Gsx2 differentially regulates transcription via independent monomer and cooperative dimer-binding sites

To identify specific Gsx2-binding sites within the 687 enhancer, we used purified mouse Gsx2 protein and DNA probes containing predicted Gsx2 sites in electrophoretic mobility shift assays (EMSAs) (Fig. 2A; Supplemental Fig. S1A). Interestingly, we observed two modes of binding depending on the probe sequence. Consistent with prior studies showing that Gsx factors bind individual AT-rich DNA sequences, the majority of probes were bound in a manner consistent with monomeric Gsx2 binding (Fig. 2B). Moreover, even if a probe contained more than one site, these probes were bound by Gsx2 in an additive manner (Supplemental Fig. S1B–F). In contrast, one probe that contained two predicted sites was unexpectedly preferentially bound by two Gsx2 proteins in a manner consistent with cooperative Gsx2 homodimer formation (Fig. 2C; Supplemental Fig. S1G). Thus, the in vitro DNA-binding assays revealed that Gsx2 interacts with DNA as either a monomer (on M sites) or as a homodimer (on D sites). Considering these distinct binding modalities, the 687 enhancer contains multiple monomer sites (M1–M9) and one dimer site (D1) (Fig. 2A).

Previous studies in Xenopus and Drosophila revealed that Gsx factors function as transcriptional repressors (Von Ohlen et al. 2007a, 2009; Von Ohlen and Moses 2009; Winterbottom et al. 2010, 2011). Thus, we developed a reporter assay to investigate the regulatory potential of mouse Gsx2 on M and D sites by creating luciferase vectors with a minimal promoter, 5xUAS sites, and either six copies of the M1 site or three copies of the D1 site from the 687 enhancer (Fig. 2D–E). In this assay, transfection with Gal4-VP16 activated luciferase expression in a mouse kidney cell line (mK4 cells) (green bar in Fig. 2D), whereas cotransfection with Gsx2 resulted in dose-dependent repression of Gal4-mediated activation on the 6xM1 reporter (orange bars in Fig. 2D). Surprisingly, the D1 site reporter behaved differently in this assay. Instead of repression, we observed enhanced Gal4-VP16 mediated luciferase activity in response to Gsx2 (Fig. 2E). However, if the D1 site was mutated so that it binds Gsx2 in a noncooperative, monomer-like manner (Fig. 2F, the D1-to-M mutation), Gsx2 mediated dose-dependent repression (Fig. 2G). These data suggest that Gsx2 differentially mediates transcription when bound as a monomer versus a homodimer. Interestingly, Gsx2 did not activate gene expression via D sites in the absence of Gal4-VP16 (Fig. 2E), suggesting that it is insufficient to activate transcription by itself, at least in mK4 cells. Moreover, the greatest stimulation of Gal4-VP16 mediated activation of the D1 reporter occurs at low Gsx2 levels, whereas increasing Gsx2 concentrations reduced luciferase activity (Fig. 2E). Analysis of the minimal promoter (minP) sequence, which is present in each luciferase vector, revealed a predicted M site (vertical red line in Fig. 2D,E,G) that may contribute to decreased reporter activity at higher Gsx2 levels (note, the impact of combining D and M sites on transcriptional output is tested later in this study). Collectively, these data support a novel model of HD-mediated gene regulation, whereby mouse Gsx2 represses transcription when bound to M sites and stimulates transcription via D sites.

Cooperative Gsx2 dimer binding to DNA requires amino acids flanking the homeodomain and precisely spaced and oriented binding sites

As Gsx factors had not previously been shown to cooperatively bind DNA, we next defined the critical domains required for cooperative Gsx2 DNA binding using deletion constructs and EMSAs to calculate their Hill binding coefficient as a measure of cooperativity (Hill coefficient of 2 = perfect cooperativity; Hill of 1 = no cooperativity) (Weiss 1997). Strikingly, we found that the mouse Gsx2 HD alone does not preferentially form dimers on a DNA probe encoding a D site, and instead binds this probe in a noncooperative monomeric manner (Fig. 3A, right). In contrast, this same DNA probe preferentially binds two Gsx2 proteins that contain short 40-amino-acid regions N- and C-terminal of the HD, consistent with the formation of cooperative Gsx2 homodimers on DNA (Fig. 3A, left). We used this data to generate a Hill plot and found that the larger Gsx2 fragment has a Hill coefficient of ~1.8, indicating significant cooperative binding, whereas the HD-only fragment has a Hill coefficient much closer to 1 (1.17) (Fig. 3B). These findings reveal that while the Gsx2 HD is sufficient to independently bind the sites that comprise the D site, it is not sufficient to mediate cooperative binding to the D site.

Open in a separate window

Figure 3.

Gsx2 requires domains flanking the homeodomain and precisely spaced and oriented DNA sites to cooperatively bind D sites. (A) EMSAs using the 687-D1 probe and purified Gsx2 proteins containing the flanking regions plus homeodomain (167–305, left) or only the HD (light blue, right). Each EMSA binding reaction had 34 nM labeled DNA probe with the following concentrations of either the Gsx2 (167–305) or Gsx2 HD protein: 0 nM (i.e., probe alone), and 2.34, 4.69, 9.38, 18.75, 37.5, 75, and 150 nM. Note, the purity of the Gsx2 (167–305) and Gsx2-HD proteins are shown by SDS-PAGE analysis and gel staining in Supplemental Figure S2A. Schematics highlight the fact that the Gsx2 + flanking region protein more readily forms dimer complexes than the HD only protein. (B) Hill coefficient calculations from EMSAs reveal that the Gsx2+flanking protein binds D sites cooperatively (Hill coefficient = 1.84), whereas the HD alone does not (Hill coefficient = 1.17). (C) Summary of scanning mutagenesis and Gsx2 EMSA data highlighting the nucleotides in the 687 D1 probe required for cooperative (C) versus noncooperative (N) binding. EMSAs are shown in Supplemental Figure S2B–M. (D) Summary of Gsx2 EMSA data using probes with insertion of different numbers of nucleotides reveals that a 7-bp spacer is required for cooperative binding to the D1 probe. Note, the +3-bp insertion generated a new “D site” with the required 7-bp spacing. EMSAs are shown in Supplemental Figure S2N–R. (E) Summary of Gsx2 EMSA data using probes engineered with a nonpalindromic D site in different orientations and spacing. Note, only the F3F probe contains the required orientation and spacing to mediate cooperative binding. EMSAs are shown in Supplemental Figure S2S–BB. (F) Optimal Gsx2 D site with a 7-bp spacer as defined by EMSAs.

We next sought to define the DNA sequence constraints that facilitate cooperative binding to D sites. Using the 687 D1 site for this analysis, we first tested a series of mutations across the D1 site (mut1 – mut10) in EMSAs and categorized each as either cooperative (C) or noncooperative (N) (Fig. 3C; Supplemental Fig. S2B–M). When sequences were aligned, the importance of 5 key nucleotides, a TA and an ATT, separated by 7 bp was evident (Fig. 3C). To test whether this spacing is required, we designed probes that increased spacing in single-base-pair increments and discovered that adding just 1 bp disrupts cooperative binding (Fig. 3D; Supplemental Fig. S2N–R). Interestingly, adding 3 bp between sites restored cooperative binding; however, this insertion shifted a different TA into a 7-bp spacing with the ATT sequence from the original site and thereby created a new cooperative D site (Fig. 3D; Supplemental Fig. S2Q). Determining the orientation of the two sites that make up a D site was complicated by the fact that the previously defined optimal Gsx2 motif is palindromic (TAATTA) (Jolma et al. 2013). To circumvent this issue, we analyzed existing protein binding microarray (PBM) data for Gsx2 and selected the highest scoring nonpalindromic 8-mer sequence (Berger et al. 2008). We then designed and tested a series of probes containing this 8-mer in different orientations and found that only one spacing and orientation combination, the forward-forward probe with 7-bp spacing, was bound cooperatively by Gsx2 (Fig. 3E; Supplemental Fig. S2S–BB). In total, by coupling EMSA data with Gsx2's known monomer binding preferences, we predict that an optimal Gsx2 D site contains two ATTA sequences separated by 7-bp (Fig. 3F).

To determine the dimer binding preferences of Gsx2 to DNA in an unbiased manner, we analyzed published HT-SELEX data that was generated using random 20-mers (Jolma et al. 2013). Importantly, while the original study only identified an enriched M motif, the SELEX assay was performed using a human GSX2 protein containing similar flanking N- and C-terminal domains that we found were required for cooperative binding by the mouse Gsx2 protein (see Fig. 3A,B). Hence, we reanalyzed the HT-SELEX data for occurrences of the optimal D site motif (Fig. 3F) after each of 4 selection cycles and observed clear enrichment of D site sequences (Fig. 4A). To control for simply enriching sequences with two M sites, we calculated the percentage of sequences with two sites in relation to spacer length from 3 to 11 bp and found a striking preference for sites separated by 7 bp (Fig. 4B), which is identical to our empirical studies (Fig. 3D; Supplemental Fig. S2). Moreover, we performed a de novo motif search for longer motifs using MEME and found that the top enriched motif contained two HD sites with a 7-bp spacer (Fig. 4C). Thus, human GSX2 enriches for D sites with the same spacing and orientation that were cooperatively bound by the mouse Gsx2 protein.

Open in a separate window

Figure 4.

Analysis of HT-SELEX data reveals that the human GSX2 protein enriches for dimer DNA binding sites. (A) The percentage of sequences that contain an optimal D site motif as a function of SELEX cycle. Note, the “0” cycle is the starting library and enrichment for D sites is observed in successive SELEX cycles using the human GSX2 protein. (B) The percentage of sequences with two sites by spacer length. Note, the highest frequency occurs with the 7-bp spacer. (C) The D site PWM motif from MEME de novo motif search on the human GSX2 HT-SELEX data after the fourth round of selection. (D) ChIP-seq for P300 and H3K27ac in E12.5 forebrain tissue revealed strong signals at the characterized Gsx2 DSG enhancer. The location of DSG is boxed and the sequence of an optimal D site (blue) as well as predicted M sites (red) within the DSG are noted. Sequence of the DSG is reported in Supplemental Document 1. (E) EMSA using Gsx2 (167-305) reveals cooperative dimer binding to the DSG D site. Each EMSA binding reaction had 34 nM of labeled DNA probe and the following concentrations of the Gsx2 (167–305) protein: 0 nM (i.e., probe alone), or 46.5, 93, 186, or 372 nM. Note, a larger image of this exact same gel is shown in Supplemental Figure S2CC. (F) UAS-3xDSGD-Luciferase activity revealed enhanced Gal4VP16 activation in the presence of Gsx2. The amounts of transfected plasmid are noted. (*) P < 0.05 using an unpaired two-tailed Student's t-test compared with Gal4VP16 alone.

To test whether the binding site spacing and orientation rules could predict sites that would be both cooperatively bound and stimulated by Gsx2, we analyzed another recently characterized Gsx2 enhancer, which we have called DSG (downstream from Gsx2) (Fig. 4D; Desmaris et al. 2018; Konno et al. 2019). We scanned the ~1200-bp region for putative D sites and identified one that perfectly matches the optimal D motif as well as eight predicted high-affinity M sites (Fig. 4D). We subsequently used comparative EMSAs to show that mouse Gsx2 cooperatively bound the DSG D site probe (Fig. 4E), and luciferase assays further revealed that Gsx2 strongly enhanced Gal4-VP16-mediated gene activation via the DSG-D site (Fig. 4F). Thus, these data show that two different cooperative D sites (the 687 D1 and DSG-D sites) both stimulate Gal4-VP16 mediated gene expression in the luciferase assay, whereas two different M sites (the 687 M1 and the D1-to-M sites) both mediate gene repression (Fig. 2D,G).

Genome-wide analysis in the mouse LGE reveals Gsx2 binding to monomer and dimer sites

To identify the in vivo genomic targets for Gsx2 in the developing LGE, we performed ChIP-seq from chromatin isolated from mouse E12.5 LGEs using a Gsx2 antibody. While this antibody worked for ChIP-qPCR (see Fig. 1E), ChIP library sequencing revealed a low signal to noise ratio compared with IgG controls. Thus, we used genome editing to insert an N-terminal 2x-FLAG tag in-frame with the Gsx2 endogenous locus (Fig. 5A). Importantly, homozygous Gsx2^2xFLAG mice are viable, breed normally, and are grossly normal compared with littermate controls. Furthermore, FLAG expression overlaps Gsx2 staining in the expected pattern (Fig. 5B), and embryonic ventral forebrain morphology is normal with no noticeable ventral expansion of dorsal telencephalic markers such as Pax6 (Fig. 5C), as would be expected in Gsx2-null mutant embryos (Corbin et al. 2000; Toresson et al. 2000; Yun et al. 2001).

Open in a separate window

Figure 5.

Genomic analysis of Gsx2 binding in the mouse LGE reveals enrichment of monomer and dimer sites. (A) Schematic of the 2XFLAG-Gsx2 with the homeodomain (HD) highlighted in light blue. (B) Immunostaining of an E12.5 Gsx2^{2XFLAG/2XFLAG} mouse telencephalon reveals extensive colocalization of Gsx2 (green) and FLAG (red) in the expected LGE expression pattern. (C) Gsx2/Pax6 double staining shows that Dorsal-Ventral patterning in the E12.5 Gsx2^{2XFLAG/2XFLAG} telencephalon appears normal. (D) Replicate CUT&RUN analysis of FLAG-Gsx2 genomic binding to the Gsx2 locus in comparison with IgG controls. Note, significant FLAG-Gsx2 binding to both the 687 and DSG enhancers. (E) Genomic annotation of Gsx2 peaks using HOMER. Note, most peaks are found in intergenic and intronic regions. (F) Top motifs identified by HOMER reveal significant M and D site enrichment. (G) The number of occurrences of two ATTA sites by spacer length in Gsx2 CUT&RUN peaks. Note, like with the SELEX assay (Fig. 4B), the strongest peak occurs with the 7-bp spacer. (H) Alignment of the top 2126 sites containing Gsx2 D sites identified by MEME and secondary filtering for footprint contrast. Sequences are color-coded and outside of the ATTA motifs, there is limited sequence similarity between sites. (I) MNase digestion footprint of the genomic Gsx2 D sites. All sequences were aligned centered on the D motif with the most highly protected sites overlapping the D motif. (J) CUT&RUN signal at Gsx2 peaks either containing at least one D site (1591 peaks, Red) or lacking a D site (1441 peaks, blue). (K) Normalized tag counts (RPM) within a 300-bp window around Gsx2 peaks containing at least one D site versus all other peaks. (*) Wilcoxon test. (L) Alignment of the top 5591 sites containing Gsx2 M sites that do not overlap with a D site. (M) MNase digestion footprint of the genomic Gsx2 M sites. All sequences were aligned centered on the M motif.

We next performed CUT&RUN assays (Skene and Henikoff 2017; Skene et al. 2018) on LGE tissue from homozygous E12.5 Gsx2^2xFLAG embryos using the anti-FLAG-M2 antibody and an IgG control. Analysis of FLAG CUT&RUN biological replicates revealed a highly reproducible peak pattern (N = 3032) against matching IgG CUT&RUN control samples (Supplemental Fig. S3A). Notably, specific peaks in the 687 and DSG enhancers were observed, confirming that Gsx2 directly binds these enhancer regions (Fig. 5D). Gsx2 peaks also overlapped with the 678 enhancer (Fig. 5D), which is insufficient to activate LGE gene expression on its own, but results in robust LGE expression when coupled with the 687 enhancer in a transgenic reporter assay (Qin et al. 2016). Thus, Gsx2 binds multiple regulatory elements near the Gsx2 locus, consistent with its direct role in autoregulation (see Fig. 1A–C).

Genomic annotation of FLAG CUT&RUN peaks (Gsx2-binding sites) revealed that the majority fall within intergenic or intronic regions with only a small subset associated with promoters (Fig. 5E). Consistent with many Gsx2 binding events being associated with regulatory elements, we found a high association with E12.5 forebrain marks of active enhancers such as p300 binding (Zhou et al. 2017), H3K27ac chromatin marks (ENCODE project ENCSR966AIB), and transposase accessible regions (i.e., open chromatin; ENCODE project ENCSR559FAJ) (Supplemental Fig. S4). Moreover, analysis with either HOMER's or MEME's de novo motif enrichment tools revealed significant enrichment of both M and D motifs (Fig. 5F; Supplemental Fig. S3B). Further analysis of called Gsx2 peaks for ATTA sequences with different spacer lengths revealed that similar to the HT-SELEX data (Fig. 4E), the Gsx2 CUT&RUN data showed a strong preference for dimer sites with a 7-bp spacer (Fig. 5G).

Based on the idea that bound TF sites are protected from MNase digestion during CUT&RUN analysis (Skene and Henikoff 2017), we performed footprint analysis to identify Gsx2 binding sites in high resolution. First, we scanned 300-bp windows of the 3032 called peaks for D and M sites using FIMO to obtain candidate binding sites (Grant et al. 2011). We then measured MNase digestion signals across each loci and defined loci displaying lower signal than flanking windows as true Gsx2-binding sites (see Materials and Methods for details). Through this analysis, we found that approximately half of the called Gsx2 peaks contained one or more D sites (1591 peaks) for a total of 2126 D sites (Fig. 5H,I). Moreover, we found that the CUT&RUN signal at the 1591 peaks with at least one D motif was significantly higher compared with the 1441 peaks lacking a D motif (Fig. 5J,K), consistent with D sites promoting robust Gsx2 DNA binding. Importantly, high-resolution Gsx2 binding analysis also identified 5591 Gsx2 binding events to M sites that did not overlap with the predicted D motifs, and as expected these binding events revealed a smaller protective footprint over a single TAATTA M motif (Fig. 5L,M). Altogether, these data demonstrate that mouse Gsx2 binds to both D and M sites in LGE progenitors.

To determine how Gsx2 binding correlates with changes in gene expression, we performed differential RNA-seq analysis on wild-type and Gsx2-null (Gsx2^EGFP/RA) E12.5 mouse LGEs and found 719 up-regulated and 289 down-regulated genes in Gsx2 mutants (fold change >1.5; FDR <0.05) (Fig. 6A). As expected, gene ontology (GO) analysis revealed a significant change in the expression of genes associated with many aspects of neural development (Fig. 6B,C). For example, in addition to Gsx2 itself (which is negatively autoregulated) (see Fig. 1A–C), a group of TFs known to regulate forebrain development were differentially expressed between WT and Gsx2 mutant animals, including Ascl1, Dlx1, Dlx2, Dlx5, Sp8, Dbx1, Gsx1, and Pax6 (Fig. 6A; Corbin et al. 2000; Toresson et al. 2000; Toresson and Campbell 2001; Yun et al. 2001, 2003; Waclaw et al. 2006, 2009). Consistent with potential direct regulation of these target genes by Gsx2, analysis of Gsx2 binding events revealed peaks near each of Ascl1, Dlx1/2, Gsx1 and Pax6 (Supplemental Fig. S5).

Open in a separate window

Figure 6.

Peaks containing D and M sites are associated with gene expression changes in the LGE of Gsx2-null embryos. (A) RNA-seq analysis from LGE tissue of E12.5 WT (Gsx2^+/+) and Gsx2-null (Gsx2^EGFP/RA) embryos reveals genes that are significantly up-regulated (red) and down-regulated (blue) in the absence of Gsx2. Significantly altered expression of transcription factors important for forebrain patterning and neurogenesis are labeled. Differentially expressed genes were defined by fold change >1.5 and FDR < 0.05. (B,C) GO analysis on genes upregulated (B) and downregulated (C) in Gsx2-null embryos reveals Biological Process GO terms related to neural development. (D,E) Genes were divided into down-regulated (blue), up-regulated (red), or unchanged (gray) groups. The cumulative percentage of genes in each group that had a Gsx2 CUT&RUN peak with an M and D site (D) or with no M or D site (E) within a certain distance up to 100 kb from their TSS is plotted. Note, only those peaks with an M and D site are significantly associated with up-regulated and down-regulated genes in Gsx2-null LGEs, whereas those lacking M and D sites are not associated with gene expression changes of nearby genes. (*) P-value < 0.05. Similar analysis for Gsx2 CUT&RUN peaks selected only on the basis of having a D site (F) or a M site (G) is shown in Supplemental Figure S6. (F) Comparative PWM logos generated from previously published HT-SELEX assays (Jolma et al. 2013) reveals nearly identical GSX2 monomer (top) and DLX1 (bottom) DNA-binding sites, consistent with these TFs binding largely the same DNA sites. (G) Comparative genomic binding analysis of the Gsx2 CUT&RUN data from E12.5 mouse LGEs and the previously published Dlx (Lindtner et al. 2019) ChIP-seq data from E11.5 (top) and E13.5 (bottom) mouse forebrains reveals significant overlap in genomic binding between Gsx2 and the Dlx factors. (H) Analysis of Gsx2 and Dlx binding to the same genomic regions associated within a 100 kb window around gene TSSs that are either down-regulated (blue bars), unchanged (gray bars), or up-regulated (red bars) in Gsx2 mutant LGEs. The percentage of Gsx2 genomic binding events that were also bound by at least two Dlx factors (see the Materials and Methods) for each group of genes is calculated using the published E11.5 (left) and E13.5 (right) ChIP-seq data for Dlx1, Dlx2, and Dlx5 (Lindtner et al. 2019). Note, those genes that are down-regulated in Gsx2 mutant LGEs are significantly enriched for nearby genomic regions that bind both Dlx and Gsx2 factors compared with the unchanged group (P-value by Fisher's exact test). In contrast, the up-regulated gene group is not significantly different from the unchanged gene group. Thus, a substantial portion of genes down-regulated in Gsx2 mutant animals is likely due to the indirect loss of Dlx transcription factor expression.

We next intersected the CUT&RUN and RNA-seq data by comparing the spatial proximity of Gsx2 peaks with distinct groups of genes that were either significantly up-regulated, down-regulated, or unchanged in Gsx2^EGFP/RA LGEs. Moreover, we divided the Gsx2-binding events into four groups: Gsx2 peaks containing at least one M and D site (Fig. 6D), Gsx2 peaks containing neither an M nor a D site (Fig. 6E), Gsx2 peaks containing at least one D site (Supplemental Fig. S6A), and Gsx2 peaks containing at least one M site (note only the first two groups are mutually exclusive) (Supplemental Fig. S6B). Importantly, while Gsx2 peaks containing D and/or M sites were all strongly associated with significant gene expression changes in Gsx2 mutant animals (Fig. 6D; Supplemental Fig. S6), those binding events that lack an M or a D site were not positively correlated with significant up-regulation or down-regulation of nearby genes in Gsx2 mutants (Fig. 6E). These data suggest that Gsx2 binding to M and/or D sites significantly contributes to the regulation of LGE gene expression. However, there was no obvious correlation between the direction of gene expression changes (up vs. down) and Gsx2 binding to D site versus M site containing Gsx2 peaks. These findings suggest that the simple presence of an M or D site within a Gsx2 peak is not sufficient to predict the direction of in vivo transcriptional output within the LGE.

Our inability to predict gene expression changes in Gsx2 mutant LGEs based solely on M versus D site binding is not unexpected given that Gsx2 mutants misregulate numerous other TFs (e.g., Ascl1, the Dlx factors, etc) that also play significant roles in LGE gene expression (Corbin et al. 2000; Toresson et al. 2000; Yun et al. 2001). Among these genes, Dlx1, Dlx2, and Dlx5 encode homeodomain TFs that bind highly similar, if not identical, DNA sequences as the Gsx2 monomer site in published HT-SELEX assays (Fig. 6F). To determine whether the Dlx factors bind the same genomic regions as Gsx2, we analyzed recently published ChIP-seq data for Dlx1, Dlx2, and Dlx5 from the E11.5 and E13.5 mouse forebrain (Lindtner et al. 2019). Since Lindtner et al. found that the intersection of Dlx1, Dlx2 and/or Dlx5 genomic binding provided a high-confidence dataset, we intersected our E12.5 Gsx2 CUT&RUN data with the combined Dlx genomic binding regions at E11.5 and E13.5. Importantly, we found that over 30% of the Gsx2 bound genomic regions are also bound by two or more Dlx TFs (Fig. 6G). Next, we asked whether those genomic regions that bind both Gsx2 and Dlx TFs within a 100 kb window of a gene's transcription start site are more highly associated with genes that are either down- or up-regulated in Gsx2 mutant LGEs. Intriguingly, this analysis revealed that down-regulated genes in Gsx2 mutant LGEs were significantly more likely to be bound by both Gsx2/Dlx factors than genes that are up-regulated in Gsx2 mutant LGEs (Fig. 6H). These findings are consistent with the idea that Dlx factors, which are known to function as activators (Stuhmer et al. 2002; Le et al. 2017), and Gsx2 regulate many of the same target genes through the same enhancer elements. Thus, these results reveal how the misregulation of other TFs in Gsx2 mutants complicates our ability to predict Gsx2-dependent output in the mouse LGE based on the simple presence of M versus D sites. Nevertheless, our CUT&RUN, RNA-seq, and bioinformatics data do clearly reveal that mouse Gsx2 binds to both M and D sites in vivo and that Gsx2 binding to regulatory elements with such sites significantly influences LGE gene expression.

The cis-regulatory logic of Gsx-mediated gene regulation

Gsx factors regulate D-V patterning in the fly and mouse nervous system by repressing the molecular identity of cells in the adjacent developmental compartment (Weiss et al. 1998; Corbin et al. 2000; Toresson et al. 2000; Yun et al. 2001). In fact, studies in flies and frogs have led to the notion that Gsx factors function primarily as repressors (Von Ohlen et al. 2007a; Von Ohlen and Moses 2009; Winterbottom et al. 2010, 2011). However, Ind can also positively regulate its own expression in embryonic Drosophila neuroblasts (Von Ohlen et al. 2007b), although the mechanisms determining positive versus negative output were unknown. Our findings that Gsx factors mediate opposing transcriptional outcomes via M versus D sites reveals an unanticipated mechanism underlying how these factors control gene expression during neural development. By focusing on binding sites within enhancers near Gsx2 in mice and Ind in Drosophila, we found that autoregulation is a common feature of Gsx factors, and their impact on autoregulation is complex. For instance, consistent with direct Ind-mediated positive autoregulation in Drosophila neuroblasts, we found that Ind cooperatively binds D sites from its autoregulatory enhancer and mutating these D sites dramatically decreased enhancer activity in vivo. However, we also found that Ind binds multiple M sites from the same autoregulatory enhancer and mutating these M sites results in a significant increase in enhancer activity in intermediate column neuroblast cells. These findings support the model that Ind autoregulation is both positive via cooperative D sites and negative via independent M sites.

Consistent with these results, we found that the D-to-M ratio within Gsx-regulated enhancers can have a profound impact on regulatory outcomes in mammalian cells. For instance, while both low and high Gsx2 levels repressed reporters with only M sites, reporters with enhancers containing a mixture of D + M sites revealed a more complex, nonmonotonous pattern of enhancer activity as a function of Gsx2 protein levels. At relatively low levels, Gsx2 stimulated the D + M enhancer activity, consistent with Gsx2 cooperativity resulting in preferential binding to D sites (Fig. 8E). However, as Gsx2 levels were increased, the D + M reporter activity decreased, consistent with higher Gsx2 levels resulting in repression via M sites (Fig. 8F,G). The net impact of this mechanism on autoregulation is that Gsx2/Ind transcription factors would maintain their own expression levels in a narrow window based on the ratio of stimulatory D sites to repressive M sites in each respective enhancer.

We used the CUT&RUN assay to show that Gsx2 binds to a substantial number of potential regulatory elements that contain M and D sites within the developing mouse forebrain. Consistent with these Gsx2 CUT&RUN peaks being in functional enhancers, a high percentage of the binding events mapped to genomic regions with active chromatin marks in E12.5 whole telencephalon samples. Moreover, transcriptomic studies revealed that Gsx2 binding events to genomic regions with D sites, M sites, or D + M sites were each significantly associated with nearby genes that exhibited altered expression in Gsx2 mutants. However, there was no direct correlation between the mere presence of a D site and gene up-regulation or an M site and gene down-regulation. In hindsight, our inability to predict gene expression changes based on only D versus M sites within each individual Gsx2 peak is not unexpected for several reasons. First, most Gsx2 bound enhancers contain a mixture of D and M sites, and as we found by studying the Gsx2 and Ind autoregulatory elements, the behavior of D + M elements is complex due to the opposing activities mediated by Gsx binding to each type of site. Second, transcriptomics data revealed the expression of a number of key LGE TFs are significantly changed in Gsx2 mutants. Hence, the germline removal of Gsx2 is likely to have numerous indirect impacts on gene regulation in the E12.5 mouse LGE. For example, analysis of existing ChIP-seq data for the misregulated Dlx TFs, which bind highly similar if not identical DNA sequences as Gsx2, revealed significant overlap with the genomic Gsx2 binding profile (Lindtner et al. 2019). Moreover, we found that only down-regulated genes in Gsx2 mutant LGEs are preferentially enriched for genomic regions bound by both Gsx2 and Dlx factors. Hence, genes that are significantly down-regulated in Gsx2 mutants could be indirectly decreased due to the loss of the Dlx TFs, which are known to function as transcriptional activators (Stuhmer et al. 2002; Le et al. 2017). This result is not surprising because the LGE contains many more Dlx-expressing cells than Gsx2-expressing cells; thus, altered gene expression in the Gsx2 mutant LGE is likely to be heavily influenced by the loss of such downstream effectors. Overall, these data highlight how the complexity of cis-regulatory elements is likely to make it very difficult to predict their transcriptional behaviors based solely upon Gsx2 D and M site binding information using bulk transcriptomic data from Gsx2 germ line mutations. Thus, future studies focused on shorter temporal windows following Gsx2 removal or addition are needed to ascertain how Gsx2 M versus D site binding directly impacts gene regulatory outcomes in the mouse LGE.

An additional unanswered question raised by our findings is how do the Gsx/Ind transcription factors mediate opposing outcomes when bound to DNA as monomers versus dimers? Prior studies found that the Gsx/Ind factors bind the Groucho corepressor protein through a conserved N-terminal domain (Von Ohlen et al. 2007a, 2009; Von Ohlen and Moses 2009; Winterbottom et al. 2010, 2011). However, Ind has been shown to have two additional regulatory activities: a second repression domain and an undefined activation domain (Von Ohlen and Moses 2009). Unfortunately, we do not know the identity of such additional corepressors/coactivators that interact with the Ind/Gsx factors, and therefore we lack an understanding of how dimer formation on DNA alters the ability of Gsx/Ind factors to recruit specific cofactors. Moreover, it should be noted that Gsx2 alone is insufficient to activate gene expression, at least in mK4 cells, but is able to stimulate gene activation in conjunction with the Gal4-VP16 protein. How the Gsx/Ind proteins synergize with other nearby transcriptional regulators to stimulate gene expression is currently unknown and an important issue to be addressed in future studies.

Mouse lines

Genotyping of the Gsx2^EGFP and Gsx2^RA mouse lines have been described previously (Waclaw et al. 2009; Wang et al. 2009). The Gsx2^2xFLAG allele was generated via CRISPR-Cas9 using a gRNA to the 5′ UTR of Gsx2. A ssDNA donor oligonucleotide, which contained cross-species nucleotide mutations 5′ to the PAM sequence to prevent retargeting by Cas9, was used to insert the 2xFLAG tag and GSG-linker coding sequence in-frame with Gsx2 (see Supplemental Document 1). To confirm 2xFLAG insertion, we used two oligonucleotide primers (5′-Primer_for_2xFLAG_genotyping and 3′-Primer_for_2xFLAG_genotyping) (see Supplemental Document 1) to identify the Gsx2^2XFLAG allele. Two founders were positive for the 2xFLAG insertion and each line was bred to homozygosity. The locus surrounding the transcription start site (TSS) and coding region of Gsx2 was amplified and sequenced to confirm that the 2XFLAG tag was properly inserted in frame and the modified Gsx2 reading frame was intact (see Supplemental Document 1). To confirm homozygous Gsx2^2XFLAG animals were “phenotypically normal,” we collected E12.5 embryos from a homozygous cross, fixed each in 4% PFA for 6 h and processed for histology as described (Wang et al. 2009). For immunohistochemistry, the following primary antibodies were used; mouse anti-FLAG (1:500; Sigma), rabbit anti-Gsx2 (1:2000) (Toresson et al. 2000) and mouse anti-Pax6 (1: 1000; Invitrogen). Fluorescent secondary antibody detection was carried out using donkey anti-rabbit IgG conjugated to Alexa 647 (1:200; Jackson Immunoreseach) and goat anti-mouse IgG₁ conjugated to Alexa 568 (1:500; Invitrogen). Sections were imaged for confocal microscopy using a Nikon A1 LSM system with a GsAsP solid-state laser.

Drosophila transgenic reporter assays

The 1.6-kb Ind autoregulatory element (Von Ohlen et al. 2007b) and proximal 980 bp of that element relative to the ind TSS were PCR-amplified from an Ind-BAC using two primers (Ind_autoregulatory_element_Forward and either Ind_1.6 kb_autoregulatory_element_Reverse or Ind_980 bp_autoregulatory_element_Reverse). Fragments were cloned into pattBLacZ (Bischof et al. 2007) using XbaI and XhoI sites. The Mmut and Dmut constructs were synthesized by GenScript and cloned using the same enzyme sites into pattBLacZ. The WT 980 bp element, Mmut, and Dmut element sequences can be found in Supplemental Document 1. To create transgenic flies, each construct was injected into Drosophila embryos by Rainbow Transgenics and the -C31 integrase system was used to insert each into the 51C locus (Bischof et al. 2007). Embryos were harvested from fly lines homozygous for each lacZ transgene for 2 h at 25°C, aged an additional 3 h at 25°C before being collected, fixed and immunostained using standard procedures (Zandvakili et al. 2018, 2019). Chicken anti-β-gal (1:1000; Abcam) and rabbit anti-Ind (1:1000) (Von Ohlen and Moses 2009) primary antibodies and AlexaFlour fluorescent secondary antibodies were used to immunostain each sample. All transgenic embryos were imaged under identical subsaturating conditions using a Zeiss Axio Imager upright microscope with an Apotome filter for optical sectioning and an AxioCam MRm digital camera. Pixel intensities for β-gal and Ind were quantified using Fiji, and the average intensity of β-gal levels in Ind-positive regions were determined for each embryo and plotted as a single data point in the box and whisker plot in Figure 7. An unpaired two-tailed students T-test was performed to compare pixel intensities from each mutant to wild-type reporter.

Protein purification

Coding DNA for all protein constructs used in EMSAs were PCR-amplified, cloned in-frame with an N-terminal 6x-His tag into the pET-14b bacterial expression vector (Novagen), and sequence confirmed. All Gsx2 constructs were from Mus musculus. All Ind constructs were from Drosophila melanogaster. The Gsx2 HD construct (amino acids 203–264) was cloned using the Gsx2_HD_F and Gsx2_HD_R primers between NdeI and XhoI sites. The Gsx2 construct containing the HD and flanking regions (amino acids 167-305) was cloned using Gsx2_167-305_F and Gsx2_167-305_R primers into NdeI and XhoI sites. The Gsx2 full-length protein (amino acids 1-305) was cloned using the Gsx2_FL_F and Gsx2_FL_R primers between NdeI and XhoI sites. The Ind construct containing the HD and flanking regions (amino acids 175–320) was cloned using the Ind_175-320_F and Ind_175-320_R primers between NdeI and KpnI sites. Full-length Ind containing amino acids 1-320 was cloned using the Ind_FL_F and Ind_FL_R primers between NdeI and KpnI sites. Full-length Gsx2 and Ind proteins were purified via Ni-chromatography under denaturing conditions and allowed to refold while still bound to Ni-beads as previously described (Zhang et al. 2019). Gsx2 and Ind subfragments were purified via Ni-chromatography under native conditions as described previously (Uhl et al. 2010). Full-length proteins were confirmed via Western blot, whereas the purity of protein subfragments was confirmed via SDS-PAGE analysis and GelCode blue staining (Thermo Scientific). The indicated protein concentrations in the Figure legends were determined via Bradford assays.

Electrophoretic mobility shift assays (EMSA)

EMSAs were performed as described previously (Uhl et al. 2016; Kuang et al. 2020). Probe sequences are listed in FASTA format in Supplemental Document 2. Binding reactions containing the indicated DNA probes and protein concentrations listed in each Figure legend were mixed and incubated in the dark at room temperature for 10 min prior to gel electrophoresis. EMSAs were imaged via a Li-Cor Odyssey CLx scanner, and quantified using the Li-Cor image studio software as described previously (Roychoudhury et al. 2020).

Luciferase assays

Sequences for all reporters can be found in Supplemental Document 1. 5xUAS sites (highlighted in green), Gsx2-binding sites (highlighted red for M sites and blue for D sites), and a minimal promoter (gray) were synthesized by GenScript and subcloned into pGL3-basic luciferase (Promega) between KpnI and NcoI sites. The Gsx2 pCDNA6 construct has been previously described (Roychoudhury et al. 2020). The Gal4-VP16 coding DNA was cloned into pCDNA6 between KpnI and XbaI sites. Luciferase assays were performed in triplicate using the mouse mK4 cell line as previously described (Roychoudhury et al. 2020). Each well was transfected with 5 ng of Renilla-luciferase, 25 ng of the indicated firefly luciferase reporter, 5 ng of Gal4-VP16 in pCDNA6 if indicated, and a titration of the indicated amounts of Gsx2 in pCDNA6. To ensure each well was transfected with the same amount of DNA, empty pCDNA6 vector was used to fill to 85 ng. Forty-eight hours after transfection, cells were harvested, lysed, and analyzed for luciferase activity using the Promega dual-luciferase assay kit and GloMax luminometer. All firefly luciferase values were normalized to Renilla-luciferase to control for transfection efficiency. Relative luciferase values presented in each bar graph represent the mean ± standard deviation with Gal4-VP16 alone condition set to 100.

ChIP-qPCR

LGE tissue was dissected from 35 E12.5 mouse forebrains, and ChIP assays were performed as described previously (Castro et al. 2011) with the following changes: Sonication was performed at 4°C using a Fisher Scientific Sonic Dismembrator model 100 at setting 5 for 25 cycles (10 sec on/30 sec off). Two percent of chromatin was set aside as input, and the remaining sample was split in half for immunoprecipitation with either a rabbit anti-Gsx2 antibody at a 1:100 dilution or 1 µg of control rabbit IgG. Quantitative PCR was performed using a QuantStudio 3 real-time PCR system and PowerUP SYBR Green Master Mix (Thermo Fisher). Primers for negative control regions, Actb and Dll1 open reading frames (ORFs), were previously published (Castro et al. 2011). Primers for the 687 locus are 687_ChIP_PCR_F and 687_ChIP_PCR_R. qPCR using two sets of negative control primers and the 687 primers were performed in triplicate on 2% input DNA, Gsx2-ChIP sample, and IgG ChIP sample. Data are presented as fold enrichment over Dll1-ORF for each condition.

RNA sequencing

E12.5 embryos were collected from Gsx2^RA/+ x Gsx2^EGFP/+ crosses in ice cold PBS and the LGE and a section of embryonic tail were dissected from each embryo. LGEs were stored in RNAlater (Ambion) and tails were used for genotyping. A total of four pairs of LGEs from Gsx2^+/+ and Gsx2^EGFP/RA genotypes were submitted for directional RNA-seq by the Genomics, Epigenomics and Sequencing Core at the University of Cincinnati. Five-hundred nanograms of total RNA was used to isolate polyA RNA using NEBNext Poly(A) mRNA Magnetic Isolation Module (New England BioLabs). Samples were further enriched using SMARTer Apollo NGS library preparation system (Takara Bio USA). A dUTP-based stranded library was generated using NEBNext Ultra II Directional RNA library preparation kit (New England BioLabs). The library was indexed and amplified for nine PCR cycles. Individually indexed libraries were proportionally pooled and sequenced at a depth of 25 million single-end 51-bp reads using a HiSeq 1000 sequencer (Illumina). The RNA-seq data are available from the GEO of the NCBI as GSE162590.

CUT&RUN assays

CUT&RUN assays were performed on dissected LGE tissue from 65 homozygous Gsx2^2xFLAG E12.5 mouse embryos (6 litters). LGE tissue from each litter was pooled, dissociated by pipetting in 0.05% Trypsin-EDTA (Gibco), passed through a 40-µm cell strainer (Corning), step-frozen in CryoStor freezing media (STEMCELL Technologies), and stored at −80°C. Samples were thawed at 37°C and washed in DMEM with 5% FBS. Cells were pooled and washed in DPBS with 0.04% BSA and protease inhibitor cocktail. After binding to Concavalin A beads, cells were divided into separate samples of ~1 × 10⁶ cells each and duplicates were incubated with either the M2-FLAG antibody (Sigma) or an IgG control antibody. CUT&RUN was performed as previously described with the following exceptions (Skene and Henikoff 2017; Skene et al. 2018). A rabbit anti-mouse secondary antibody was used due to the low affinity of protein A for mouse IgG1 (M2-FLAG antibody). MNase digestion after the addition of CaCl₂ was performed for 45 min at 0°C. After stopping MNase digestion, released DNA fragments were purified using the Nucleospin DNA cleanup kit. Libraries were prepared using the NEBNext Ultra II DNA library preparation kit for Illumina sequencers, and PCR products between ~150 and 350 bp were size selected using AMPure XP beads. Bar-coded libraries were sequenced on an Illumina NextSeq 550 sequencer at a depth of ~20 million paired-end 75-bp reads.

We thank Aaron Zorn, Ron Waclaw, and members of the Gebelein and Campbell laboratories for comments on this work. We thank Dr. Matthew Weirauch (Cincinnati Children's Hospital Medical Center [CCHMC]) for bioinformatics advice. We thank Dr. Yueh-Chiang Hu and the CCHMC genome-editing core for helping with the design and creation of the Gsx2^2xFLAG mouse line, and the University of Cincinnati Sequencing Core for RNA sequencing. This work was supported by the National Institutes of Health (R01 NS044080 to K.C. and B.G., R01 NS069893 to M.N. and K.C., and R01 GM079428 to B.G. and H.-W.L.), the National Science Foundation (1715822 to B.G.) and Cincinnati Children's Research Foundation (Trustee Award to H.-W.L.).

Author contributions: This scientific study was conceived and planned by J.S., K.C., M.N., and B.G. The design and making of the Gsx2^2xFLAG mouse allele were done by S.Q., J.S., and K.C. in conjunction with the CCHMC genome-editing core. The transgenic Drosophila constructs and experiments were designed and performed by J.S. and B.G. Protein purification and gel shift analysis were performed by J.S., E.F., and B.G. Embryonic mouse dissections and RNA-seq experiments were performed by S.Q. and K.C. CUT&RUN assays were designed and performed by T.D.F., B.G., R.B.H., and J.S. ChIP-PCR experiments were performed by J.S. Mouse immunostaining was performed by S.Q. Bioinformatics analysis was performed by B.C., J.S., and H.-W.L. The manuscript was written by J.S., K.C., and B.G.