quadruplex, published data clearly show that it binds more strongly to the corresponding duplex (40), so that bound MAZ will preferentially stabilize its binding site in the duplex conformation.

In resolution of this question, two kinds of evidence support the conclusion that important properties of MAZ sites depend on bound MAZ, but not the MAZ motif alone. First, the ability of a MAZ binding site to act as an enhancer-blocking insulator depends on binding of the MAZ protein (Fig. 5). Overexpression of MAZ increases blocking activity, and depletion of MAZ reduces it. Second, we find that, at many sites in the genome, bound MAZ, but not the MAZ motif alone, is responsible for pausing a transcribing RNA Pol (Fig. 6 A–C). We identified many sites occupied by both MAZ and the elongating form of RNA Pol II, but not occupied by CTCF (SI Appendix, Fig. S7A). Such a configuration is consistent with a role for MAZ in RNA Pol II pausing; the further observation that knocking down MAZ results in loss of both RNA Pol II and MAZ confirms that MAZ protein is both necessary and sufficient for pausing of Pol II at a MAZ binding site.

It has been shown that pausing of RNA Pol II can result in alternative splicing choices during transcription, and this has been demonstrated explicitly in the case of CTCF (28). It seems likely that MAZ would have similar effects on splicing outcomes. An example of the effect of MAZ knockdown on splicing choices is shown in Fig. 6F. We note that among the proteins that coprecipitate with MAZ (SI Appendix, Table S1) are the U1, U2, U4, U5, and U6 members of the spliceosome complex (41, 42), a property shared with CTCF.

It is interesting to compare the properties of MAZ with those of YY1, another zinc finger protein that contributes to genome organization (43–45). YY1 interacts with the SMC1 subunit of cohesin (44) and helps stabilize promoter–enhancer interactions in the genome (43, 45). In K562 cells, most YY1 binding sites (77%) are close to MAZ sites (SI Appendix, Fig. S6A). However, most (73%) of these YY1/MAZ sites are not close to CTCF (SI Appendix, Fig. S6B). Thus, one of the properties of MAZ, in stabilizing CTCF binding, is distinct from the properties of YY1. The stabilization of enhancer–promoter interactions by YY1 is proposed to involve contact between YY1 proteins bound at those sites and further strengthened by interaction with cohesin (45). Unlike YY1, many MAZ sites are located near promoters, but few are near enhancers (SI Appendix, Fig.S6 C and D). Although, like YY1, MAZ helps stabilize short-range interactions, it also affects TAD organization (see below), probably because of its ability to arrest cohesin sliding.

MAZ Contributes to Genome Organization.

CTCF plays the predominant role in organizing the genome, probably because it is so effective at arresting sliding, both of cohesin and RNA Pol II. MAZ, though it shares these properties, is probably less efficient. Nonetheless, the Hi-C data in Fig. 7 show that MAZ also plays an important role in genome organization on a different scale from CTCF. Depletion of MAZ primarily disrupts interactions within TADs, as well as reducing the contacts that establish the boundaries of individual TADs. It seems likely that the effect on intra-TAD contacts reflects the ability of MAZ to stop sliding cohesin complexes. As we have shown above, MAZ is capable not only of arresting cohesin on its own, but also contributes significantly to the ability of CTCF to do so; some effects of loss of MAZ on long-range organization may reflect loss of bound CTCF stabilized by adjacent MAZ binding. Perhaps surprisingly, the rearrangements in genomic organization seen on MAZ depletion are not reflected in widespread changes in gene expression (Fig. 4D). Significant changes are confined to a relatively small number of genes. It seems likely that large-scale genome architecture is resilient.

There is reason to think that some of the properties of MAZ are shared with other zinc finger proteins. For example, our laboratory has shown that Vezf1 is capable of pausing RNA Pol II and affecting RNA splicing outcomes (46). However, these proteins do not appear to have formed the collaborative arrangement with CTCF that is reflected in the high abundance in the genome of paired CTCF/MAZ sites. That pairing results in stabilization of the CTCF/DNA interaction and must also contribute to the arrest of cohesin. If CTCF is the “master organizer of the genome,” MAZ may be the master’s apprentice.

Methods

The details of data sources are provided in the reference list of the SI Appendix Cited Data and Table S6.

The plasmid expressing human MAZ protein was purchased from OriGen (RC219635). The human MAZ gene with Myc-Flag tags fused in the C terminus was cloned in the pCMV6 vector. Knockdown of MAZ and/or CTCF in K562 and HEK293T cells used On-TARGET plus SMARTpool siRNA oligonucleotides.

For MAZ pull-down assays, a DNA fragment encoding the human MAZ gene was amplified with PCR. All PCR products were digested with restriction enzymes Xba I and Hind IlI and cloned into XbaI/HindIII-digested pMal-C2X (New England Biolabs) bacterial expression vector, which carries an MBP tag.

ChIP was performed with a ChIP-IT Express kit (Active Motif) following the manufacturer’s instructions. The insulation assay was performed as described by Ishihara et al. (25). Hi-C libraries were prepared as previously described (47). Details of these and other methods and data analysis are contained in the SI Appendix.

Data Availability.

ChIP-seq, RNA-seq, and Hi-C data have been deposited in the National Center for Biotechnology Information Gene Expression Omnibus database (GSE143937) and are accessible.