Author response: Binary outcomes of enhancer activity underlie stable random monoallelic expression

Article Figures and data Abstract Editor's evaluation Introduction Results Discussion Materials and methods Appendix 1 Data availability References Decision letter Author response Article and author information Metrics Abstract Mitotically stable random monoallelic gene expression (RME) is documented for a small percentage of autosomal genes. We developed an in vivo genetic model to study the role of enhancers in RME using high-resolution single-cell analysis of natural killer (NK) cell receptor gene expression and enhancer deletions in the mouse germline. Enhancers of the RME NK receptor genes were accessible and enriched in H3K27ac on silent and active alleles alike in cells sorted according to allelic expression status, suggesting enhancer activation and gene expression status can be decoupled. In genes with multiple enhancers, enhancer deletion reduced gene expression frequency, in one instance converting the universally expressed gene encoding NKG2D into an RME gene, recapitulating all aspects of natural RME including mitotic stability of both the active and silent states. The results support the binary model of enhancer action, and suggest that RME is a consequence of general properties of gene regulation by enhancers rather than an RME-specific epigenetic program. Therefore, many and perhaps all genes may be subject to some degree of RME. Surprisingly, this was borne out by analysis of several genes that define different major hematopoietic lineages, that were previously thought to be universally expressed within those lineages: the genes encoding NKG2D, CD45, CD8α, and Thy-1. We propose that intrinsically probabilistic gene allele regulation is a general property of enhancer-controlled gene expression, with previously documented RME representing an extreme on a broad continuum. Editor's evaluation Kissiov et al., show that enhancers can play an instructive role in controlling stable random monoallelic expression (RME). In order to do so, they initially focus on a limited set of natural killer (NK) receptor genes that are subject to RME, which they investigate using several in vivo genetic models. Furthermore, they show that RME can be considerably more prevalent than previously thought, applying to other gene function than receptors and independently of promoter CpG content. Finally, they provide evidence that enhancer strength and/or number might influence the extent of RME, by affecting the probability of promoter activation. Overall, this is a highly relevant manuscript with major implications in gene regulation and enhancer biology and, thus, of broad scientific interest. https://doi.org/10.7554/eLife.74204.sa0 Decision letter Reviews on Sciety eLife's review process Introduction In most cases, both alleles of autosomal genes are co-expressed. In recent years random monoallelic expression (RME) has emerged as an important exception that may apply to ~0.5–10% of expressed genes in a given tissue and has been characterized as the autosomal analog of X-inactivation (Gendrel et al., 2016). In RME, different cells of a given type express only one allele, both or neither, and this expression pattern is mitotically stable. Notably, RME genes do not share an overarching unifying feature or function (Deng et al., 2014; Eckersley-Maslin and Spector, 2014; Gendrel et al., 2016; Gimelbrant et al., 2007; Reinius et al., 2016; Xu et al., 2017), and the biological role for RME is in most cases not known. RME is distinct from X-inactivation, genomic imprinting, and allelic exclusion of antigen receptor and odorant receptor genes, in that biallelic expression occurs at an appreciable frequency, and expression is largely stochastic rather than imposed by strict feedback regulatory mechanisms (Gendrel et al., 2016). The molecular determinants of RME are poorly understood, in part because of difficulties analyzing single primary cells (Reinius and Sandberg, 2015). Chromatin analysis of RME alleles in primary populations has not been possible due to the difficulty of isolating pure cell populations ex vivo with defined RME expression patterns (Gendrel et al., 2016; Xu et al., 2017). As a result, previous analyses have been limited to clonal cell lines derived from F1 hybrids, where allelic expression is known and clonally stable (Eckersley-Maslin et al., 2014; Gendrel et al., 2014; Levin-Klein et al., 2017; Xu et al., 2017). These analyses revealed that enhancers of RME genes are constitutively accessible irrespective of gene or allelic expression status, whereas promoters are accessible only at active alleles (Levin-Klein et al., 2017; Xu et al., 2017). Therefore, promoter accessibility, rather than enhancer opening and activation, might be the ‘gatekeeper’ of RME, whereas enhancers, being constitutively open, were proposed to permit rather than impose expression of RME alleles (Xu et al., 2017). Enhancers may play more than a permissive role in RME, however, in light of evidence that enhancers primarily influence the probability of mitotically stable expression, rather than the amount of expression per cell (Walters et al., 1995; Weintraub, 1988). In fact, deletions of enhancers resulted in mitotically stable gene variegation in both cell lines and normal tissues—notably Igh in B cell hybridomas and Cd8a in primary thymocytes, among others (Ellmeier et al., 2002; Garefalaki et al., 2002; Ronai et al., 1999; Sleckman et al., 1997; Walters et al., 1995; Weintraub, 1988; Xu et al., 1996). Collectively, these data support the binary or ‘on/off’ model of enhancer action (Blackwood and Kadonaga, 1998; Fiering et al., 2000), where an increase in enhancer activity at a genetic locus results in an increase in the probability of gene expression, rather than an increase in expression per cell. Conversely, weak or reduced enhancer activity results in a lower likelihood of expression, but the cells that express the gene express a similar amount of gene product. We reasoned that the binary action of enhancers—when limiting—might be a driving principle of RME, and sought to test this in examples of RME with a clear biological purpose: the Klra genes (which encode the Ly49 family receptors) and the Klrc1 gene (which encodes the NKG2A receptor) (Chess, 2012; Eckersley-Maslin and Spector, 2014; Gendrel et al., 2016). These genes, clustered in a ~ 1 Mb stretch of the NK gene complex (NKC) on chromosome 6 in mice, encode cell surface receptors expressed by NK cells that bind MHC I molecules. They are expressed in a variegated (Raulet et al., 1997; Yokoyama et al., 1990), monoallelic (Held et al., 1995), stochastic and largely mitotically stable fashion (Raulet et al., 2001), resulting in subpopulations of NK cells that express random combinations of the receptors and consequently exhibit distinct reactivities for cells expressing different MHC I molecules. Regulation of each gene allele is independent, and expression of one Klra gene has minimal effects on expression of others (Tanamachi et al., 2001). While a clear biological purpose for RME at many genes is lacking, RME at Klra genes underlies the basis of the ‘missing self’ mode of NK cell target recognition (Kärre et al., 1986). Furthermore, the system represents a powerful in vivo genetic model of RME, where allelic expression states can be easily assessed at the population level in primary cells using allele-specific antibodies that we and others previously generated (Tanamachi et al., 2001; Vance et al., 2002), circumventing previous technical limitations to studying RME in single primary cells. Importantly, competition between Klra genes for interaction with a shared enhancer or locus control region is not required for variegation of Klra genes, as a Klra1 (encodes Ly49A) genomic transgene ectopically integrated in different genomic sites was usually expressed with a frequency similar to that of the native Klra1 gene (~17% of NK cells) (Tanamachi et al., 2004). We previously identified a key DNase I hypersensitive element, Klra1Hss1, ~ 5 kb upstream of the Klra1 gene that is conserved in other Klra genes and required for expression of the Klra1 transgene (Tanamachi et al., 2004). Our central hypothesis is that enhancers, rather than simply being permissive for RME, both limit and directly control the probability of expression of Klra genes—and RME alleles generally—in a stochastic and binary fashion. Binary enhancer action, when limiting, may represent a causal mechanism of RME, explaining the pervasiveness of RME across genes and cell types. We have carried out genetic dissection and population analyses to demonstrate that enhancers control the probability of allelic expression and have provided a more general model of the role of enhancers in RME as well as in other developmentally regulated genes. Results Elements upstream of the Klra, Klrc, and Klrk family genes are transcriptional enhancers with activity in mature NK cells Klra family genes are expressed in a mitotically stable RME fashion by NK cells. Each harbors an accessible chromatin site (Hss1) ~5 kb upstream of the transcription start site (TSS) (Figure 1A and B; Figure 1—figure supplement 1A). We noticed that related NK receptor genes, including the variegated Klrc1 gene (encodes NKG2A) and the Klrk1 gene (encodes NKG2D and is expressed by ~all NK cells), harbor similar elements which we named Klrc15′E and Klrk15′E, respectively (Figure 1A, B; Figure 1—figure supplement 1A). All Hss1 and 5′E elements are bound by a similar suite of factors including Runx3, T-bet, and the enhancer-associated acetyltransferase p300 (Figure 1—figure supplement 1A). Figure 1 with 2 supplements see all Download asset Open asset The Klra1Hss 1, Klrc15′E and Klrk15′E elements display chromatin features of enhancers. (A) ATAC-seq and H3K4me1:me3 log2 ratio ChIP-seq data of relevant NKC genes in primary NK cells; red denotes positive me1:me3 ratios (enhancer-like) while blue indicates negative values (promoter-like). Approximate gene locations are indicated (bottom). Standard gene names (Klr nomenclature) are indicated followed by names derived from the gene products (Ly49 or Nkg2) for reference. Gray ovals represent additional undiscussed Klra genes. Vertical yellow bars and arrows denote the positions of the Hss1 and 5′E enhancers at the indicated genes. Data are sourced from ref (Lara-Astiaso et al., 2014). Normalized data ranges are indicated on the left. (B) Normalized ChIP-seq and ATAC-seq results (sourced from Lara-Astiaso et al., 2014), showing enhancer and promoter histone modifications at Klra1, Klrc1, and Klrk1. Approximate locations of sgRNAs used in this study to delete enhancers are shown. All datasets are presented with the same vertical scale across sub-panels, which are indicated in normalized signal per million reads (SPMR) in the left sub-panel. (C) Heatmaps depict 51,650 ATAC-seq peaks in primary NK cells (excluding peaks ranking in the bottom 5% for either H3K4me1 or H3K4me3) ranked according to H3K4me1:me3 ratio of average ChIP-seq signal calculated over a 2kb window centered on the ATAC-seq peak midpoint. The indicated data are displayed over these peaks in each heatmap. The locations of selected NK receptor gene Hss1, 5′E and promoter elements within the me1:me3 ranking are shown. H3K4 methylation data are sourced from ref (Lara-Astiaso et al., 2014) while p300 is sourced from ref (Sciumè et al., 2020). Figure 2 with 1 supplement see all Download asset Open asset Klrc15′E and Klra7Hss1 are constitutively accessible, while promoters are accessible only at expressed alleles. (A) FACS plot depicting splenic NK cells from a (B6 x BALB/c)F1 hybrid mouse stained with allele-specific antibodies, allowing separation of NK cells expressing both, either, or neither NKG2A allele. (B) (left) Normalized ATAC-seq data generated from the 4 cell populations depicted in (A) aligned to the mm10 reference genome. (right) Allele-informative reads were binned according to chromosome of origin, and displayed as signal mapping to the B6 or BALB/c chromosome. The Klrc15′E enhancer and promoter (Pro.) are boxed (dotted line). Vertical data range in SPMR is indicated for each track. (C and D) Data are as in (A and B), but using an allele-specific staining protocol with respect to the Ly49G2 receptor. Klra7Hss1, Klra7Hss5 and the dominant TSS (Pro3, Gays et al., 2011) are boxed. (E) CUT&RUN data depicting each of 4 indicated histone modifications at the Klra7 gene in IL-2 expanded NK cells sorted to express neither ‘N’ or both ‘B’ alleles of Klra7. Negative control CUT&RUN data were generated using a mouse IgG2aκ (cIgG) antibody, and a 50:50 mixture of IL-2 expanded NK cells that expressed the B6 or BALB/c alleles. These data are displayed in the bottom track in each sub-panel. The ATAC-seq patterns are shown for reference above each analysis; Klra7Hss1 is denoted as ‘1’, Klra7Hss5 is denoted as ‘5’. Arrows depict the locations of the dominant Pro3 TSS. All ATAC-seq and CUT&RUN data within a sub-panel are presented with the same vertical scale. The KlraHss1 elements were hypothesized to serve as upstream bidirectional promoters active only in immature, developing NK cells (which do not yet express Ly49s); it was proposed that the direction of transcription predetermines the subsequent on or off state of the gene in mature NK cells, which is driven by a distinct downstream promoter (Saleh et al., 2004). Recent analysis of Klra gene expression in cell lines suggested instead that the KlraHss1 elements are transcriptional enhancers (Gays et al., 2015), and that the Hss1 transcripts represent enhancer RNAs (eRNAs). Those conclusions were in turn contested in a subsequent paper (McCullen et al., 2016). To address this issue in vivo, we analyzed chromatin features associated with the Hss1 elements in primary NK cells with published ChIP-seq data generated in mature primary splenic NK cells, using the H3K4me1:me3 ratio as an indicator of regulatory element identity (Calo and Wysocka, 2013). The Hss1 and 5′E elements are all enriched in H3K4me1 relative to H3K4me3 (Figure 1A), indicating enhancer identity. The putative NK receptor gene enhancers all ranked in the top 32% of ATAC-seq accessible peaks with respect to the H3K4me1:me3 ratio. In contrast, known promoters of the respective genes ranked in the bottom 21% (Figure 1C). In a deeper analysis, we independently defined enhancers and promoters in mature NK cells. NK cell promoters were defined as previously annotated mouse promoters from the EDPNew database (Dreos et al., 2017) enriched in H3K27ac in NK cells, and enhancers were defined as ATAC-seq peaks bound by the p300 histone acetyltransferase that do not overlap with the promoter list. Enhancers defined in this manner were highly skewed to high H3K4me1:me3 ratios, and promoters to low ratios (Figure 1—figure supplement 1B). All Hss1 and 5′E elements were classified as enhancers based on the p300-bound enhancer dataset (Figure 1—figure supplement 1B). These findings support the conclusion that the Hss1 and 5′E elements elements are enhancers in primary mature NK cells. To test whether the Klra7Hss1 (Klra7 encodes Ly49G2) and Klrc15′E enhancers are required in mature NK cells, we adapted a CRISPR/Cas9 nucleofection protocol developed to edit primary human T cells (Roth et al., 2018; Figure 1—figure supplement 2). We used NK cells from (B6 x BALB/c)F1 hybrid mice and sorted NKG2AB6+ or Ly49G2B6+ cells using B6-allele reactive monoclonal antibodies against each receptor (Tanamachi et al., 2001; Vance et al., 2002) in order to follow the fate of a single allele in each case (Figure 1—figure supplement 2A). Editing efficiencies of NK cells were lower than that of T cells, resulting in only 30% or fewer cells with disruption of the control Ptprc locus encoding CD45 (Figure 1—figure supplement 2B). Targeting Klrc15′E increased the percentage of NKG2AB6-negative cells from ~10% to ~20%–40%. (Figure 1—figure supplement 2C-E), in line with our theoretical maximum editing efficiency. Similarly, targeting Klra7Hss1 resulted in marked loss of Ly49G2B6 expression, with minimal (<5%) loss of expression in non-targeting or non-nucleofected (no zap) control conditions (Figure 1—figure supplement 2F-H). These data show that Klrc15′E and Klra7Hss1 play key roles in maintaining expression of active alleles in mature NK cells, arguing against the proposal that Hss1 elements are only required in immature NK cells (Saleh et al., 2004; Saleh et al., 2002). The results provide functional evidence that Hss1 functions as an enhancer in mature NK cells. The Klra7Hss1 and Klrc15′E enhancers are constitutively accessible Analysis of bulk NK cells did not reveal a correlation between the gene expression frequency of an NK receptor gene and the accessibility, TF occupancy, or H3K27ac modifications of Hss1 and 5′E enhancers (Figure 1—figure supplement 1A). This lack of concordance raised the possibility that these enhancers were similarly active and occupied by TFs upstream of both silent and active alleles, as has been observed for RME genes genome-wide in F1 clones (Levin-Klein et al., 2017; Xu et al., 2017). It has not previously been possible, however, to address this issue for an RME gene in freshly isolated ex vivo cell populations. To purify populations of cells expressing different alleles of Klrc1, we stained (B6 x BALB/c)F1 NK cells with allele-specific antibodies (Vance et al., 2002), allowing us to sort and perform ATAC-seq on NK cell populations expressing all four configurations of alleles: expressing both alleles of Klrc1, only B6, only BALB/c, or neither (Figure 2A and B). While the cells expressing both alleles and those expressing only the B6 allele are closely juxtaposed in the cytometry plots, post sort analysis demonstrated that they could be efficiently separated (Figure 2—figure supplement 1A), consistent with previously published data where allelic expression was confirmed by mRNA analysis (Rogers et al., 2006). SNPsplit (Krueger and Andrews, 2016) analysis of reads demonstrated that the enhancer element Klrc15′E was accessible on both active and inactive alleles in all four populations, whereas the Klrc1 promoter was accessible only at active alleles (Figure 2B). We used a similar allele-specific staining protocol (Tanamachi et al., 2001) to sort and analyze cells expressing either, both or neither Ly49G2 allele (Figure 2C, D). The Klra7Hss1 enhancer was accessible on both active and inactive alleles in all four populations, whereas the dominant promoter Pro3 (Gays et al., 2011) was accessible only on the active allele (Figure 2D). Notably, the Klra7 gene harbors a second minor enhancer element, Klra7Hss5 (Figure 1C; Figure 1—figure supplement 1B), which was similarly accessible at all alleles (Figure 2D). These data demonstrated that enhancers within the Klra and Klrc gene families behave similarly to those of other RME genes analyzed in F1 hybrid clones (Levin-Klein et al., 2017; Xu et al., 2017), exhibiting an accessible configuration whether or not the gene was expressed. Importantly, this analysis further validated the NK receptor genes as a model for RME. While initially surprising, the decoupling of enhancer and promoter accessibility seen at NK receptor genes and other RME loci is consistent with a binary model of enhancer action, where enhancer activation occurs in all cells of a given type and acts stochastically to raise the binary “on or off” probability of gene expression, rather than regulate the per-cell amount of expression (Blackwood and Kadonaga, 1998; Walters et al., 1995). We extended these observations by analyzing the pattern of active enhancer associated marks at silent and active alleles of Klra7 (which encodes Ly49G2). In order to obtain sufficient cell numbers of rare populations, we expanded NK cells from (B6 x BALB)F1 mice with IL-2 containing medium and then sorted cells that expressed neither (N) or both (B) Ly49G2 alleles and performed CUT&RUN for the enhancer-associated H3K4me1/2/3 and H3K27ac modifications. The Klra7 promoter and gene body displayed striking enrichment of H3K4me2/3 and H3K27ac in cells that expressed both Klra7 alleles, and as predicted lacked these modifications in cells where neither allele was expressed (Figure 2E). Notably, the Klra7Hss1 and Klra7Hss5 enhancers displayed equal enrichment of H3K27ac in cells expressing both alleles or neither (Figure 2E). As H3K27ac delineates active as opposed to poised enhancers (Calo and Wysocka, 2013), these data suggest constitutive enhancer activation on both silent and active alleles. These data are consistent with previous results demonstrating that enhancer-derived transcripts are produced at Hss1 elements in cells that do not express the target Klra gene (Gays et al., 2015). Thus, whether measured by accessibility, H3K27ac enrichment or eRNA production, Hss1 activity may be decoupled from target Klra expression status. and Klrc15′E are required for gene expression in vivo, and in We the for in the locus by it in the B6 CRISPR/Cas9 editing (Figure Figure supplement 1A, B). mice lacked expression (Figure Figure supplement 1B), but expression of other Ly49 receptors was (Figure supplement 1B), the that the variegated NK receptor genes are regulated and independently of each mice displayed an percentage of cells (Figure and Figure supplement 1B). Figure with 2 supplements see all Download asset Open asset The and Klrc15′E enhancers are required for gene (A) of sgRNAs used to delete in the B6 NK cell ATAC-seq are displayed for reference with the vertical data scale in SPMR staining of the indicated deletion of staining are depicted in In data are from Data as in for Data in are from with the allele (Figure supplement and were in analysis of the allele using with multiple As with Klra1Hss1, deletion of both allelic of Klrc15′E in the NKG2A expression, and mice displayed a reduced frequency of NK cells (Figure Figure supplement D). the activity of constitutively accessible enhancers of RME genes is in a epigenetic mechanism that RME is not known. We the activity of Klrc15′E and in F1 using F1 hybrid mice between mice and BALB/c mice were generated with F1 The of cells expressing all four combinations of NKG2A alleles were using allele-specific NKG2A antibodies we previously that the Klrc15′E acts only in we calculated the in the of these cells in the The data closely the (Figure supplement Therefore, the constitutively accessible Klrc15′E in and independently of the activity of the other Similarly, in the BALB/c allele was when expression of the B6 allele was (Figure supplement enhancer in the Klra7 locus to the high expression frequency of Klra7 the Klra1 and Klrc1 gene loci harbor only a single site (Figure 1A and B), and are on those enhancers for expression (Figure analysis of the role of and Klrc15′E in the RME of target genes. We reasoned that analysis of an RME NK receptor gene with multiple enhancers could reveal the role of enhancer strength in expression We in with the binary that the of multiple enhancers, enhancer activity at loci is resulting in RME. We predicted that enhancer activity further by a enhancer in a natural RME gene but not gene expression The Klra7 locus is expressed by of NK cells and both an Hss1 element and constitutively accessible Klra7Hss5 (Figure 2D). the region of the highly related Klra1 gene, which is expressed by only of NK cells, is accessible and active (Figure Figure 4 with 1 supplement see all Download asset Open asset minor enhancer Ly49G2 expression (A) Normalized ATAC-seq of Klra1 and Klra7 in bulk NK cells; the vertical data range in SPMR is displayed on each track. Hss1 and enhancers and the Pro3 promoter are sgRNAs used to alleles are shown (B) Ly49G2 staining of NK cells in the indicated Klra7Hss5 deletion allele, Figure supplement 1A). Ly49G2 and of the positive populations results were with the allele (Figure supplement 1B). using with multiple (E) cytometry of NK cells using and antibodies (right) and a and observed of populations depicted in in F1 mice with the is is or Klra7 allele. were calculated stochastic regulation of alleles Materials and of genetic Data are of All bars represent deletion of in resulted in a percentage of cells to mice with only a minor in expression per cell by of (Figure Figure supplement 1A and B). mice displayed an percentage of cells (Figure Figure supplement 1B). To test whether Klra7Hss5 acts in we to BALB/c The NK cell populations in the F1 mice expressing the Ly49G2B6 alleles were and the populations expressing neither allele or only were in the probabilistic action of Klra7Hss5 in (Figure Thus, the constitutively active enhancer Klra7Hss5 is directly in Ly49G2 expression frequency, and at in the high expression frequency of Ly49G2 in to other receptors including of Klrk15′E is sufficient to stable RME in Klrk1 Our hypothesis that RME of NK receptor genes is by binary enhancer activity that a receptor expressed by all NK cells may be into a variegated receptor by enhancer for by one of multiple associated enhancer We this for the Klrk1 gene encoding the NKG2D which is expressed by all NK cells et al., is related to the Klrc1 and Klra genes, and is on both by suggesting possible regulation by multiple enhancers (Figure 1B). of the site ~5 kb upstream of the Klrk1 gene (Figure Figure supplement 1A, B), resulted in variegated NKG2D expression in (Figure and B). of NK cells expressed NKG2D in The expression level per cell was only and to an extent consistent with largely monoallelic expression expression in some cells of both alleles (Figure and Figure supplement 1C). These data suggest that the primary role of Klrk15′E is to increase the probability rather than the degree of Klrk1 expression, in line with the binary model of enhancer Figure 5 with 2 supplements see all Download asset Open asset Klrk15′E deletion results in mitotically stable RME, recapitulating natural NKG2D staining of from Klrk15′E deletion and an mouse Results are of four with deletion alleles (Figure supplement 1, and D). from mice were with IL-2 for and NK cells, which were expanded in IL-2 medium for analysis NK cells are shown in (E) of splenic NK cells from mice of and to gene and Klrk15′E deletion alleles, results in (E) from and observed of in expression is calculated based on observed in