Characterization of Gene Regulatory Elements in Human Fetal Cortical Development: Enhancing Our Understanding of Neurodevelopmental Disorders and Evolution

Abstract The neocortex is the region that most distinguishes human brain from other mammals and primates [Annu Rev Genet. 2021 Nov;55(1):555–81]. Studying the development of human cortex is important in understanding the evolutionary changes occurring in humans relative to other primates, as well as in elucidating mechanisms underlying neurodevelopmental disorders. Cortical development is a highly regulated process, spatially and temporally coordinated by expression of essential transcriptional factors in response to signaling pathways [Neuron. 2019 Sep;103(6):980–1004]. Enhancers are the most well-understood cis-acting, non-protein-coding regulatory elements that regulate gene expression [Nat Rev Genet. 2014 Apr;15(4):272–86]. Importantly, given the conservation of both DNA sequence and molecular function of the majority of proteins across mammals [Genome Res. 2003 Dec;13(12):2507–18], enhancers [Science. 2015 Mar;347(6226):1155–9], which are far more divergent at the sequence level, likely account for the phenotypes that distinguish the human brain by changing the regulation of gene expression. In this review, we will revisit the conceptual framework of gene regulation during human brain development, as well as the evolution of technologies to study transcriptional regulation, with recent advances in genome biology that open a window allowing us to systematically characterize cis-regulatory elements in developing human brain [Hum Mol Genet. 2022 Oct;31(R1):R84–96]. We provide an update on work to characterize the suite of all enhancers in the developing human brain and the implications for understanding neuropsychiatric disorders. Finally, we discuss emerging therapeutic ideas that utilize our emerging knowledge of enhancer function.


Introduction
A significant proportion of neuropsychiatric disorders, such as autism spectrum disorders (ASD), attentiondeficit/hyperactivity disorder (ADHD), and schizophrenia (SCZ), have their roots in disrupted neurodevelopment [1][2][3][4][5].This is in addition to an assembly of thousands of rare, medical genetic disorders that more severely affect the development of the brain [6,7].Therefore, there has been a keen interest in understanding the molecular mechanisms of human brain development.Further, from an evolutionary standpoint, understanding human cortical development will provide keys to understanding the development of human cognitive specializations [8].
The bulk of the work dissecting mechanisms of cortical development has been done using mice as a model, given their accessibility to genome engineering and the conserved repertoire of most genes across mammals [9].Developmental principles discovered in mice or nonhuman primates (NHP) have been largely validated in human, such as the mechanisms of neuronal migration along radial glia, the laminated structure, and many cell type markers [10][11][12][13].However, there are also dramatic differences between human and mouse brain, as well as human and other primates: the size, neocortical expansion, especially layers II/III, and most importantly, the complex cognitive functions unique to human [14].The origin of these phenotypic differences resides in the noncoding genome: whereas the coding proportion, which accounts for 1-2% of the genome, is highly conserved, the noncoding landscape is highly divergent between species.Within the noncoding genome, enhancers are the most understood class of cis-regulatory sequences, which regulate gene expression by recruiting transcriptional machinery to promoters through longrange chromatin interactions [15].While gene promotors are generally conserved across mouse and human, enhancers are far less so [16,17].
Through decades of work on biochemistry and with the advent of next-generation sequencing technologies, we are now in possession of robust methods to identify enhancers genome-wide [18], as well as high-throughput assays of enhancer function [19][20][21].Using these technologies, large, collaborative efforts have been initiated to map and validate enhancers in the brain, including fetal brain, generating vast data sets [22][23][24][25][26].This is not only important for understanding normal brain development in humans as well as human brain evolution but also to understand the impact of genetic risk for disease.Over the last 2 decades, researchers have used genome-wide association studies (GWAS) to identify many hundreds of loci harboring common genetic variants associated with brain disorders.The vast majority of these GWAS loci lie in non-protein-coding regions whose function is not well understood and until recently, poorly annotated; what appears so far, is that common genetic risk for many childhood and adult onset neuropsychiatric disorders seem to be most enriched in fetal brain enhancers [5,23,25].Moreover, some of the functions of evolutionary divergent sequences between human and other mammal species also reside in these enhancers [27].In the sections following, we will elaborate these points, providing more detailed evidence for these claims and end with the promise of harnessing the understanding of developmental brain enhancers to treat genetic psychiatric diseases.

Transcriptional Control of Cortical Development
Mammalian cortical development is initiated upon specification of neuroepithelium in mammalian ectoderm [28].This process, for which we only have a basic understanding, is under a complex set of cis-regulatory elements and transcription factors (TFs).This initial neuroepithelium is populated with neural stem cells (NSC) that proliferate and differentiate into postmitotic neurons [29].Tangentially across the developing cortex, the NSCs are located in the ventricular zone (thus also referred to as the ventricular radial glia, where they either divide symmetrically to maintain the pool of stem cells or divide asymmetrically to give rise to intermediate progenitor cells or postmitotic neurons [28,30].Notably distinct from mice, primates have an additional outer radial glia (oRG) cell population, which is localized in the sub-ventricular zone and a result of asymmetric division of NSCs, but with retained progenitor potency [31]; this oRG layer is substantially expanded in humans and is predicted to primarily give rise to intratelencephalic (IT) neurons in upper cortical layers, which are also expanded in humans [28].

Neuronal Differentiation
NSCs are characterized by expression of a group of TFs such as SOX2 and PAX6; oRG cells specifically express the TF HOPX [13].As neurogenesis begins, newborn neurons migrate radially and form a laminated structure that is refined over the course of development [11].Formation of the cortical layers occurs in an inside-out fashion, i.e., the earliest-born neurons occupy the deepest layers, with subsequent waves of neurons destined to progressively more superficial layers [32].Each of the cortical layers is defined by a specific set of TFs.For example, at early stages, deep-layer excitatory neurons express TFs SOX5, TBR1, BCL11B, and FEZF2, whereas upperlayer excitatory neurons express TFs SATB2 and NEU-ROD6; a more detailed description of the lamination of neocortex and layer-specific markers is reviewed in Molyneaux et al. [33].Notably, knockout of Fezf2 [34] or Satb2 [35] in mice result in loss of deep layer or upperlayer excitatory neurons, respectively.Although most of the initial studies were performed in mice, most of the major observations have been confirmed in recent, systematic studies in human fetal tissues using single cell RNA-Seq profiling [12,25].Moreover, the early phases of lamination and expression of the corresponding markers are recapitulated in human 3D cortical spheroid models [36], suggesting that the basic functions of the transcriptional programs controlled by these factors are robust and evolutionarily conserved to some extent.

Arealization
Other than neuronal differentiation along the radial direction, the developing cortex in mammals is also arealized along the anterior-posterior and mediallateral axes [37] and in humans, includes the left-right axis.NSCs are exposed to gradients of morphogens, including fibroblast growth factor expressed by the commissural plate, Wnt and bone morphogenetic protein from the cortical hem, and Sfrp1 and transforming growth factor alpha from the antihem [38].Gradients of morphogens induce expression of TFs such as EMX1, EMX2, PAX6, LHX2, and COUP-TFII, and genetic ablation of these factors leads to dramatic changes of the areal map of the cortex [37], suggesting critical roles of transcriptional regulation in specialization of the neurons.On the other hand, the same developmental genes are regulated differently in different areas.For example, DARP-32 is a marker of cortical-thalamic neurons in medial prefrontal cortex (mPFC) [39], but not visual cortex [40].C-Met is enriched in layer 2-3 IT neurons in all sensory cortical regions, but not expressed in layer 2-3 neurons in mPFC; the receptor is expressed in corticothalamic neurons in mPFC, but not in primary sensory cortical regions [40].An interesting question is how to trace brain arealization at later stages, such as differences between prefrontal cortex and V1, back to these early events driven by developmental signaling pathways, so that we can generate better area-specific in vitro models.Increasing transcriptomic information [41] collected throughout development will inform the answers to this question.A common theme of cell type specification in neocortex, including neuronal differentiation and arealization, is distinct expression pattern of TFs, pointing to the delicate underlying transcriptional control.In the following sections, we will discuss how such control might be exerted at DNA level and technological evolution accompanying the work in this area.

Enhancers: Noncoding Elements That Regulate Gene Expression
Transcriptional regulation occurs at the interface of protein and DNA.For most protein-coding genes, their expression is initiated by engagement of RNA polymerase II at promoters.However, in almost every mammalian cell, transcription of individual genes must follow a certain spatial-temporal pattern, i.e., only a specific set of genes can be turned on in a given scenario to fulfill cell type-specific functions.This delicate regulation is largely driven by a class of DNA sequences called enhancers.Four decades ago, the first cellular enhancer was discovered within a mouse immunoglobin heavy chain gene; the enhancer, located from several hundreds to thousands of base pairs up-or downstream of several promoters, elevates expression of a number of beta-globin genes by at least two orders of magnitude [42].Remarkably, the enhancer functions only in lymphocyte-derived cells [42].This example highlights basic properties of enhancers that were subsequently discovered: (1) they are localized in noncoding regions distal to the promoters of their target genes and (2) they often function in a cell type, or stage-specific manner, which is partly driven by the presence of binding sites of TFs expressed in particular cell types.The basic and current model of enhancer function is that the enhancer, through binding to TFs, recruits the transcriptional machinery (mediators, transcriptional co-activators, RNA Pol II, etc.) and brings them to the promoters of its target genes to activate transcription [15].
The study of enhancers entered a new era in the first decade of the 21st century, as a result of a completed human genome project [43] and next-generation sequencing technologies.Earlier biochemical studies supported transcriptional regulatory roles of certain modified forms of histones [44].Specifically, histone acetylation has been shown to drive transcriptional activation [45] through mobilizing local chromatin structure; histone H3, Lysine 4 (K4) methylation positively regulates transcription, as it recruits nucleosome remodelers [46] and histone acetylases [47], whereas Lysine 27 (K27) methylation plays an opposite role by condensing chromatin structure and reducing its accessibility [48].Based on these observations, initial work [49] mapped promoters [50], followed by enhancers [51] in a genome-wide manner using H3K4 methylation and/or acetylation markers.Subsequent sequencing based analysis revealed a "bivalent" chromatin modification pattern in embryonic stem cells, where overlapping H3 Lys4 and Lys27 trimethylation demarcates genes that are temporarily silenced, but poised for activation upon receiving developmental signals [52,53].More accurate characterizations of active enhancers were achieved with ChIP-Seq and integration of H3K27ac and p300 markers [54].Bivalency was also observed in developmental enhancers: although H3K4me1 is a generic marker of enhancers, those with H3K27ac are active, as opposed to the poised (temporarily silenced) enhancers marked by H3K27me3 [55].Aside from histone marks, open chromatin has also been used as a hallmark of all types of cis-regulatory sequences, including enhancers [56] (Fig. 1).Transitioning into the genomic era, open chromatin information was initially exploited with DNase I hyper-sensitivity assays followed by NGS [57].This has been replaced by ATAC-Seq [58], which turned out to be more efficient and technically tractable.Taken together, with extensive development in the past 2 decades, there is a mature set of technologies to detect and profile enhancers and other regulatory sequences genome-wide.Indeed, most of the technologies mentioned above are now available at single cell resolution [59][60][61][62].Since enhancer activity is largely cell type specific, there has been a tremendous effort in annotating noncoding genomes systematically to identify enhancers in specific cell and tissue types, as exemplified by the ENCODE [63], PsychENCODE [24], Epigenetics Roadmap [64,65], and most recently, the 4D nucleome projects [66].
Profiling enhancer activity in specific biological paradigms and in specific tissues or cells is classically done by installing a reporter gene, e.g., GFP or galactosidase, directly downstream of the enhancer sequence in a plasmid, which would then be delivered into a relevant cell or animal model to measure its activity [54,67,68].However, this is not a high throughput method and can be costly when performed in animals.A high throughput version of the reporter assay described above is the massively parallel reporter assay (MPRA), where a pool of reporter constructs is assembled, each expressing a regulatory sequence (e.g., enhancer driving a reporter tagged with a barcode).Each enhancer drives tens to hundreds of unique barcodes to mitigate any transcriptional artifacts introduced by different barcodes [69,70].
It is worth noting that a single cell version of MPRA was recently described [71].Another class of methods, i.e., STARR-seq, uses random [72] or captured [73] genomic DNA instead of barcodes.
Once enhancers are mapped, the next key question is: what genes do they regulate?A popular and predictive method is based on chromatin conformation capture (3C) [74], which later gave rise to 4C [75], 5C [76], and eventually HiC [77], in order of increasing genome coverage.Theoretically, chromatin conformation data can inform the majority of enhancer-promoter loops.However, the method is relatively expensive due to requirement of high coverage of the genome and its resolution is practically limited to about 1 kb [78,79].HiC data need to be combined with other methods since many loops do not result in functional, transcriptional output, and some of their function is not yet known [80].Despite these limitations, the field is moving forward rapidly, with adapted methods that selectively capture interactions from regulatory regions only [81,82], or that allow low input material [83,84], both increasing efficiency substantially.Notably, several single cell-based HiC methods have been recently developed [85][86][87].Beyond identifying individual chromatin loops, HiC data also inform the identification of topologically associating domains (TADs), which are defined by inherent barriers within the nucleus that limit chromatin interaction across domains [88].Thus, the majority of enhancers regulate genes within the same TAD.
Chromatin conformation methods, which are primarily biochemical in nature, can be complemented with other primarily statistical methods to identify gene regulatory interactions.One such method is expression quantitative trait loci (eQTL) analysis [89], where statistical tests are performed to associate allelic differences in noncoding, common genetic variants with gene expression levels in a cis-regulatory window (e.g., +/− 1 Mb from the variant tested).In this way, one can connect enhancers to target genes by identifying genetic variants within them that are predicted to regulate a protein or non-protein-coding gene.Although identification of eQTLs is technically more feasible than biochemical methods, it normally requires a large number of human samples to measure allelic variability and gene expression, which can be difficult to obtain.Lastly, one can correlate the degree of activity, or the degree of chromatin openness, around a promotor with nearby enhancers to identify those most likely to regulate a given gene [23].When combined with eQTL and HiC, this approach can result in very high confidence predictions that can be experimentally validated [5,19].
While both HiC and eQTL analysis are both highly predictive of target genes regulated by an enhancer, the ultimate way to prove the link between enhancers to target genes is through genetic perturbation experiments.Several studies in mice have shown that individual enhancers are required for target gene expression in specific cell types/tissues, or at least required for expression to reach certain levels [5,[90][91][92].There has been significant progress on in this area, primarily CRISPR-based screening technologies, recently reviewed in ref [19].In summary, thanks to the rapid development of functional genomics in the last 2 decades, we have an arsenal of tools to either profile or functionally validate enhancers in high throughput (Fig. 1), which is opening a new window to understanding gene regulatory mechanisms in neurodevelopment.

Developmental Enhancers in the CNS
Just as anatomical, histological, and connectivity mapping have been essential for developing our current understanding of the brain, so are maps necessary for molecular scale data.In this vein, large-scale transcriptional analysis of human fetal brain [93,94] has laid foundational molecular maps that distinguish features associated with developmental stages, brain areas, and cortical layers.Notably, such work is driven by large-scale collaborations exemplified by the Psy-chENCODE [24] and BICCN [95] consortiums.Prominent output of these consortia include ATAC-Seq data generated from germinal zone (GZ) and cortical plate (CP) in mid-gestation [5] and histone ChIP-Seq data generated from a larger cohort of samples across developmental stages [23], as well as ATAC-Seq data generated from various areas in telencephalon during mid-gestation [96].More recent work at the single cell level used scATAC-Seq [25,26], defining cell typespecific enhancers in human fetal brains.In conjunction with single-cell transcriptomic profiling [12,26,41], these efforts comprise a comprehensive, albeit first generation, regulatory map of the first two trimesters of human brain development, since there are few samples available from late fetal (24-38 PCW) stages.
To map fetal brain enhancers to target genes, one might use HiC data generated from the GZ and CP [97], H3K4me3 (marking active promoters) PLAC-Seq data [98] from purified cell types (radial glia, intermediate progenitors, excitatory neurons, and interneurons), as well as eQTLs [1] from mid-gestational fetal brain.For example, Walker et al. [1] identified 7,962 cis-eQTLs linked to 6,546 genes, averaging 1.2 eQTLs per gene, as well as more than 4,000 splice QTLs in 200 distinct fetal brain samples.Based on fetal brain ATAC-Seq data with support from HiC data from similar samples, 3,184 enhancers were mapped to 1,487 genes [5], averaging 2.1 enhancers per gene [5].Both estimations can be considered very conservative, given the limited power from relatively low sample numbers.In general, enhancers outnumber genes by almost an order of magnitude [63]; therefore, it can be estimated that each gene is regulated by 6-10 enhancers on average, combining various scenarios.Below, we expand on how these data inform our understanding of human evolution and disease vulnerability, and how we can harvest such information for therapeutic purposes.A summary of the studies that identified enhancers in human fetal brain and related in vitro models can be found in Table 1.
In parallel to data generation in fetal brain, similar efforts have been carried out for adult human brain [22,[99][100][101][102].Not much has been published comparing these two major stages of human brain development, so we performed some additional analyses for this review to highlight the major differences between epigenomes at these two stages.We identified differentially active enhancers between fetal and adult brains using H3K27ac ChIP-Seq data sets generated by PsychEN-CODE [22,23], which comparable data sets of reliable quality (Fig. 2a).We found that 45% of fetal and 16% of adult enhancers to be specifically enriched at their respective stage; notably, fetal cortex has a substantially higher number of stage-specific enhancers than their adult counterparts (Fig. 2b).Moreover, there is likely substantial difference in the functions of enhancers of the two stages.The predicted target genes (see Methods) of fetal-specific enhancers are most significantly enriched in relevant developmental genes such as "neuron development," "head development," "central nervous system development," and "brain development" (Fig. 2c).In contrast, the adult-specific enhancers are enriched near genes involved in specific cellular functions related to neurons, such as "intracellular transport," "vesicle-mediated transport," "Golgi vesicle transport," pointing to synaptic vesicle functions (Fig. 2c).The distinction between fetal and adult epigenomes is evident, as reflected by both the genome-wide correlation between samples (Fig. 2a) and specific loci, such as SATB2 (Fig. 2d).Result of this analysis points to substantially distinct regulatory landscapes between fetal and adult human brains.This distinction very likely contributes to the critical time window underlying the etiology and intervention neurodevelopmental disorders.The epigenomic information here can serve as a reference to assess the level of maturation of in vitro models of human brain [103], which has becoming increasingly important in studying CNS disorders.It can also be used as a resource to rescue certain genetic disorder affecting proper function of CNS, as will be elaborated in the final part of this review.

Evolution
The major principles of mammalian cortical development have been established primarily in mice, but it is critical to be aware of differences between mice and human.In mice, neurogenesis lasts about 1 week, starting on embryonic day 11 [104], whereas in human it begins at 6 weeks postconception (wpc) and extends through mid-gestation, possibly continuing into the early third trimester [105].Consistent with the extended duration of cortical development, the human neocortex displays the highest number of cortical neurons (around 16 billion) relative to brain mass [106,107].Among primates, the human cortex contains the highest number of neurons, with about twofold more than the chimpanzee and greater than tenfold more than the macaque counterparts [106].Most strikingly, the human cortex also displays the highest relative and absolute number of upper-layer (layers II and III) pyramidal neurons compared with other species, although evidence of this expansion of upper layers is observable in other primates [108].As upper-layer neurons underlie most intracortical projections, expansion of this area is likely to lead to higher levels of cortical-cortical connectivity.
There is limited understanding of the cellular and molecular mechanisms that shape the human-specific aspects of cortical development.One class of hypothesis has to do with the special properties of human NSCs, such as longer lasting proliferation period of human NSCs [109][110][111][112], and a significantly higher abundance of oRG [113], the latter of which gives rise to the expanded layers II/III.There has also been very interesting data showing how human-specific gene [114,115] or paralogs [116] may contribute to cortical development.We will not expand on these subjects here as it has been systematically reviewed elsewhere [14].
Significant efforts have been undertaken to connect human-specific sequences to biological function in the developing brain.Comparative genomics has identified ≈3,000 noncoding genomic sequences that are ultraconserved across vertebrates but that display a significant excess of human-specific substitutions, called human accelerated regions (HARs) [27].Those include 721 HARs [117,118], 992 human accelerated conserved noncoding sequences [119], 1,356 accelerated noncoding conserved sequences (ANCs) [120], and 63 accelerated elements [121], each identified using different methods.
A significant proportion of HARs overlap with brain enhancers by epigenomic annotation [3,[122][123][124] and are located near genes implicated in CNS functions [119].Importantly, enhancer activities of HARs have been measured in vitro using capture MPRA, where 61% of active HARs showed divergent activities between human and chimpanzee sequences [124].Another MPRA study examined all the human-specific nucleotide substitutions in HARs and HGEs (described below) [125].Fetal brain HiC data informed putative target genes of HARs, which are enriched in fetal brain oRG cells as well as adult neurons and astrocytes [3].Interestingly, PPP1R17, a gene with unknown function, but is highly expressed in neural progenitors, is predicted to be a target gene of HAR2635 and specifically expressed in primate developing cerebral cortex [124].CRISPR activation experiment in primary human neural progenitors also validated regulation by HARs of major patterning genes including, GLI2, GLI3, and TBR1 [3].Another report demonstrated HAR5, a predicted enhancer of FZD8, drives reporter activity in mouse fetal brain, with species-specific activity [126].These data suggest an integral role of HARs in regulating transcriptional programs in neural progenitor cells.In addition, among the first HARs identified, HAR1 actually encodes an long noncoding RNA that is expressed specifically in Cajal-Retzius neurons in the developing human neocortex [127].Lastly, a recent study identified enrichment of copy-number variants in HARs in subjects with ASD versus their siblings [123].That being said, HARs account for only a very limited proportion of the regulatory genome and so their overall composite contribution to neuropsychiatric disorders should be understood in this context.However, they do comprise substantial humanevolved elements that may help us understand humanspecific cognitive and behavioral features.
Epigenetic changes that do not involve changes in sequence also occur on the human lineage.For example, there are 8,996 enhancers with gained activity in humans (HGEs) [16].More concretely, these regions display quantitatively higher H3K27ac or H3K4me2 signal in fetal human versus macaque and mouse orthologous sites [16].Originally, these regions were assigned to the closest genes, but HiC data-predicted HGE target genes are mostly distal, and likely not the closest ones, and are enriched in human oRG and adult neurons and astrocytes [3].Interestingly, this pattern matches the cellular enrichment of HARs.An independent study also showed enrichment of HGE target genes in the NOTCH signaling pathway, which is key to neural progenitor self-renewal [98].In general, HARs and HGEs represent human-evolved elements that likely impart new gene regulatory functions in fetal brain, with likely specific roles in the transcriptional networks driving neural progenitor cell proliferation, expansion, differentiation, and lineage decisions.
In this regard, a convergent feature of almost all the epigenomic maps of human fetal brain is enrichment for genetic risk for many childhood and adult onset neuropsychiatric disorders in active elements.Using partitioned heritability analysis, common genetic variants associated with SCZ or ADHD were found to be enriched in open chromatin regions that are more active in GZ relative to CP [5], as well as those specific to newborn and upper-layer excitatory neurons [25].Similar observations were made in studies of fetal brain eQTLs [1], as well as H3K27ac-marked regions [23].It is important to note that enrichment of SCZ GWAS hits is only significant in regions marked by H3K27ac, indicative of active enhancers, in fetal and infant brains more so than adult, and in dorsal frontal cortex but not cerebellum [23].While not all GWAS studies are well powered to generate significant results in partitioned heritability analysis, and that the fetal epigenomic data sets are biased toward early or mid-gestation due to limitations of sample acquisition at later stages, these data suggest that genetic etiology of these disorders impacts early developmental events, specifically neural progenitors and cortical neurogenesis.That is, common genetic polymorphisms that increases risk across multiple neuropsychiatric disorders are enriched in regulatory regions that impact fetal brain development, primarily cortical neurogenesis, and progenitor function.
Other than direct annotation of GWAS hits, epigenomic data can also be used to predict the target genes that are impacted by human genetic variation.In one example, an association of enhancers and promoters was calculated by correlation of ATAC-Seq signals; the predicted target genes of SCZ GWAS hits were enriched broadly in fetal brain, while BIP GWAS target genes were enriched only in oRG [12].In another example investigating the genetic relationships between neuropsychiatric disorders, using a combination of eQTL and HiC resource, target genes of common variants were determined; pleiotropic variants impacting many disorders were found to be highly expressed in prenatal stages, in contrast to disorder-specific variants [143], suggesting that disruption of early developmental processes have profound effects that can result in diverse phenotypes.
A major challenge remains in identifying the actual functional variants from GWAS [19].Using MPRA, we recently surveyed 34 genome-wide significant loci associated with Alzheimer's disease and Progressive Supranuclear Palsy, testing 5,706 variants and identifying 320 functional regulatory variants (frVars) that show significant, allelic effects on gene expression [144].CRISPRi screening and CRISPR/Cas9 knockout experiments were used to further validate target genes of 47 of the frVars with a combination of [144].This work provides a framework for us to systematically characterize the transcriptional regulatory functions of enhancers and noncoding variants associated with diseases.Downstream analysis might include editing individual genetic variants in a relevant context before studying both the transcriptional and physiological responses.This has become feasible with iPSC-based human brain 3D models [145].

Therapeutic Potential
While there is an obvious rationale, as discussed above, to use enhancer information as a window for understanding the biological mechanisms of neuropsychiatric disorders, there is an opportunity to treat genetic diseases directly by modulating enhancer activity.This topic has been thoroughly reviewed elsewhere recently [146].Although most of the relevant research so far was performed in rodents, it is still worth highlighting a few examples relevant to neurodevelopmental disorders.
Most of the genes associated with risk of ASD have mutations that are predicted to knockout one copy of the gene, leading to a reduction in dosage of the gene, despite having another normal copy, so-called haploinsufficiency [147].Theoretically, in the scenario of haploinsufficiency, enhancers can be targeted to activate the wildtype allele of the gene to compensate the loss of the other allele.In a prominent example, a hypothalamus-specific enhancer to Sim1 was activated in mice with rAAV-packaged dCas0-VP64 CRISPRa effector [148].The activation rescued the obesity in animals with haploinsufficiency of Sim1, which encodes a TF.Notably, the activation was specifically in hypothalamus, while Sim1 is highly expressed in both hypothalamus and kidney [148].In a second study of Dravet syndrome which is characterized by severe epileptic encephalopathy, SCN1A haploinsufficiency is the primary known cause.In this case, the promoter of Scn1a was targeted by CRISPRa, both in vitro and in vivo, which rescued both the gene expression level and epilepsy [149].
In the most recent study, CRISPRa was enlisted to activate retinoic acid induced 1 (RAI1) via promotor activation in Smith-Magenis syndrome; rescue by AAV-delivered CRISPRa to brain led to correction of repetitive behavior as well as delayed onset of obesity [150] in a mouse model, which indicates potential modification of the disease progression.None of these have yet been used in human, but these results are very promising.
Other than the examples above, similar strategies can be used to upregulate an alternative gene that has a similar function [151], or downregulate genes whose mutated version results in gain of function [152].Modulating gene expression from enhancers has the advantage of tissue specificity and in theory, a more moderate control of the expression level to avoid overexpression.However, there is only one study so far that modulates gene expression from enhancers to treat disease in model systems [148].This is possibly due to a lack of systematic enhancer resource for each tissue/cell type, especially for mapping enhancer-gene relationships comprehensively.The combination of MPRA and CRISPRi-QTL analysis in relevant cell types which would measure enhancer activity and determine the target genes is one promising high throughput strategy that can be pursued in vitro in human cells or in vivo in animal models.With the development of such functional genomics resources related to human cortical development and improvement of viral and non-viral delivery methods, precision treatment of genetic diseases is becoming feasible.

Conclusions
Understanding the basic biology of cortical development is essential to developing mechanism-based therapies.The work described here shows how this knowledge can be transformed into therapeutics using modern genomics, focusing on understanding of gene regulatory mechanisms in cortical development.These data have elucidated how genetic risk for neuropsychiatric disorders is embedded in the functional, noncoding genome, and how this knowledge informs our understanding of disease etiology.More excitingly, examples have emerged to show how tissue-specific enhancers may be utilized to effectively treat certain genetic diseases.In parallel, basic maps of gene regulation in the fetal human brain have begun to reveal new factors driving the evolution of human-specific features.That HARs are enriched in fetal brain enhancers is consistent with the notion that human-specific aspects of brain development have roots in basic developmental neurobiology.Here, we focus on genetic regulation of human cerebral cortical development.However, it is worth nothing that environment also plays a critical role in this process and more specifically in influencing the transcriptional regulatory events involved, which has been reviewed in [153].An interesting frontier will be to understand how environment impacts enhancer activity.
There still remain important areas of work to be done to further our understanding of gene regulation in treating brain disorders.First, there is a need to refine enhancer maps to specific brain regions and cell types and generate a consensus map over developmental time.Epigenomic profiling of the bulk fetal brain tissue has been successful [5,23,96]; the data generated there can serve as a reference for future maps.The limitation there is that the samples assayed were numerically limited and the data were only generated from bulk tissue [5,23,96].Excitingly, single cell epigenomics work on this area has already started [25,26], pinning down enhancers in each major cell type in the fetal brain.While current studies include a limited number of samples [25,26], it is encouraging to see that multiple brain regions were interrogated, with area-specific regulatory elements identified [25].
It is admittedly difficult to collect postmortem tissues, especially fetal tissues.Therefore, a promising direction is to profile iPSC-derived brain models; since iPSC lines can be derived from easy-to-access tissues such as skin and blood, this strategy in theory can provide unlimited samples.Initial work showed that about 50% of enhancers discovered in iPSC-derived organoids overlapped with those from human fetal brain [25,154].Interpreting these data is challenging, as inconsistencies could either be technical artifacts from in vitro experimental conditions [25], or reflect true biological signal that were not captured in vivo [23].Finally, it is worth noting that in addition to organoids, primary neural progenitor cells derived from human fetal brain also generated cell type-specific, useful insight, with a sizeable collection of samples (n = 87) [155].
Second, there is very little work to date that has validated the functional impact of predicted enhancers annotated with epigenomic methods.As outlined in previous sections, two parallel questions need to be answered: (1) whether the enhancers can drive reporter expression and in what cell types; (2) what the target genes of the enhancers are and the levels of redundancy in regulating the genes.Technological innovations such as MPRA and CRISPRi-QTL screening have made it increasingly feasible to answer these questions [19].
In summary, there is already enough evidence supporting the importance of enhancers in regulating human brain development, and variation within these sequences is clearly connected to neuropsychiatric disorders and humanspecific aspects of brain evolution (Fig. 3).With the rapid advancement of high through-put functional genomics assays and in vitro brain models, it is reasonable to predict that translation into humans will occur in the near future.

Processing of ChIP-Seq Data Sets
Human fetal, postnatal, and adult brain H3K27ac ChIP-Seq data sets were downloaded from the PsychENCODE portal for human brain development (http://development.psychencode.org/#).Human adult brain prefrontal cortex (PFC) H3K27ac ChIP-Seq data sets were downloaded from the PsychENCODE data collection on Synapse (https://www.synapse.org/#!Synapse:syn9941133). Broadpeaks were called by MACS3 (https://github.com/macs3-project/MACS)and bigwig tracks were generated using bamCoverage in the deepTools package.To call differential peaks between fetal and adult samples, a consensus peak list was first generated using Diffbind [156], based on which a count matrix was generated.The count matrix was then normalized with the cqn package [157] followed by differential analysis by DESeq2 [158].To retrieve enhancers that are not in proximity of any promoters, peaks in promoter regions were removed by intersecting with regions +/−500 bp centered on TSS using bedtools2.

Predicting Target Genes of Enhancers
We linked target genes to the enhancers by either eQTL or HiC loops.For the fetal brain-specific enhancers, we used fetal brain eQTL data set [1] and HiC data set [97] (merged CP and GZ data sets).For the adult brain-specific enhancers, we used adult brain eQTL and HiC data sets [22].The target genes of each set of enhancers were subsequently merged and submitted to the ToppGene Suite [159] for gene set enrichment analysis with ToppFun.

Fig. 1 .
Fig. 1.Methodologies to identify and functionally validate enhancers.a Methods to identify enhancers based on illustration of an enhancer in action that loops to its target gene.TFs bind to and help activate the enhancer through specific sequence patterns.The histone residuals next to an active enhancer and promoter are modified in a specific manner, concomitant with the open chromatin where the enhancer is localized.Based on these observations, active enhancer can be identified through ChIP-Seq of histone modifications, in addition to ATAC-Seq which cuts and tags open chromatin.To map target genes of enhancers, HiC can be used and identifies chromatin interaction by ligating the sequences in physical contact followed by sequencing.b Experiments to functionally validate enhancers.The first class of experiments is reporter assay (upper panel), which puts a reporter gene directly downstream of the enhancer; the presence of reporter signal indicates the capacity of the enhancer to drive expression.The high throughput version of reporter assay is the MPRA, which is based on a pool of reporters, the activity of each is measured by transcription of a uniquely associated barcode.The second class of experiments is modulation-of-function assay (lower panel).This can be knockout of enhancer by CRISPR/Cas9, or repression or activation of enhancer by CRISPRi or CRISPRa.The goal is to measure change of gene expression in response to the modulation of enhancer activity.The high through-put version of these assays is the CRISPR-QTL analysis, where a pool of guide RNAs (gRNAs) is delivered to the experimental model, then scRNA-Seq was performed to identify differential gene expression events that correlate with specific gRNA(s), thus identifying the genes regulated by the enhancer targeted by the gRNA(s).

Table 1 .
Resources of epigenomics data sets related to human fetal brain enhancers c Results of gene set gene ontology (GO) Dev Neurosci 2024;46:69-83 DOI: 10.1159/000530929