Functional genetic variants can mediate their regulatory effects through alteration of transcription factor binding

Abstract

Functional variants in the genome are usually identified by their association with local gene expression, DNA methylation or chromatin states. DNA sequence motif analysis and chromatin immunoprecipitation studies have provided indirect support for the hypothesis that functional variants alter transcription factor binding to exert their effects. In this study, we provide direct evidence that functional variants can alter transcription factor binding. We identify a multifunctional variant within the TBC1D4 gene encoding a canonical NFκB binding site, and edited it using CRISPR-Cas9 to remove this site. We show that this editing reduces TBC1D4 expression, local chromatin accessibility and binding of the p65 component of NFκB. We then used CRISPR without genomic editing to guide p65 back to the edited locus, demonstrating that this re-targeting, occurring ~182 kb from the gene promoter, is enough to restore the function of the locus, supporting the central role of transcription factors mediating the effects of functional variants.

Fig. 1

Identification of high-confidence multifunctional variants. In a we show the steps involved in proceeding from 5.5 million high-confidence variants in this family to the 26 likely to have effects on gene expression, chromatin accessibility, and DNA methylation. An ATAC-seq peak for which the number of reads is influenced by a specific parental haplotype was required to include a polymorphic sequence within the peak, a polymorphism for which that parent had to be heterozygous. We then further filtered by requiring the locus to have an intermediate DNA methylation value, consistent with only one allele being active, and then filtered to overlap with the eGenes that we found to be in common with Li et al. This list is shown in b with those validated by qRT-PCR marked with an asterisk, and those genes in common to (a) and (b) shown in black. We proceeded with TBC1D4 for further testing

Fig. 2

Overview of the TBC1D4 haplotype. The TBC1D4 gene is at the centromeric end of an ~6 Mb haplotype on chromosome 13q. The genes and three-dimensional interaction domains are depicted, with the lower part of the panel showing the detail in the immediate TBC1D4 region. Three promoters are located within the major loop containing TBC1D4. We show how the promoters and the multifunctional variant are predicted to be organized in three dimensions in vivo, revealing a relative proximity of the multifunctional variant (x) to the TBC1D4 (c) and UCHL3 (d) promoters, despite the multifunctional variant being 182 kb centromeric to the TBC1D4 promoter. Source data of the interaction domains are provided in a Source Data file

Fig. 3

Characterization of the multifunctional variant at the TBC1D4 locus. a Comparative sequence analysis shows three variants within 7 bp, the insertion of a C and two C>T transitions that are only found in Homo sapiens, not in any other primate or Denisovan DNA, while the one Neanderthal read at the locus also shows the ancestral C. The ancestral haplotype encodes a canonical NFκB binding site, with the human-specific haplotype disrupting it in three locations. In b, using data from the GGV browser, we observe that while the ancestral allele is the more common in European and African populations, in South and East Asia the human-specific allele becomes substantially more common in these populations

Fig. 4

Expression of genes within local contact domain. The intronic TBC1D4 cis-regulatory locus encodes an expression quantitative trait locus (eQTL) that acts on the TBC1D4 gene but not the immediate upstream COMMD6 or UCHL3 genes. For TBC1D4, the amount of expression is higher when the ancestral allele (AA) is present compared with the human-specific (HS) allele. Boxplots represent the interquartile range (IQR) with the median denoted by the middle line; whiskers extend to the largest/smallest value no further than 1.5 times the IQR. Expression values are provided in a Source Data file

Fig. 5

Chromatin accessibility within the TBC1D4 multifunctional variant interaction domain. The intronic TBC1D4 cis-regulatory locus encodes a chromatin accessibility quantitative trait locus (caQTL) that is located at a site of strong ChIP-seq enrichment for H3K27ac and of read pileup for the p65 component of NFκB, but is only significantly associated with the overlying ATAC-seq peak, with a higher normalized read count (summit signal) associated with the ancestral allele (AA) compared with the human-specific (HS) allele. Boxplots represent the interquartile range (IQR) with the median denoted by the middle line; whiskers extend to the largest/smallest value no further than 1.5 times the IQR. Summit signal values are provided in a Source Data file

Fig. 6

Allelic chromatin accessibility at the TBC1D4 multifunctional variant. The TBC1D4 multifunctional variant is associated with open chromatin present on the ancestral allele (AA) but not on the human-specific (HS) allele. In a we show the 5 kb region flanking the local ATAC-seq peaks, with the peak (summit ± 250 bp) reflecting the caQTL shown in red. In b we represent polymorphic AA sequences where C>T transitions are present with orange, showing that in the heterozygous GM(128)85 cell line, all of the reads are from the AA, and none from the HS allele in the same cell

Fig. 7

Allele-specific binding of NFkB at variant locus ChIP-seq read pileups for NFκB components p65, RelB, RelC, p50, and p52 from the heterozygous AA/HS GM12878 cell line, showing the 5 kb context above (a), and a ~200 bp detail (below; b) showing most reads of each NFκB component are from the AA

Fig. 8

DNA methylation in the 5 kb context of the TBC1D4 multifunctional variant. We show the DNA methylation (5mC) for each cytosine in green with the corresponding unmethylated (C) fraction of alleles in orange, so that loci found to have very low levels of DNA methylation do not appear to be missing in the representation. We show the mean values for the 2 HS/HS, 7 AA/HS, and 2 AA/AA individuals. There are no cytosines sufficiently close to the multifunctional variants to allow observation of haplotypic patterns of DNA methylation, but there is a pattern of less DNA methylation at loci closest to the multifunctional variant in individuals with the AA haplotype. 5mC values, along with median coverage, are provided in a Source Data file

Fig. 9

Allelic editing leads to local functional changes among edited clones. In a we show the ancestral and human-specific alleles, and a 7 bp deletion that was also generated by CRISPR-Cas9 editing. In b TBC1D4 gene expression levels are shown in the unedited GM12881 (AA/AA) and GM12880 (HS/HS) LCLs, compared with two human-specific clones (HSC1, HSC1) and the 7 bp deletion clone generated from GM12881 (n = 3 independent experiments each). As expected, there is a strong decrease in expression of TBC1D4 in the HS/HS GM12880 LCL compared with AA/AA GM12881, while the edited HSC1, HSC2, and 7 bp deletion clones show levels of expression comparable with the HS/HS unedited control. In c we show that this editing is associated with a decrease in chromatin accessibility (n = 3 independent experiments each), and in d with a decrease in the p65 component of NFκB using quantitative ChIP (n = 3 independent experiments each). Source data are provided in a Source Data file

Fig. 10

Targeting of NFκB to the edited intronic locus helps to restore TBC1D4 expression. In a we show the effect of targeting of the full-length p65 component of NFκB using CRISPR and catalytically-inactive Cas9 to the edited locus (HSC1 cell line), using a guide RNA targeting the edited NFκB binding site. This was associated with the increase in expression levels shown (dots show individual replicates, for clarity). Because we were concerned that this targeting of a locus ~182 kb from the gene’s transcription start site (TSS) was working unexpectedly efficiently, and the possibility that we may be having the introduced p65 binding at other NFκB motifs, including the TSS itself, we added two additional measures. We introduced a p65 construct lacking its DNA-binding domain, preserving its trans-activating domains, and used a control guide RNA targeting a known NFκB binding site at the OXTR gene, with the goal of having this locus act as a sink for any ectopic p65 in the cell nucleus. In b we again show individual results from HSC1 replicates, and also the 7 bp deletion clone as an independent set of replicates. Once again, we see enhancement of TBC1D4 expression, without the same concerns for ectopic p65 activity. The TF replacement strategy at this locus thus appears to be having effects on transcription acting over a distance of 182 kb, which is unusually far for typical epigenetic editing experiments. Source data are provided in a Source Data file