Members of an array of zinc-finger proteins specify distinct Hox chromatin boundaries

Abstract

Partitioning of repressive from actively transcribed chromatin in mammalian cells fosters cell-type-specific gene expression patterns. While this partitioning is reconstructed during differentiation, the chromatin occupancy of the key insulator, CCCTC-binding factor (CTCF), is unchanged at the developmentally important Hox clusters. Thus, dynamic changes in chromatin boundaries must entail other activities. Given its requirement for chromatin loop formation, we examined cohesin-based chromatin occupancy without known insulators, CTCF and Myc-associated zinc-finger protein (MAZ), and identified a family of zinc-finger proteins (ZNFs), some of which exhibit tissue-specific expression. Two such ZNFs foster chromatin boundaries at the Hox clusters that are distinct from each other and from MAZ. PATZ1 was critical to the thoracolumbar boundary in differentiating motor neurons and mouse skeleton, while ZNF263 contributed to cervicothoracic boundaries. We propose that these insulating activities act with cohesin, alone or combinatorially, with or without CTCF, to implement precise positional identity and cell fate during development.

Keywords: CTCF; Hox genes in development; MAZ; PATZ1; chromatin domains; cohesin; gene regulation; genome organization; insulation; motor neurons; zinc-finger proteins.

Figure 1.. Loss of MAZ in ΔCTCF mESCs reduced RAD21 re-localization, indicating a possible barrier function for MAZ.

(A) Venn diagram showing RAD21 binding in mESCs with CTCF intact (WT, Untreated) versus CTCF degraded (ΔCTCF, Auxin treatment, 48 hr). (B) Venn diagram showing MAZ and RAD21 binding in ΔCTCF mESCs. (C) Heat maps of RAD21, CTCF, and MAZ ChIP-seq read density grouped as Cluster 1 (n=7881), Cluster 2 (n=12451), and Cluster 3 (n=10018) based on the alteration in RAD21 signal upon CTCF degradation within a 4 kb window. ChIP-seq data is from one representative of two biological replicates for CTCF and MAZ, and one biological replicate for RAD21. Additional RAD21 ChIP-seq datasets were reported earlier ^,. (D) Western blot analysis of CTCF, MAZ, RAD21, OCT4, and Histone H3 under WT (Untreated), ΔCTCF (Auxin treatment, 48 hr), ΔMAZ, and ΔCTCF/ΔMAZ conditions in ESCs. ΔMAZ represents two independent Maz KO clones. (E) Heat maps of RAD21 and MAZ ChIP-seq read density in RAD21 re-localized sites (Cluster 3, n=10018) within a 4 kb window under WT, ΔCTCF, ΔMAZ, and ΔCTCF/ΔMAZ conditions in mESCs (see Figures S1C–D for all clusters). Average density profiles for ChIP-seq under each condition is indicated above the heat map. ChIP-seq data is from one representative of two biological replicates for CTCF and MAZ, and one biological replicate for RAD21. (F-G) Normalized ChIP-seq densities for RAD21 and MAZ at (F) Nat9 and (G) Dnajb12 wherein re-localized RAD21 is reduced upon MAZ KO. RAD21-relocalized regions are indicated within dashed-lines. Arrows indicate RAD21 signal in WT that is proximal to the RAD21 re-localized regions. (H) Motif analysis of RAD21 re-localized regions in the absence of CTCF and MAZ by de novo MEME motif analysis, along with the corresponding top matches by Tomtom motif comparison (see Figure S1E for the detailed list).

Figure 2.. PATZ1 and other zinc finger proteins, ZNF263, ZNF341, and ZNF467, co-localize with RAD21 on chromatin and at loop anchors in HEK293 and HepG2 cells

(A) Venn diagram showing RAD21, CTCF, and PATZ1 binding in HEK293 cells. (B) Heat maps of RAD21, CTCF, PATZ1, MAZ, and other zinc finger proteins, ZNF263, ZNF341 and ZNF467 in HEK293 cells. ChIP-seq read density was grouped as Cluster 1, Cluster 2, and Cluster 3 based on the indicated overlaps with RAD21 signal within a 4 kb window. (C) Western blot analysis of RAD21, FLAG, and CTCF upon FLAG-PATZ1 immunoprecipitation from mESCs (n=2, see Figure S2H for biological replicate). (D) Visualization of Hi-C contact matrices for a zoomed-in region around the TBC1D1 locus in HepG2 cells. Shown below are loops with PATZ1 at both anchors in HepG2 cells, ChIP-seq read densities for RAD21, CTCF, PATZ1, and MAZ, and gene annotations. ChIP-seq data in HepG2 cells is from two combined biological replicates. (E) Percentage of Hi-C loops in HepG2 cells overlapping with RAD21, CTCF, and PATZ1 ChIP-seq peaks. (F) Heat maps of RAD21, PATZ1, ZNF263, ZNF341, and ZNF467 in HEK293 cells. ChIP-seq read density was grouped as Cluster 1, Cluster 2, Cluster 3, and Cluster 4 based on the combinatorial overlaps of zinc finger proteins with RAD21 within a 4 kb window in HEK293 cells. The model on the right side indicates combinatorial binding of the indicated factors in each cluster (see Figure S4B). ChIP-seq data in HEK293 cells is from one replicate for RAD21 and one representative of two biological replicates for others (see Table S1 for datasets).

Figure 3.. PATZ1 co-localizes with RAD21 on chromatin in mESCs

(A) Schematic of PATZ1 protein domains including zinc fingers (ZnFs), AT hook domains, and the BTB/POZ domain. (B) Western blot analysis of PATZ1, CTCF, RAD21, MAZ, OCT4, and GAPDH in WT, Patz1 KO, and FH-PATZ1 mESCs (see Figure S5A for four independent Patz1 KO clones). (C) Venn diagram showing RAD21, CTCF, and PATZ1 binding in mESCs. (D) Heat maps of RAD21, CTCF, PATZ1, and MAZ ChIP-seq read densities grouped as Cluster 1, Cluster 2, and Cluster 3 based on the indicated overlaps with RAD21 signal within a 4 kb window in mESCs. (E-F) Normalized ChIP-seq densities for RAD21, CTCF, PATZ1, and MAZ, wherein co-localizing peaks were visualized. ChIP-seq data is from one representative of two biological replicates.

Figure 4.. Loss of PATZ1 impacts RAD21 and MAZ chromatin binding in mESCs and gene expression of developmental processes

(A) Heat maps of RAD21, CTCF, MAZ, and PATZ1 ChIP-seq read densities in WT and Patz1 KO ESCs at RAD21 peaks, clustered into RAD21-decreased and -unchanged sites. Each ratiometric heat map plotted to the right shows the log₂ fold change (Patz1 KO/WT) of the signals. (B-C) Normalized ChIP-seq densities for RAD21, CTCF, MAZ, and PATZ1 at the indicated loci wherein MAZ and RAD21 signal was altered. ChIP-seq data is from one representative of two biological replicates. (D) Differentially expressed genes by RNA-seq upon Patz1 KO in ESCs from three biological replicates (see all in Table S2). (E) Gene Ontology (GO) analysis shows the categories of biological processes that were enriched in the differentially expressed genes in Patz1 KO versus WT mESCs. PANTHER overrepresentation test tools were used for GO analysis (see all in Table S3).

Figure 5.. Combinatorial binding of insulation factors at distinct Hox gene borders determines rostral-caudal patterning in MNs

(A) Normalized ChIP-seq densities for RAD21, CTCF, and PATZ1, and ZNF263 in WT, Patz1 KO, and Znf263 KO mESCs at the indicated regions in the HoxA cluster. ChIP-seq data represents one representative replicate of two biological replicates for RAD21, CTCF, and one replicate for FH-PATZ1 and FH-ZNF263. (B-D) RT-qPCR analysis for the indicated Hox genes in (B) the HoxA, (C) the HoxC, and (D) the HoxD clusters in WT and Patz1 KO cervical MNs. RT-qPCR signal was normalized to Atp5f1 and ActB levels. The fold-change in expression was calculated relative to WT MNs. All RT-qPCR results are represented as mean values and error bars indicating log₂(SE) across three biological replicates (two-sided Student’s t-test without multiple testing correction; ***P ≤ 0.001, **P ≤ 0.01, *P < 0.05). (E) Differentially expressed genes by RNA-seq upon Patz1 KO in MNs from two biological replicates (see all in Table S4). (F) GO analysis showing the top biological processes enriched in the differentially expressed genes in Patz1 KO versus WT MNs. PANTHER overrepresentation test tools were used for GO analysis and top 15 categories having a fold enrichment > 2.5 were plotted (see all in Table S5). (G-I) RT-qPCR analysis for the indicated Hox genes in (G) the HoxA, (H) the HoxC, and (I) the HoxD clusters in WT and Znf263 KO cervical MNs. RT-qPCR signal was normalized to Atp5f1 and Gapdh levels. The fold-change in expression was calculated relative to WT MNs. All RT-qPCR results are represented as mean values and error bars indicating log₂(SE) across three technical replicates. Two independent Znf263 KO clones are shown (see Figure S7).

Figure 6.. Loss of PATZ1 impacts genome organization in vitro, and leads to homeotic transformation at thoracolumbar boundaries in skeletal patterning in vivo

(A-B) Normalized APA plots of all loops in WT versus PATZ1 KO (A) ESCs and (B) MNs. The resolution of APA is 5 kb. P2LL (Peak to Lower Left) is the ratio of the central pixel to the mean of the mean of the pixels in the lower left corner. (C-D) Visualization of Micro-C contact matrices for a zoomed-in region around (C) the HoxA cluster in WT versus PATZ1 KO ESCs, and (D) Grb10 locus in WT versus PATZ1 KO MNs. Examples of the differential loops detected have been indicated with the circles. The resolution is 5 kb. Shown above and left are gene annotations. (E) Bar plot showing the percentage of lethality in Patz1 gRNA injected embryos. The numbers represent pups at postnatal day 2 by which lethality has been observed. (F) Representative Alcian blue–Alizarin red staining of axial skeletons indicating homeotic transformations in WT versus Patz1 KO mice at postnatal day 0.5. Additional thoracic vertebra, L1 gain of rib, (middle) or loss of T12 (right) are shown with arrows. (G) Bar plot demonstrating the percentage of pups (postnatal day 0.5) with the homeotic transformation phenotype in Patz1 KO compared to WT. Raw numbers of mice are shown in orange (see Figure S11 for genetic deletions).

Figure 7.. Combinatorial binding of insulation factors at boundaries are critical to the regulation of cellular identity

(A-B) Models depicting (A) the anterior-posterior MN identity in the WT setting in the presence of different accessory/insulation factors and (B) the transition of anterior-posterior MN identities impacted upon the loss of different accessory/insulation factors. The illustrations on the left side of each panel demonstrate the regulation of gene expression through the combinatorial binding of accessory factors in addition to cohesin and CTCF at loop anchors. The right side of each panel shows MN identity determination/transition based on Hox gene expression in the absence of various accessory/insulation factors, as compared to WT MNs. (C) Model depicting the homeotic transformations in the skeletal structure of mice. Cervical, thoracic, lumbar, sacral, and caudal vertebra are shown in indicated colors. Stars mark the vertebrae impacted at cervicothoracic and thoracolumbar boundaries in CTCF/MAZ Hox5|6 binding deficient mice and Patz1 KO mice, respectively.