Structural Evolution of Gene Promoters Driven by Primate-Specific KRAB Zinc Finger Proteins

Abstract

Krüppel-associated box (KRAB) zinc finger proteins (KZNFs) recognize and repress transposable elements (TEs); TEs are DNA elements that are capable of replicating themselves throughout our genomes with potentially harmful consequences. However, genes from this family of transcription factors have a much wider potential for genomic regulation. KZNFs have become integrated into gene-regulatory networks through the control of TEs that function as enhancers and gene promoters; some KZNFs also bind directly to gene promoters, suggesting an additional, more direct layer of KZNF co-option into gene-regulatory networks. Binding site analysis of ZNF519, ZNF441, and ZNF468 suggests the structural evolution of KZNFs to recognize TEs can result in coincidental binding to gene promoters independent of TE sequences. We show a higher rate of sequence turnover in gene promoter KZNF binding sites than neighboring regions, implying a selective pressure is being applied by the binding of a KZNF. Through CRISPR/Cas9 mediated genetic deletion of ZNF519, ZNF441, and ZNF468, we provide further evidence for genome-wide co-option of the KZNF-mediated gene-regulatory functions; KZNF knockout leads to changes in expression of KZNF-bound genes in neuronal lineages. Finally, we show that the opposite can be established upon KZNF overexpression, further strengthening the support for the role of KZNFs as bona-fide gene regulators. With no eminent role for ZNF519 in controlling its TE target, our study may provide a snapshot into the early stages of the completed co-option of a KZNF, showing the lasting, multilayered impact that retrovirus invasions and host response mechanisms can have upon the evolution of our genomes.

Significance.

Previous research has investigated the role of KRAB zinc finger genes as repressors of transposable elements, recent data revealed that a number of KZNFs also bind to gene promoters but their role at these genomic sites is not well understood. In this study, we investigate the impact of this KZNF-promoter relationship and how the emergence of new KZNFs that recognize promoters has shaped gene-regulatory networks. We observed a higher-than-average sequence turnover at bound promoters and an increase in the expression of genes with promoter KZNF sites when the binding KZNF was removed. We conclude that members of this gene family have a subtle yet important influence on the shaping of primate gene-regulatory networks and shed light upon a transitional state in genome evolution which may contribute to primate-specific traits such as increased brain size.

Introduction

Krüppel-associated box (KRAB) zinc finger proteins (KZNFs) are a type of transcription factor (TF) encoded by a large gene family consisting of more than 400 loci. These genes have evolved over time through tandem duplication followed by sequence divergence to create a large collection of DNA-binding proteins that mostly function as transcriptional repressors (Emerson and Thomas 2009; Thomas and Schneider 2011; Bruno et al. 2019; Wolf et al. 2020). A typical KZNF protein contains a highly conserved KRAB domain that interacts with the co-repressor protein TRIM28 (KAP1) and a zinc finger domain, containing a highly variable array of zinc fingers that recognize and bind to specific DNA sequences (Groner et al. 2010). Members in the KZNF gene family are primarily distinguished by differences in the composition and size of the zinc finger domain, although other structural variations, including in the KRAB domain, are also represented (Looman et al. 2002). DNA interaction is encoded in just four amino acids in each zinc finger and a single nucleotide substitution can alter the DNA binding capacities of the zinc finger unit, and the KZNF protein in total (Looman et al. 2002). The structure of KZNF genes, amplified by the fact that most KZNFs reside in highly unstable genomic loci, provides a source of rapidly flexible gene-regulatory mechanisms that have high potential to fall under evolutionary mechanisms (Nowick et al. 2010).

In recent years it was established that a large number of individual KZNFs recognize and suppress different classes of transposable elements (TEs) (Thomas and Schneider 2011; Jacobs et al. 2014; Najafabadi et al. 2015; Schmitges et al. 2016; Imbeault et al. 2017; Helleboid et al. 2019). Although this was shown in detail for just a few specific examples, the general concept is that the initial role of KZNF genes is to target and prevent newly arising TE families from replicating in genomes in order to safeguard the integrity of the host genome (Helleboid et al. 2019). However, due to their repetitive nature, TEs are prone to rapidly mutate and sometimes escape binding by the KZNF which can give rise to a new wave of TE insertions (Jacobs et al. 2014). Over time, a new or adapted KZNF gene will evolve to repress this new class of TEs, only to force it to develop new mutations to escape its repressor once again. This pattern of genomic evolution, remarkably similar to a classical evolutionary arms race, has been taking place independently in many different species, to the extent that many species have their own species-specific KZNF genes (Thomas and Schneider 2011; Castro-Diaz et al. 2014; Jacobs et al. 2014; Ecco et al. 2017).

The regulatory effects of KZNFs are not only limited to the repression of TEs. After, or in parallel to their initial role, KZNFs are often co-opted for other functions. Some remain associated with TEs to form KZNF-controlled TE-mediated regulatory elements (Ecco et al. 2016; Pontis et al. 2019). These transposable element-embedded regulatory sequences (TEeRS) and their associated KZNFs have been shown to regulate gene expression during neuronal development in humans (Turelli et al. 2020). KZNFs also play a role in the regulation of TE-mediated cryptic gene promoters (Sundaram and Wysocka 2020; Haring et al. 2021). Here, the KZNF regulates gene expression by binding directly to a TE-derived gene promoter. ZNF57 and ZNF445 have been shown to play important roles in imprinting (Li et al. 2008; Quenneville et al. 2011; Takahashi et al. 2019) and ZNF568 has been shown to regulate IGF2 in the placenta (Yang et al. 2017).

It was recently shown that in addition to their prime TE target sites, a subgroup of KZNFs also has the ability to recognize and bind to gene promoters independent of TEs (Frietze et al. 2010; Schmitges et al. 2016; Imbeault et al. 2017; Helleboid et al. 2019; Farmiloe et al. 2020). The frequency of promoter occupancy varies widely between KZNFs, with some KZNFs showing binding capacity to over 2,000 gene promoters. Comparative analysis of the KZNF binding sites in TEs and the TE-independent binding sites in gene promoters revealed clear similarities in DNA sequence (Farmiloe et al. 2020). The binding of the KZNF to gene promoters may have been an inevitable side effect of the acquired recognition potential of the KZNF to the target TE, but once established, this secondary and coincidental regulatory involvement of the KZNF could have become indispensable for normal gene regulation. In fact, after the TE itself has lost its capacity to retrotranspose, the KZNF may have become redundant for the control of the TE, but not for the genes it has evolved to regulate directly through their promoters (Ecco et al. 2017). We previously showed that in general, the binding of KZNFs to promoters correlates well with brain-developmental gene expression patterns (Farmiloe et al. 2020), indicating that binding of KZNFs influences neuronal gene expression.

From an evolutionary perspective, the recent emergence of many primate-specific KZNFs raises the important question of how genes have coped with these new regulators that may have created a temporary imbalance creating an evolutionary impetus for stabilization. Such newly emerged regulatory roles of KZNFs on gene regulation would likely have evoked structural changes in the gene promoters they developed the ability to bind, which may have required further “stabilizing” genomic adaptations to cope with the newly acquired KZNF-mediated level of gene regulation. We therefore hypothesize that gene promoters bound by KZNFs have been under a similar pressure to change and “escape” binding by KZNFs as TEs. In this study, we investigated the relationship between KZNFs and KZNF-bound gene promoters to elucidate the extent of the evolutionary pressure that KZNFs have exerted on human gene expression patterns.

Results

ZNF519, ZNF441, and ZNF468 are Widely Expressed Primate-specific KZNFs, Binding to Thousands of Gene Promoters Genome-Wide

51 KZNFs that bind to 100 or more gene promoters were outlined in Farmiloe et al. (2020). To investigate genomic adaptations of the host genome in response to the emergence of the KZNF-mediated level of gene regulation, we focused on recently evolved promoter-binding KZNFs that are expressed in a variety of tissues. Out of the set of promoter-binding KZNFs, we selected three KZNFs based on the following criteria; 1) these KZNFs emerged specifically in primates, 2) they interacted directly with a high number of gene promoters, and 3) they showed high levels of expression in hESCs and/or hESC-derived cortical tissues (fig. 1A and B). The KZNFs that fit our criteria best were ZNF519, ZNF441, and ZNF468 and they were selected for further in-depth investigation. All three KZNFs are expressed in the brain and, to a lesser degree in a range of human tissues (supplementary fig. 1, Supplementary Material online). These selection criteria led to the exclusion of some KZNFs, including those that were shown to bind to the highest number of gene promoters (ZNF202, ZNF534).

Fig. 1.

Co-evolution of three KZNFs with TEs is paralleled by co-option for direct gene-regulatory properties. A, The top ten promoter binding KZNFs by number of binding sites present in promoter regions (promoter defined as 5,000 bp upstream and 1,000 bp downstream of transcription start site). Adapted from Farmiloe et al. (2020). B, DESeq2 basemean expression values of KZNFs in human and rhesus embryonic stem cells (ESC) and weeks 1–5 wk1-wk5) of cortical organoid development from differential analysis of RNA-seq data. Data points show the average of two replicates for each species and time point. C, ChIP-Seq density plots showing ZNF and KAP1 binding at TEs and gene promoters (ZNF519—MER52; ZNF441—AluY, AluYa5; ZNF468—MER11A). D, The emergence and expansion of the TE classes associated with ZNF441, ZNF468, and ZNF519. Colored sections of the bars show approximately when the TEs emerged and density of color shows peak transposition activity E, evolutionary tree showing approximate time of emergence of TE classes and KZNFs.

Comparative genomics analysis confirmed that all three KZNFs are primate-specific and not present in any of the species that diverged before the last common ancestor with new world monkeys (NWMs) (Imbeault et al. 2017) (supplementary fig. 2, Supplementary Material online). Traces of the ZNF519 locus can be found in the squirrel monkey genome which suggests the presence of a ZNF519 gene in the common ancestor of humans and NWMs that has been lost in the earlier lineages. Now ZNF519 is only present in gibbons and great apes including humans (fig. 1E). ZNF441 is detected in the genomes of NWMs, old world monkeys (OWMs), gibbons, and great apes. ZNF468 emerged later and is only detected in the genomes of OWMs, gibbons, and great apes.

Like many other KZNFs, ZNF519, ZNF441, and ZNF468 are associated with a specific class of TEs (Table 1, Imbeault et al. 2017). For each TE class studied we see a rise in the number of elements at the same time as we see the emergence of each KZNF, suggesting their emergence is linked to the respective TE invasions. The number of MER52 elements, a class bound by ZNF519 (Imbeault et al. 2017), increases in NWMs and OWMs and then remains relatively consistent in the great apes indicating a lack of transposition activity (fig. 1D, supplementary fig. 2A, Supplementary Material online). ZNF441 has been shown to target AluY elements, specifically subclasses AluY and AluYa5 (Imbeault et al. 2017). AluY elements are still active in the human genome (Bennett et al. 2008). Their numbers show large species-specific expansions in OWMs (fig. 1D, supplementary fig. 2C, Supplementary Material online). MER11A elements, bound by ZNF468 (Imbeault et al. 2017), follow a similar evolutionary pattern as MER52 elements. They emerge in OWMs and we see a rapid expansion in these species followed by stabilization of numbers in apes and humans (fig. 1D, supplementary fig. 2C, Supplementary Material online).

Table 1.

Promoter Binding KZNFs and the TE Families They Recognize

ZNF	# Promoters Bound	TE Family Recognized
ZNF519	1843	Mer52
ZNF441	816	AluY
ZNF468	3545	MER1A

Further analysis of the ChIP data for these KZNFs shows a defined peak at associated TEs (fig. 1C). It is possible that the KZNF remains important to control the TE's regulatory potential, even if the TE itself has lost the capability to retrotranspose. To address this possibility we tested the activity of MER52 elements in a luciferase assay in the absence and presence of ZNF519, its partner KZNF. The luciferase assay was performed in mouse embryonic stem cells (mESCs), which lack all primate-specific KZNFs including ZNF519. A MER52D element cloned upstream of luciferase had a mild repressive effect on luciferase expression. (supplementary fig. 3, Supplementary Material online, two-way analysis of variance (ANOVA), Tukey's multiple comparison test P < 0.0001). When ZNF519 was ectopically expressed, there was no change to the regulatory effect of the MER52D element on luciferase activity (supplementary fig. 3, Supplementary Material online), noting that this may be related to the very limited regulatory potential of MER52 elements. Further analysis of endogenous KAP1 binding data shows a depletion of KAP1 at KZNF-bound gene promoters suggesting that KAP1 is not recruited at these loci (fig. 1C). Previous analyses published in Farmiloe et al. (2020) supported the endogenous binding of KZNFs at gene promoters in the absence of KAP1 suggesting endogenous binding of KZNFs at these loci. The fact that ZNF519 is fixed in gibbons and great ape-lineages raises the possibility that ZNF519 was co-opted for a TE-independent role in our genome.

High Similarity Between ZNF Binding Sites in TEs and Promoters

De novo motif discovery analysis of the ZNF519 binding motif in highly bound MER52 elements and TE-independent promoters revealed high compatibility of ZNF519 core binding motifs (fig. 2A). Mapping of ZNF519 ChIP-seq data to a consensus MER52 sequence also showed the presence of the ZNF519 transcription start site (TSS) motif at the summit of the MER52 binding peak (fig. 2B). A similar pattern was observed for ZNF441 and ZNF468; de novo motif prediction for TE-independent TSS binding sites for both ZNF468 and ZNF441 closely matches the motif found in TE binding sites (fig. 2C and E). These promoter motifs can also be found at the summit of the ZNF441 peaks mapped to the AluY consensus sequence and for ZNF468 summits mapped to the MER11A consensus sequence (fig. 2D and F). Therefore, ZNF519, ZNF441, and ZNF468 seem to be clear examples of KZNFs that were initially recruited to control transposable elements, but because the binding domain for each KZNF also recognized numerous gene promoters, they have become integrated into our gene-regulatory networks through the phenomenon of co-option.

Fig. 2.

Transposable element motifs recognized by KZNFs are also seen in TSS binding sites. A, C, E, HOMER de novo motif discovery in DNA sequences ±50 bp from the KZNF summits in TEs and gene promoters. (A) The ZNF519 summits in MER52 elements and promoter regions. (C) The ZNF441 summits in Alu elements and promoters with and without an Alu element. (E) The ZNF468 summits in MER11A elements and promoter regions. B, D, F, KZNF summits ±7 bp lifted over to the UCSC repeat browser, shown at their recognized repeat family's consensus sequence show the presence of the KZNF-bound motif recognized in promoters for (B) ZNF519, (D) ZNF441, and (F) ZNF468.

In theory, it is possible that after the KZNFs became redundant for TE control, its binding specificity changed under pressure of some of the genes it evolved to regulate as part of its co-opted function. This does not seem to be the case however: A multiple sequence alignment of the protein sequences shows very few differences in the zinc finger (ZNF) domains and contact residues of all three KZNFs between the oldest lineages and humans (supplementary figs. 5–7, Supplementary Material online). Despite these minor differences in their DNA binding domains, the predicted binding motifs of all three KZNFs are identical between the different orthologs (supplementary fig. 4, Supplementary Material online). The observed high level of conservation suggests that these KZNFs still serve an essential purpose in the human genome and may have been doing so since their emergence. In addition, the high occupancy of these three KZNFs on gene promoters suggests that any subsequent adaptive changes to restore the balance of gene expression must have come from modifications to the KZNF-bound promoters rather than the KZNFs themselves. We, therefore, expect that a genome that is confronted with a newly emerged KZNF that binds to a lot of gene promoters will show additional adaptations in these promoters to balance out the influence it has on gene regulation.

Accumulation of Indels and Substitutions at the KZNF Target Site in Gene Promoters

The consequences of a TE invasion are clearly not limited to the impact of TE insertions. Alongside the consequences of KZNFs which have an effect on gene regulation through TE-mediated insertions, the impact of KZNFs directly on gene promoters in the absence of TEs may be substantial as well. We investigated the presence of nucleotide substitutions and indels around the summits of the KZNF peaks in promoters. Because promoters often show high levels of conservation, the presence of mutations at the site of KZNF binding that corresponds to the time point of emergence of the KZNF gene could be a signature of local evolutionary adaptations. To identify promoters with hominid-specific insertions or deletions, multiple sequence alignments of KZNF-bound promoter sequences were made between humans and two OWM species: Rhesus Monkey and Green Monkey. To distinguish between OWM-specific indels or hominoid-specific indels, the NWM species marmosets were taken along in the alignments as a outgroup. The regions analyzed centered around the summit of the KZNF peak, generated from ChIP-seq data, in the human genome (Imbeault et al. 2017), flanked by 140 bp upstream and downstream promoter sequence for ZNF519 and ZNF441 and 175 bp for ZNF468. These alignments were then coded for substitutions, insertions, deletions, and identical sequences, respectively. Only alignments that had complete sequence data for each species were included, which led to the exclusion of promoters with incomplete coverage in any one of the four species. Despite an overall high sequence similarity between the promoter sequences of each species, a large number of promoters that remained after the screening process displayed a hominoid-specific insertion or deletion around the ZNF-bound promoter sequence while the adjacent sequence did not show this level of mutation (fig. 3). This was the case ZNF519 in 164 out of 256 promoters (∼64%) (fig. 3A), for ZNF441 in 56 out of 86 (∼65%) promoters (fig. 3B) and for ZNF468 in 38 out of 59 (∼64%) promoters (fig. 3C). Plotting of the relative coverage of indels along the promoters showed an increased likelihood for an indel to be present around the core binding site of the ZNF compared to the adjacent sequence (coverage graphs under each of the heatmaps, fig. 3). These data suggest that indels in ZNF-bound promoters are prevalent, and often occur at the core site of ZNF binding.

Fig. 3.

ZNF-bound promoters with hominid-specific insertions or deletions around the summit. Showing ZNF summits and adjacent bases containing indels specific to the hominid line after alignment of the human sequence with that of rhesus, green monkey, and marmoset. Composite values of all promoters are shown in the profile plots below each heatmap (A) ZNF519, (B) ZNF441, (C) ZNF468 (gray = no change, red = insertion in human, blue = deletion in human, black = substitution in human).

Promoter Regions Bound by KZNFs Show Higher Sequence Turnover at the Site of KZNF Binding

We further expanded our comparative analysis of KZNF-bound promoters by analyzing the levels of evolutionary sequence conservation using PhyloP 100 way conservation data available through the University of California Santa Cruz (UCSC) genome browser; 31 KZNFs that bind to >100 gene promoters were included in the analysis. For each individual KZNF, the average PhyloP value was taken at each base pair around summits in gene promoters. The regions analyzed centered around the ZNF-peak-summit and included 100 bp flanking sequences downstream and upstream. The values were then normalized and visualized in a heatmap for all of the KZNFs (fig. 4A). As a control, the analysis was repeated for the same sized-region 500 bp downstream from each of the KZNF summits (fig. 4A). The expectation is that because of the vicinity to the KZNF summit, these control regions are still in the vicinity of the TSS and still have overall high conservation.

Fig. 4.

Collective conservation scores at KZNF summits in bound promoters. (A) Normalized PhyloP conservation scores averaged across all KZNF summits in gene promoter regions ±100 bp and control regions 500 bp downstream of summits. blue = lower, orange = higher normalized PhyloP score than the average across summits for each KZNF. (B–D) Averaged, unnormalized PhyloP scores for ZNF519, ZNF441, and ZNF468 around the ChIP-seq summit and around control regions.

Twenty-two out of the 31 KZNFs included in this analysis show a pattern of reduction in the PhyloP conservation value at the binding site of the promoter-bound KZNF compared to the flanking sequences up and downstream (fig. 4A). A lower conservation value is an indicator of increased sequence turnover. Whereas in the control region, 500 bp downstream of the KZNF-binding site the range of conservation values was very similar, the pattern was completely homogenous, indicating no local increase or decrease of conservation in the control region. For 10 KZNFs this pattern is particularly pronounced; the region approximately 30 bp up and downstream of the summit of the ZNF-binding site shows a clear reduction in conservation when compared to the surrounding region. Whereas the reduced conservation was clearest for ZNF519 (fig. 4A and B) and some other KZNFs a more modest reduction of conservation values was observed for ZNF441 (fig. 4A and C) and ZNF468 (fig. 4A and D). To determine the number of promoters under the influence of positive (positive phyloP score) versus purifying (negative phyloP score) selection, the bound promoters for ZNF519, ZNF441, and ZNF468 were split based on their average score across the 200 bp (supplementary fig. 8A–C, Supplementary Material online). The summit-focused reduction in scores was maintained for both groups of promoters. These results support our initial findings from the indel analysis and show that the summits of the KZNF binding sites in gene promoter regions show an increased sequence turnover compared to the surrounding bases in the same gene promoters, or a control region downstream. Because this pattern is observed for most of the KZNFs that bind to gene promoters directly, the focal reduction in sequence conservation precisely at the site of KZNF binding seems to be a more general phenomenon that may point to local evolutionary adaptations as a direct response to KZNF binding.

To validate these findings, we repeated the analysis for a subset of known TFs using ChIP-Seq data from the Encyclopedia of DNA Elements (ENCODE) in either HEK293 or H1 cells (supplementary fig. 9, Supplementary Material online). In well-documented TFs we expect to see the opposite pattern to that observed in the KZNF data, with high levels of sequence conservation at the summits. Indeed for four out of seven TFs, AFT2, EP300, REST, and CTCF there is a strong peak showing very high conservation centered around the summit of the binding site (supplementary fig. 9, Supplementary Material online). Only one of the other TFs, POLR2A, shows a similar pattern to the KZNFs (supplementary fig. 9C and E, Supplementary Material online). POL2RA is known to bind to transcription start sites in gene promoters which show a high rate of turnover in primates (Taylor et al. 2006), this is reflected in the reduced conservation scores for POLR2A binding sites. It is unlikely that the turnover at KZNF sites is explained by turnover at POLR2A sites as only 0.02% of the KZNF summits ±100 bp overlapped 50% or more with a POLR2A summit ±100 bp (supplementary fig. 9F, Supplementary Material online). Taken together, the analysis of relative conservation values around the binding sites of KZNFs and known TFs supports our hypothesis that the emergence of KZNFs can exert selective pressure on the KZNF binding sites in gene promoters.

Genetic Deletion of KZNF Genes Reveals Widespread Effects on Gene Expression in Neuronal Tissues

We previously found a correlation in expression profiles of KZNFs and KZNF-bound gene promoters in the brain, suggesting KZNFs are widely integrated in neuronal gene-expression networks. Indeed, gene ontology analysis of genes bound by each of the KZNFs showed a significant enrichment of genes expressed in the brain for ZNF519 and ZNF468 (supplementary Tables S1 and S2, Supplementary Material online).To test this relationship further, we used Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR)/Cas9 to generate genetic deletions of either ZNF519, ZNF441, and ZNF468 in hESCs. If the KZNFs have a regulatory effect on genes where they bind, we would expect to see changes in the expression of bound genes when the binding KZNF is removed from the genomic context.

To assess the impact of KZNF binding on gene expression in the brain, the three KZNF knockout (KO) hESC cell lines were directed into a neuronal fate by generating cortical organoids. Correct genetic deletion and complete absence of expression for each KZNF were confirmed by RNA-seq in hESCs for ZNF519 and ZNF468. Confirmation of ZNF441 KO was done in D35 cortical organoids because of low ZNF441 expression levels in hESCs (fig. 5A, C and E). ZNF519 and ZNF441 cortical organoids were grown for 35 days before harvest. ZNF468 KO cortical organoids began to display phenotypic changes compared to wild type (WT) at day 8 and were only grown for 5 days before harvest, a time point at which the ZNF468-KO and WT organoids were morphologically comparable.

Fig. 5.

Knock out of KZNFs results in changes inbound gene expression. A, C, E, Overview of ZNF519, ZNF441 and ZNF468 loci, gRNAs used for CRISPR-Cas9 KO and RNA of WT and KO hESCs/Day 35 cortical organoids, scaling based on the number of mapped reads. B, D, F, Boxplot showing comparison of log₂ fold change of expressed (baseMean >10), high-confident KZNF-bound genes (gray) compared to unbound genes (white) after (B) ZNF519 KO in 5-week old cortical organoids (bound n = 2,311, unbound n = 10,598), (D) ZNF441 KO in 5-week old cortical organoids (bound n = 850, unbound n = 11,170) (F) ZNF468 KO in 5-day old cortical organoids (bound n = 397, unbound n = 11,345). **** = P < 0.0001, ** = P < 0.01, * = P < 0.05, Wilcoxon rank sum test with continuity correction. Red dashed lines were calculated independently and show the 95% CI of a 10,000 times bootstrapped median of a set of unbound genes with the same sample size as the target genes. Individual data points are not shown.

We next assessed the collective change in expression of genes bound or not bound by each of the KZNF in WT and KO cortical organoids. Genes bound by ZNF519 show a small but significant increase in expression compared to unbound genes in ZNF519-KO organoids (median log2 fold change (log2FC) D35: 0.018 versus −0.0055, P < 2.2e-16, Wilcoxon rank sum test with continuity correction) (fig. 5B). ZNF441-KO organoids also show increased collective expression of ZNF441 bound genes compared to genes without binding of ZNF441 in their promoters (median log2FC D35: 0.0066 vs. 0.00011, P < 0.01) (fig. 5D). Finally, a similar relative increase in expression was also observed for ZNF468-bound genes in ZNF468-KO organoids (median log2FC D5: 0.023 vs. 0.0079, P < 0.05) (fig. 5F). The analysis was repeated with using a random set of genes the same size as the sets of KZNF-bound genes (supplementary figs. 10 and 11, Supplementary Material online), the results of this comparison were significant for ZNF519 and ZNF441 (P > 0.0001, P < 0.05) but not for ZNF468.

The differential expression analysis was repeated for the TE classes recognized by ZNF519 (MER52), ZNF441 (AluY), and ZNF468 (MER11A). No significant differences were observed between the WT and KO at individual TEs. A collective comparison of normalized read counts showed no significant changes in the expression of MER52 or MER11A elements in the ZNF519 and ZNF468 knockouts respectively (supplementary fig. 12A and C, Supplementary Material online). The slight increase in expression of AluY elements after ZNF441 KO was significant (median normalized read count WT: 2.033, KO: 2.13, P = 0.0001459, supplementary fig. 12B, Supplementary Material online). DESeq2 results and normalized read counts can be found in supplementary data files 1 and 2, Supplementary Material online.

In summary, cortical organoids derived from all three KZNF-KO hESC lines showed a modest collective increase of expression of their respective KZNF-bound target genes, supporting our hypothesis that KZNF-binding to gene promoters influences the expression of affected genes. Furthermore, the observed increase in expression of KZNF-bound genes in KZNF-KO organoids, shows that endogenous levels of ZNF519, ZNF441, and ZNF468 affect their target genes under normal physiological conditions.

In-Depth Analysis of ZNF519 KO and Overexpression Experiments Shows the Opposite Trend for Changes in the Regulation of Bound Genes

To further investigate the extent of the influence of KZNFs on gene expression we continued our analyses with ZNF519 and examined the effects of ZNF519 KO on gene expression in hESCs and day 14 cortical organoids. Similar to the observations in day 35 organoids, the collective expression of ZNF519-bound genes in day 14 cortical organoids is increased (median log2FC D14: 0.012 vs. −0.0017, P = 1.41e-11) (fig. 6A). Conversely we see a reduction in the collective expression of ZNF519-bound genes in hESCs (median log2FC of −0.026 vs. −0.0042, P = 4.981e-06) (fig. 6B). These data suggest a regulatory effect of ZNF519 on gene expression which is active at certain developmental time points and in specific tissues only. Based on the gene networks that are expressed at these times and in these tissues the regulatory effect of ZNF519 could be subtly different.

Fig. 6.

In-depth analysis of ZNF519 KO and overexpression experiments shows reciprocal changes in the regulation of bound genes. (A) Boxplot showing comparison of log₂ fold change of expressed (baseMean >10) high-confident KZNF-bound genes (gray) compared to unbound genes (white) in ZNF519 KO cortical organoids of 2 weeks old (day 14, bound n = 2,293, unbound n = 10,926), (B) ZNF519 KO hESCs (bound n = 2,282, unbound n = 9,937). **** = P < 0.0001, Wilcoxon rank sum test with continuity correction. Red dashed lines were calculated independently and show the 95% CI of 10,000 times bootstrapped median of a set of unbound genes with the same sample size as the target genes. Individual data points are not shown. Mind difference in y-axis. (C) Schematic showing ZNF519 overexpression experiment set up in HEK293 cells. (D) RNA-seq at ZNF519 locus confirms overexpression in HEK293 cells. Mean of three replicates shown, scaled on number of mapped reads (excluding ZNF519 locus). (E) Boxplot showing a comparison of log₂ fold change of expressed (baseMean >10) high-confident ZNF519-bound genes (gray, n = 2,268) compared to unbound genes after overexpression of ZNF519 (white, n = 8,222) after ZNF519 overexpression, **** = P < 0.0001.

Finally, we analyzed the effect of ZNF519 over expression (OE) on the expression of ZNF519-bound genes. (fig. 6E). Ectopic expression of ZNF519 was verified using RNA-seq (fig. 6F). Overall, we see a collective decrease in the expression of ZNF519-bound genes when compared to unbound genes (median log2FC OE: −0.029 vs. −0.0022, P < 0.0001) (fig. 6G). This observed change mirrors the increase in expression of ZNF519-bound genes we see in ZNF519-KO cortical organoids suggesting that the effect of ZNF519 on bound genes is related to levels of ZNF519 in cells. These results establish that at endogenous expression levels, the promoter-bound KZNFs subject to this study are able to regulate gene expression directly by binding to gene promoters, independent of the TEs they initially evolved to recognize.

Discussion

Previous studies have established an arms-race model for some specific KZNFs and the TEs they recognize (Jacobs et al. 2014; Imbeault et al. 2017). Under this model, KZNFs could have come to bind gene promoters by chance as a side effect of the arms race with TEs. Indeed, for ZNF519, ZNF441, and ZNF468 we found a high similarity between the binding sites recognized in TEs and gene promoters. Our data offers a view into a “transitional” state affecting gene-regulatory networks suggesting that these KZNFs are in the process of co-option for gene-regulatory functions that are unrelated to their capacity to recognize and bind TEs that have long lost their capability to retrotranspose. Data from the KO and overexpression experiments show that these KZNFs are capable of influencing gene expression and most likely exert a repressive effect upon bound genes. Our analyses show binding of the KZNFs to promoters in the absence of KAP1, this raises the question of the mechanism behind the gene-regulatory effect of the KZNFs we studied. Further study is needed to truly understand this process, however, it is possible that the physical presence of the KZNF binding at gene promoters interferes with other transcription factor activities and through this mechanism gene expression is regulated.

We further discovered a high turnover of sequence at the core ZNF-binding site in promoters that display generally high sequence conservation. The increased likelihood of insertions, deletions, and substitutions and the higher rate of sequence turnover at KZNF binding sites in gene promoters suggests that KZNFs could be driving changes in DNA sequence at promoter-binding sites. Our analysis of this process is limited in that we are only able to study promoters that are still bound by KZNFs, we do not have information on promoters that may have changed so much that they are no longer recognized by the KZNFs. Information on previously bound promoters would give us more insight into this process, however, based on our observations, we propose a model of “promoter adaptation” where a newly emerged KZNF is able to recognize motifs in gene promoters as well as TEs. The subsequent repressive effect of the KZNF on these promoters exerts selective pressure on the promoters to modulate the consequential gene-regulatory influence of the KZNF. This becomes evident by a higher rate of sequence turnover at the site of KZNF binding while the promoter adapts to the new KZNF-mediated regulatory influence (fig. 7). The collection of genomic adaptations to regain a new gene-regulatory balance, drives a wave of subtle changes in gene expression at many genetic loci simultaneously. At first sight, the effect of any particular KZNF on any particular promoter may seem noticeable but modest in absolute terms. However, it's important to consider that our analysis shows a modest change in expression for a large set of KZNF-bound genes, meaning that the expression of many genes has been affected in a modest way. Our analysis of these three KZNFs gives us a unique insight into evolutionary processes that are ongoing in the human and primate genomes. If this process were also to hold true for the other ∼160 primate-specific KZNFs, their contribution to the evolution of gene expression in humans and primates should not be underestimated. Although evolutionary time runs slowly we highlight the importance of recognizing the genome as an entity that is constantly in flux. Our data suggests that KZNFs are able to contribute to this change and could also be drivers behind the features that differentiate primates and humans from other species.

Fig. 7.

Promoter escape model of evolution at promoter KZNF binding sites. Promoter escape model of evolution at promoter KZNF binding sites: 1) Binding motif present in TE and recognized by KZNF is also present in the gene promoter region. 2) KZNF also recognizes promoter motifs and binds there, affecting gene expression. 3) The effect of KZNF binding on gene expression exerts a selective pressure at the locus, either leading to a loss of binding site or strengthened binding by the KZNF.

Our study emphasizes that the invasion of a genome by a new class of TEs has an even more far-stretching and long-lasting impact on a species’ gene-regulatory network than was previously considered, and it happens on clearly distinguishable regulatory levels: First by the addition of new TE- and KZNF-TE-mediated regulatory functions; second by the evolution of KZNF-mediated control over gene promoters, and lastly by the increase sequence turnover in gene promoters at the sites of KZNF binding. Whereas it's impossible to know the sequence of events that led KZNF to bind both TEs and promoters, the most likely scenario is that most promoter-binding KZNFs evolved to bind gene promoters at the same time as the TEs, simply because the targeted TE sequence has similarity to some gene promoters. A careful dissection of the evolutionary histories of both the TEs, promoters, and the KZNFs could sometimes give clues about that: Whereas such analysis did not yield clues for the KZNFs in this study, the evolutionary history was analyzed in much detail for ZNF91/SINE-VNTR-Alu (SVA) and ZNF93/LINE1 (Jacobs et al. 2014) revealing that both KZNF gene ZNF91 and ZNF93 were around quite some time before their structural changes allowed it to recognize and repress the newly invading SVA and L1PA classes, respectively. In short, earlier forms of the KZNF were not able to repress the TE, but later versions were. This shows that with regards to KZNFs and TEs, the KZNFs were present earlier than the TEs but adopted their structure quickly once the TE invasion started. Keeping in mind that KZNF binding properties evolved as a result of TE invasions, we believe that the most likely scenario is that most KZNFs evolved to bind gene promoters at the same time as the TEs. A less likely scenario, in our view, is that KZNFs evolved to bind gene promoters before TEs because this could have happened straight after the KZNFs first emerged in primate genomes, which often preceded the emergence of TEs. We do not intend to claim that this is the case for every KZNF in our genome and some KZNFs are likely to have evolved gene-regulatory properties independent from TEs, for which some of the older KZNFs (without clear TE targets as identified by Imbeault et al. 2017) may be good examples for that.

Altogether, our study reveals the multilayered impact of the emergence of primate-specific KZNFs on human neuronal gene expression patterns and raises the exciting question of how the combination of these regulatory effects is influencing the evolution of species-specific gene-regulatory networks.

Materials and Methods

Human and Rhesus RNA-Seq Data

Basemean values from DESeq2 differential expression analysis data published by (Field et al. 2019) were used for analyses between human and Rhesus cortical organoid development. The data plotted is the average of two replicates.

Analysis of Published ChIP-Seq Data

The ChIP-exo data for this analysis was generated by Imbeault et al. (2017) (NCBI gene expression omnibus (GEO) database accession number GSE78099). Using Trimmomatic (Galaxy v0.36.5) (Bolger et al. 2014), the reads were processed and adaptor and illumina-specific sequences were removed. The reads were mapped to the human genome (assembly GRCh37/hg19; (Lander et al. 2001) using Bowtie2 (Galaxy v2.3.4.2) (Langmead and Salzberg 2012) with single-end, very sensitive end-to-end settings. Summits were generated using Model-based Analysis of ChIP-seq 2 (MACS2) (Zhang et al. 2008) with default settings and intersected with the list of peaks in gene promoters generated in Farmiloe et al. to produce a final list of KZNF summits in gene promoter regions. The TE families most recognized by each KZNF were taken from published data (Imbeault et al. 2017). TE family locations were extracted from the UCSC genome browser repeatmasker track for MER11A, MER52 elements and Alu subclasses AluY and AluYa5. ChIP density plots were generated using computeMatrix (Galaxy Version 3.1.2.0.0) and plotHeatmap (Galaxy Version 3.1.2.0.1) (Ramírez et al. 2016).

KZNF Evolutionary History and Binding Motif Analysis

The human exonic sequences from each KZNF transcript of interest (ZNF519: ENSG00000175322, ZNF441: ENSG00000197044, ZNF468: ENSG00000204604) were extracted from the UCSC genome browser hg38 using the table browser tool (Kent 2002; Karolchik et al. 2004) http://genome.ucsc.edu. The equivalent primate sequences from panTro5, panPan2, gorGor5, ponAbe3, nomLeu3, rheMac10, macFas5, papAnu4, calJac3, and saibol1 were retrieved using the UCSC BLAST-like alignment (BLAT) tool (https://genome.ucsc.edu/cgi-bin/hgBlat (Kent 2002). This comparison was confirmed by lifting over the coordinates of the sequences returned in the BLAT search from primate to human using the UCSC liftOver tool. If all the exons for each gene were found in proximity in the primate genome this was taken to be a paralogous gene and the sequence was extracted. The genomic sequences were then converted to peptide sequences using the ExPAsy translate tool (Gasteiger et al. 2003) and a multiple sequence alignment was performed on these sequences using ClustalW (1.2.4) (Sievers et al. 2011). Predicted zinc finger domains and motifs were generated using the Persikov tool (http://zf.princeton.edu/, Persikov and Singh 2014).

TE Evolutionary Analysis

The TE families associated with each of the three KZNFs were identified from previously published data (Imbeault et al. 2017). Coordinates for the loci of all the TEs from each family recognized by a candidate KZNF (MER51, MER11A, ALuY) were downloaded from the RepeatMasker track on the UCSC genome browser for the human and other primate genomes. The human coordinates were then converted to each of the primate genomes using the UCSC liftOver tool to find the number of homologous loci in other genomes. The minimum amount of bases set to remap was set to 0.95. The analysis was done pairwise between the human genome and each primate genome available. The flanking region, 1000 bp immediately downstream of the TE coordinates was also generated and lifted over from human to each of the primate genomes and this number was used to correct for discrepancies in annotation accuracy. Charts were made using the dotplot function of the ggplot2 package in R (Wickham 2016). Species were grouped as follows:

Great apes: Chimpanzee (panTro5), Bonobo (panPan2), Gorilla (gorGor5), Orangutan (ponAbe3)

Apes: nomLeu3

Old-world monkeys: Rhesus macaque (rheMac10), Crab eating macaque (macFas5), Baboon (papAnu4), Green monkey (chlSab2), Golden snub-nosed monkey (rhiRox1), Proboscis monkey (nasLar1)

New-world monkeys: Marmoset (calJac3), Squirrel monkey (saiBol1)

Basal primates: Tarsier (tarSyr2), Bushbaby (otoGar3), Mouse lemur (micMur1)

KZNF Expression GTEx

The tissue expression graphs for ZNF519, ZNF441, and ZNF468 were generated on the Genotype-Tissue Expression (GTEx) Portal (https://gtexportal.org/home/) using GTEx Analysis Release V8 (dbGaP Accession phs000424.v8.p2).

KZNF De Novo Motif Analysis

De novo motif analysis for ZNF519, ZNF441, and ZNF468 in highly bound MER52s, MER11As, Alus, and TSSs (coverage > 100 on the UCSC genome browser) was performed with Hypergeometric Optimization of Motif EnRichment (HOMER; Heinz et al. 2010). As input, a bed file was used with locations of the KZNF-binding summit plus/minus 50 bp.

KZNF Binding at Repeat Elements

ZNF519, ZNF441, and ZNF468 summits were extended with 7 bp on each side and subsequently a lift over to the hg38 repeat browser was performed using the liftOver tool from the UCSC genome browser (http://hgdownload.soe.ucsc.edu/admin/exe/) using the hg19_to_hg38reps.over.chain as provided by the UCSC repeat browser (https://repeatbrowser.ucsc.edu/). The file was sorted using bedSort and a coverage track was generated using bedtools genomecov (-bg -split) followed and the bedGraphToBigWig tool to specifically visualize the summit of ZNF519, ZNF441, and ZNF467 binding at the consensus MER52A, AluY, and MER11A, respectively.

Cloning of MER52 Elements into Luciferase Reporter Plasmid

A MER52D with a peak height >100 reads was selected and amplified by polymerase chain reaction (PCR) using flanking primers (chr10:56667794-56670008, assembly GRCh37, forward 5′-3′: CCATACCCTATGAAAGCTGGTC, reverse 5′3′: GGGAGATTGTACCTTGATGAC) with LongAmp^® Taq DNA Polymerase (NEB) with an annealing temperature of 57 °C. Amplicons were purified using the QIAquick Gel Extraction Kit (QIAGEN), and cleaned with the DNA Clean & Concentrator™-5 Kit (ZYMO Research). Blunting of 3′ ends was done using DNA Polymerase I, Large (Klenow) Fragment (NEB) before phosphorylation of 5′ ends using T4 Polynucleotide Kinase (NEB). Inserts were ligated upstream of the luciferase reporter in the pGL4.12[luc2CP]SV40 plasmid that was digested using EcoRV (Thermo Scientific™), purified using the QIAGEN, cleaned with the DNA Clean & Concentrator™-5 Kit (ZYMO Research) and treated with Shrimp Alkaline Phosphatase (rSAP) (NEB) to remove 5′- and 3′- phosphates. Inserts and vectors were concentrated using the DNA Clean & Concentrator™-5 Kit (ZYMO Research) and ligated using the Quick Ligation™ kit (NEB). Cloning resulted in one orientation of the MER52, and, therefore, another PCR was performed on generated plasmids using primers with restriction site for size-oriented cloning (forward 5′-3′: AGATGAGCTCCCATACCCTATGAAAGCTGGTC, reverse 5′-3′: CCTCAGATCTGGGAGATTGTACCTTGATGAC). Amplicons were cleaned using the DNA Clean & Concentrator™-5 Kit (ZYMO Research) and, together with the pGL4.12[luc2CP]SV40 plasmid, digested using corresponding restriction enzymes (SacI and BglII, Thermo Scientific™). Inserts and plasmid were purified, cleaned, and ligated as described above. Sanger sequencing was performed to confirm correct cloning.

mESC Cell Culture and Transfection and Luciferase Assay

The mouse embryonic stem cell and luciferase assay protocols followed were published in van Bree et al. (2022).

Insertion and Deletion Analysis at Gene Promoter KZNF Binding Summits

The DNA sequence the size of average peak of each KZNF was extracted from the UCSC reference assemblies around the ZNF519, ZNF441, and ZNF468 summits for human, rhesus, green monkey, and marmoset. Alignments were made between the human, rhesus, green monkey and marmoset using the European Molecular Biology Open Software Suite (EMBOSS):6.6.0.0 stretcher tool (Rice et al. 2000) to assess mutations specific to the human lineage. Alignments with indels specific to the human lineage were confirmed with a multiple sequence alignment using ClustalW on the ebi online portal (https://www.ebi.ac.uk/Tools/msa/clustalo/) (Sievers et al. 2011). The remaining human alignments were coded 1 = no difference, 2 = insertion in human sequence, 3 = substitution, and 4 =deletion in human sequence. The coded files were then visualized in R using the heatmap2 function from the gplots package using the default clustering method (Warnes et al. 2016).

KZNF Summit Conservation Score Analysis

PhyloP conservation data was accessed from UCSC genome browser and intersected with KZNF summit coordinates ±100 bp and transcription factor summits ±100 bp using Python (see supplementary script, Supplementary Material online). The transcription factor summits used for this analysis were taken from ENCODE (ENCODE Project Consortium 2004) using the hg19 narrow peaks files and selecting all peaks with a signal value greater than 100, in cases where the average signal value was lower, the top 2,500 peaks were selected. This analysis was repeated for control regions 500 bp downstream of all summits ±100 bp. The data was then compiled into tables for each KZNF, transcription factor, and control region from which averages at each base location relative to the summit were calculated across all binding sites for each KZNF. These averages were visualized in a heatmap made using Python (see supplementary script, Supplementary Material online).

Guide RNA Design and Cloning

gRNAs to KO exon 1 of ZNF519 (upstream: ATGCTAAAAAAATGACCCCT; downstream: ATGTATGGGAGCGTAGAAGT), exon 1 of ZNF441 (upstream: AAATCAGGGGATAGCTCCAC; downstream: CTGGTGTCCCGACGCGTGAG), and exon 1 of ZNF468 (upstream: GCCCTTTCTGGGCGGAACGT; downstream: AAACCCTCTTGTGATCGTGT) were designed using Benchling (Biology Software), with masked regions included and protospacer adjacent motif (PAM) sequence set to NGG in the search. CHOPCHOP (Labun et al. 2016), CRISPR design (http://crispr.mit.edu/), and the BLAT tool of the UCSC Genome Browser (Kent 2002) were used for verification of the efficiency and specificity of the gRNAs. Complement gRNA oligos were ordered with CACCG added at the 5′ end while reverse-complement oligos contained AAAC at the 5′ and C at the 3′ end, to facilitate cloning. Equal amounts of complement and reverse-complement gRNA oligos were combined in annealing buffer (10 mM Tris pH7.5–8.0, 50 mM NaCl, 1 mM Ethylenediaminetetraacetic acid (EDTA)) and annealed using a thermocycler. The pX330-U6-Chimeric_BB-CBh-hSpCas9 plasmid (Addgene) was digested using BbsI (Thermo Scientific™) and cleaned using the DNA Clean & Concentrator™-5 Kit (ZYMO Research). gRNAs were ligated into the pX330 plasmid using the Quick Ligation™ kit (NEB).

KO of KZNFs in hESCs

KO of ZNF519, ZNF441, and ZNF468 in hESCs was performed as described previously by our lab (Haring et al. 2021). ZNF441 and ZNF468 KO and WT lines were transfected with pX330 plasmid + guide RNAs. WT clones were selected from clones transfected with the complete CRISPR construct which did not show a deletion as in Haring et al. (2021). ZNF519 WT hESCs, cells were transfected using the pX330 plasmid without guide RNAs. Initial genotyping of hESCs was performed similarly as described above according to the protocol of Hendriks et al. (2015). Three KO and three WT hESC clones were analyzed by RNA sequencing for ZNF519 and ZNF468. Due to the low expression of ZNF441 in hESCs, this KO was validated in the cortical organoids.

Primers used for the genotyping were:

• ZNF519 F: GCCTAATAAGGGCGTTTGTG; R: GAAATACAAAAAAAAAGAGGTGTTC

• ZNF441 F: CCAGACTGGTCTCGAATTTCT; R: GCAGAAGAATGCGGTTTCTT;

• ZNF468 F: CCTTCGTCGCAAAGATGCA; R: GGATGTCTCTGAAGCTGAGCACT

Cortical Organoid Culture

Cortical organoids were grown from WT or KO hESCs based on the methods of Eiraku et al. (2008). In short, hESCs colonies were grown in a 10 cm dish on mitomycin C-treated mouse embryonic fibroblasts (MEFs, GlobalStem) in hESC medium (Dulbecco's Modified Eagle Medium, DMEM-F12 (Gibco) supplemented with 20% KO Serum Replacement (Gibco), 100 U/ml penicillin/100 µg/ml streptomycin (Gibco), 2 mM GlutaMAX (Gibco), 1× MEM Non-Essential Amino Acids solution (Gibco), 100 µM 2-mercaptoethanol (Gibco)) with 8 ng/ml fresh basic fibroblast growth factor (bFGF, Sigma). The medium was changed to differentiation medium (99% hESC medium supplemented with 1 mM sodium pyruvate (Gibco)) and colonies with a diameter of 1–2 mm were detached from the plate using a cell lifter (Corning) adjusted to approximately 2–3 mm width. Lifted colonies were collected in 5 ml medium, transferred to a 60 mm ultra-low attachment dish (Corning), and embryoid bodies were formed overnight at 37 °C, 5% CO₂. The next day, medium was refreshed for differentiation medium with freshly added 3 µM IWR-1-Endo, 1 µM Dorsomorphin, 10 µM SB-431542 hydrate, and 1 µM Cyclopamine hydrate (day 0 of differentiation), which was repeated every other day. Organoids were placed on a rocker on days 3–4, to enhance growth and prevent the merging of organoids. From day 18 onward, Neurobasal/N2 medium (Neurobasal (Gibco) supplemented with 100 U/ml penicillin/100 µg/ml streptomycin (Gibco), 2 mM GlutaMAX (Gibco), 1× N-2 supplement (Gibco)) supplemented with 1 µM Cyclopamine hydrate was used for growth or organoids. From D24, no inhibitors were added anymore to the medium until organoids were harvested at day 35 and day 5. For organoid formation, one KZNF-KO line was used for each KZNF; For the ZNF441 and ZNF468 organoids, two replicates (each replicate containing >10 organoids) were taken from two independent batches of organoids made at different times for a total of four replicates for each condition. For the ZNF519 organoids, three replicates (each replicate containing >10 organoids) were taken from a single batch of organoids for each condition.

Overexpression ZNF519 HEK293 Cells

HEK293 (ATCC) cells were grown in DMEM, high glucose, GlutaMAX™, supplemented with 10% heat inactivated fetal bovine serum (HIFBS Gibco™), and 100 U/ml pen/strep (Gibco™). 24 h before transfection, cells were plated on a 60 mm dish at a density of 50.000 cells/cm² to ensure 70–90% confluency at the time of transfection. Transfection was performed in a complete growth medium without pen/strep using Polyethylenimine (PEI) (Polysciences) for six hours with 6.4 ng pCAGEN-ZNF519 (Jacobs et al. 2014) or pCAGEN-empty vector (Addgene #11150), combined with 335.6 ng pCAGEN-GFP (green fluorescent protein). Transfected cells were grown in a complete growth medium for 48 h, with medium refreshment after 24 h.

Before Fluorescence-activated cell sorting (FACS) sorting, cells were washed with warm phosphate buffered saline (PBS) and harvested by incubation for five minutes at 37 °C, 5% CO₂ in 0.25% Trypsin, 0.5 mM EDTA. Trypsinization was deactivated by addition of a complete growth medium, after which cells were pelleted and resuspended in 500 µl FACS buffer (PBS supplemented with 3% HIFBS, 0.5 mM EDTA). 300.000 GFP+ cells were sorted in resuspension buffer (0.5 mM EDTA in PBS), and centrifuged. Supernatant was removed and samples were ready for RNA isolation.

RNA Isolation and Sequencing

For ZNF519 overexpression and ZNF519, ZNF441, and ZNF468 KO experiments, RNA was isolated in 400 µl TRIzol Reagent (Invitrogen™) according to manufacturer's recommendations. Potential DNA contamination was removed with DNAseI (Roche) and samples were cleaned using the DNA Clean & Concentrator™-5 Kit (ZYMO Research). Libraries were prepared with the TruSeq Stranded Total RNA (Illumina) with Ribo-Zero ribosomal RNA depletion, and sequenced paired-end, 75 bp on a NextSeq 550 system (Illumina) by molecular analysis department (MAD): Dutch Genomics Service & Support Provider (Swammerdam Institute for Life Sciences, Amsterdam).

RNA-Seq Data Analysis

The public Freiburg Galaxy server (Goecks et al. 2010) (usegalaxy.eu) (Afgan et al. 2018) was used for processing data. Adapters were removed and reads were trimmed with trimmomatic (Bolger et al. 2014) version 0.36.5 for paired-end reads (ILLUMINACLIP TruSeq3 paired-end), cutting if the average per base quality in a four-base sliding window was below 20, dropping reads below 30 bases. Mapping of reads was performed with HISAT2 (Kim et al. 2019) (Galaxy Version 2.1.0 + galaxy3) against the built-in reference hg19 Full genome. Reads were assigned to gencode V19 and rmsk features using featureCounts (Liao et al. 2014) (Galaxy Version 1.6.3) with -p, -d 75 -D 900 -B -C. Output was analyzed using DESeq2 (Love et al. 2014) (Galaxy Version 2.11.40.3) with default settings. Coverage tracks were generated using bamCoverage (Galaxy Version 3.0.2.0, with deepTools2 (Version 3.0.2) and samtools (Version 1.7)) from the deeptools2 package (Ramírez et al. 2016) (bin size 1) and scaled on UCSC with a scaling factor based on the number of uniquely assigned reads from featureCounts, or scaled using bamCoverage for merging of replicates (HEK293). Scaled coverage tracks were merged using wiggletools (Zerbino et al. 2014) mean, and wig files were transformed into bigwig files using the wigToBigWig script (http://hgdownload.soe.ucsc.edu/admin/exe/). DESeq2 output and read counts available in supplementary Data Files 1 and 2, Supplementary Material online.

KZNF Target Analysis OE and KO

Peak data provided by Imbeault et al. (2017) was extracted from NCBI (GSM2466578). Peaks with a MACS score >500 were taken as “high-confident ZNF519-bound” regions while all peaks provided by Imbeault et al. (2017) were taken as “ZNF519-bound” regions. TSSs were generated from gencodeV19 annotation by selecting the first bp of protein-coding is known transcripts. The center of the MACS peaks was calculated to perform an overlap where at least 50% of the KZNF peak is in the promoter region. Using bedtools ClosestBed, the 50 closest TSSs to the MACS peaks were selected and reported with the distance of the center of the MACS peak to the TSSs. These were subsequently filtered to only keep those peaks that overlap with the promoter regions of TSSs as defined by a window of 5,000 bp upstream and 1,000 bp downstream of the TSS.

Log2FC of bound and unbound genes was compared using R (R Core Team 2019) and visualized using ggplot2 (Wickham 2016). The number of bound and unbound genes expressed differed greatly and so the 95% confidence interval (CI) of the median log2FC of unbound genes was calculated by 10.000 times bootstrapping the median of a random set of unbound genes with a similar sample size as the bound genes. For comparison of log2 fold change of expressed target genes versus nontarget genes, a Wilcoxon rank sum test with continuity correction was performed.

Supplementary Material

Supplementary data are available at Genome Biology and Evolution online (http://www.gbe.oxfordjournals.org/).

Author Contributions

Conceptualization, F.M.J.J., E.J.v.B., and G.F.; Methodology, F.M.J.J., E.J.v.B., G.F; Investigation & Validation, G.F., E.J.v.B., S.R., L.J.M.J., L.M.; Data Curation, F.M.J.J., E.J.v.B., G.F. ; Writing—Original Draft, G.F.; Writing—Review & Editing; F.M.J.J., E.J.v.B., G.F.; Visualization, G.F., E.J.v.B., S.R.; Supervision, F.M.J.J.; Project Administration, F.M.J.J.; Funding Acquisition, F.M.J.J.

Data Availability

Raw sequence reads and processed gene expression values from this study have been submitted to the NCBI Sequence Read Archive (SRA; http://www.ncbi.nlm.nih.gov/sra) under accession number PRJNA830860 and to the NCBI Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo) under accession number GSE201315, respectively.