Divergence and rewiring of regulatory networks for neural development between human and other species

ABSTRACT

Neural and brain development in human and other mammalian species are largely similar, but distinct features exist at the levels of macrostructure and underlying genetic control. Comparative studies of epigenetic regulation and transcription factor (TF) binding in humans, chimpanzees, rodents, and other species have found large differences in gene regulatory networks. A recent analysis of the cistromes of REST/NRSF, a critical transcriptional regulator for the nervous system, demonstrated that REST binding to syntenic genomic regions (i.e., conserved binding) represents only a small percentage of the total binding events in human and mouse embryonic stem cells. While conserved binding is significantly associated with functional features (e.g., co-factor recruitment) and enriched at genes important for neural development and function, >3000 genes, including many related to brain and neural functions, either contain extra REST-bound sites (e.g., NRXN1) or are targeted by REST only (e.g. PSEN2) in humans. Surprisingly, several genes known to have critical roles in learning and memory, or brain disorders (e.g., APP and HTT) exhibit characteristics of human specific REST regulation. These findings indicate that more systematic studies are needed to better understand the divergent wiring of regulatory networks in humans, mice, and other mammals and their functional implications.

There are key structural and functional innovations in the evolution of human brains. The human neocortex, which is involved in higher functions including reasoning and thought, contains various developmental and structural features distinct from those of rodents and other mammalian species, such as significant enlargement and complex elaboration.¹ For example, the surface of the mouse neocortex is smooth (lissencephalic), while the surface of the human neocortex is highly convoluted (gyrencephalic). At the cellular level, human neocortex expansion is related to the expansion of progenitor cells in the outer subventricular zone (OSVZ) during development.² The OSVZ grows progressively thicker, due to a dramatic increase in the number of specific kinds of basal progenitor (BP) cells.³ The increase in BPs is followed by considerable changes in the cell type composition of the subventricular zone (SVZ). It is thought that differences in the composition and cell proliferation capability of the BP pools, including basal intermediate progenitors (bIPs), basal radial glia (bRG), and apical radial glia (aRG) in the SVZ and OSVZ eventually contribute to the specific expansion of the OSVZ and neocortex in humans.⁴ How these differences emerge and evolve remains a fascinating and active research area, but it is generally considered that this kind of structural or phenotypic difference is mainly the result of a difference in the genetic program regulating brain development, a view in line with the general idea that gene expression change is a major driver of species differentiation and evolution⁵ (Fig. 1).

Figure 1.

Cartoons illustrate distinct TF binding patterns leading to divergent gene expression during the development of human and mouse neural systems. (A) Conserved TF binding sites and similar gene expression profiles in humans and mice, (B) duplicated TF binding site (light green) in humans or (C) gain of human specific TF binding site (red) resulting in divergent gene expression profiles between humans and mice. The color circles represent 3 distinct TF binding sites.

Previous studies have identified both conserved and divergent gene expression patterns in the brains of mice and humans. Weighted gene co-expression network analysis of >1,000 expression microarrays from human and mouse brains identified conserved modules related to glial and neuronal cell function, but also uncovered human-specific modules involved in the Alzheimer disease.⁶ A comparison of ∼1,000 genes important for neural functions found a high degree of expression concordance between mouse and human visual cortices, but also observed that 21% of genes exhibited differential expression between the 2 species.⁷ Most of the differentially expressed genes showed distinct spatial expression patterns, suggesting that the control and regulation of these genes in the visual cortices of the 2 species have subtle differences.⁷ These studies, together with many other similar reports in the literature, show that gene regulatory networks may have undergone a large degree of rewiring between mice and humans. The findings also indicate that divergent gene regulation could be an issue, or at least needs to be considered, when mice are used as a primary model for studying the human nervous system. One way to systematically interrogate the extent of regulatory network rewiring is to compare the genome-wide occupancy of transcription factors that are important for regulating neural development and functions.

REST (Repressor Element 1 Silencing Transcription factor; also called Neuron Restrictive Silencer Factor - NRSF) is considered a critical transcriptional regulator for neural development and neural homeostasis. It has been mostly studied as a suppressor of important neural genes in embryonic stem cells (ESCs), neural progenitors, and other non-neural systems. Its function in neuronal cells, however, seems to be more complex and dependent on contexts^8-14 To better understand the roles of REST in regulating neural programs and how its cistrome has evolved, Rockowitz and Zheng recently compared genome-wide REST occupancy in human and mouse ESCs, using data from chromatin immunoprecipitation coupled with deep sequencing (ChIP-seq). They identified significant expansion of REST binding in human ESCs (hESCs) compared with its binding in mouse ESCs (mESCs) (8,199 occurrences in human versus 4,107 in mice).¹⁵ A small core group of REST binding sites (15.3% of hESC REST peaks and 30.4% of mESC peaks) was conserved between humans and mice (i.e., occurring in syntenic genomic regions). The conserved binding was higher for REST than for many ESC specific transcription factors, e.g. NANOG, OCT4, and SOX2, probably reflecting either the importance of REST in regulating neural genes or its broad function as an epigenetic modulator; the binding conservation of REST was similar to that of CTCF, another key epigenetic regulator. Note that at the chromatin level, REST binding was less conserved than histone posttranslational modifications, such as H3K4me3. Conserved REST binding sites exhibited several features important for REST functions: increased DNA sequence conservation, higher cofactor (e.g., SIN3 and CoREST) co-occupancy, enrichment of specific types of histone modifications, reduced DNA methylation, increased association with genes that responded to knockdown of REST expression, and most importantly a significant association with genes implicated in neural development and function (Fig. 2). Many human-specific REST binding sites also displayed some of these features, suggesting those sites are likely to confer gene regulation as well. The authors' analysis of the sequence features of REST peaks showed that the biochemistry of the REST-DNA interaction was similar in the 2 species, as similar percentages of REST peaks (95.3% in hESC and 99.1% of mESC) contained the REST canonical binding motifs (the Repressor Element 1 (RE1) motif) or its close variants. Interestingly, a large portion of human-specific REST peaks also contained RE1 motifs. Further analysis showed that many of these RE1 sites have emerged recently during evolution; some are primate-specific and derived from transposable elements or repetitive sequences, including short interspersed nuclear elements (LINEs) or short interspersed nuclear elements (SINEs). There were cases where the RE1 motif was conserved in mice but not bound by REST in mESCs, an observation needing further study for factors underlying the binding turnover. Analysis of genes with human-specific REST peaks showed that REST also targets many of these same genes in other human cell types,¹¹ suggesting that the targets are likely genuine and that REST can regulate those genes in diverse human cells and tissues, including differentiating neurons.

Figure 2.

A model of REST cistrome rewiring between humans and mice. (A) Conversed REST binding sites, highly enriched with cognate RE1 motif, co-factors, specific histone modifications and DNA hypomethylation. (B) Expanded REST binding in human. (C) Human specific REST binding. All three modes of REST regulation are enriched with genes important for neural functions but with distinct characteristics.

Very interestingly, some of the ∼3000 human-specific REST targets have a broad array of known roles in brain function. For example, gene ontology and pathway analysis indicated that some of the genes bound by REST specifically in hESCs were related to learning and memory (e.g. APP and HTT), axon guidance, GNRH signaling, CREB signaling in neurons and CRH signaling.¹⁵ The authors also found many genes containing more REST binding sites in hESCs than in mESCs. For example, AUTS2 and NRXN1/2/3 have 4 and 6/6/19 REST peaks, respectively, in hESCs, but only 1–3 peaks in mESCs. The increase of REST binding to the human NRXNs is quite interesting, because the migration of neuroblasts from the ventricular zone to the cerebral cortex is altered, leading to lissencephaly, when the binding activity of ligands such as neurexin to dystroglycan is abolished.¹⁶ Redundant TF binding (such as the kind particularly prominent in the human neurexin genes) has been proposed to backup or be in an epistatic relationship with other TF binding sites.¹⁷ Thus, it is conceivable that the expansion of REST binding to these genes might allow REST to fine-tune transcription and achieve greater regulational complexity for key human neural genes.

Rockowitz and Zheng also noted that several well-characterized causal genes for Alzheimer disease (APP, PSEN2 and SORL1), Huntington's disease (HTT and SLC2A3), and Parkinson disease (PARK2, PARK7) were bound by REST in hESCs but not mESCs. More detailed comparison of human-specific REST targets with genes previously implicated in various brain disorders suggested that human-specific REST regulation may play a role in oxidative-stress-involved brain disorders and Amyotrophic Lateral Sclerosis. This finding is consistent with a recent report that REST, as a protector against neurodegeneration, plays an important role in oxidative stress response, especially in the aging human brain and Alzheimer disease.¹⁸ The functional consequence of the human specific REST binding to these genes is, of course, yet to be established.

Overall, the study by Rockowitz and Zheng underscores significant differences between the REST regulatory networks of humans and mice, highlighting the divergence of the transcriptional networks between these 2 species. As the study was carried out on ESCs, which represent the ground state of development, it provides only a static view of the REST regulatory network. In addition, unlike the mESC data, only one set of human ESC ChIP-seq data was used in the study. It will be interesting to repeat the same analysis with an independent REST ChIP-seq dataset to uncover potential influences of any other experimental factors not considered in the original study, such as potential differences in the efficiencies of antibodies or ChIP protocols. More importantly, the authors have previously shown that REST cistromes are quite diverse across human cell types,¹¹ and others have also reported that TF binding events with greater cell specificity are less conserved between species.⁷ Therefore it would be quite interesting to expand the same analysis to neural progenitors, differentiated neurons, and other cell types that are more directly related to brain development and function. As the greater proliferative capacity of human neural progenitors contributes to neocortical expansion⁴ and REST is known to be important for maintenance of stem cell plasticity, it would be valuable to know if the expansion of REST regulation plays any role in controlling the transition from neural progenitor proliferation to differentiation. We should note that the partial conservation of REST cistromes across species has been reported previously,^19-21 independent of binding profiles identified by ChIP-seq analysis.^19,20 Future studies that can provide novel insights as to the mechanisms underlying differential REST binding to human and mouse chromatins will be valuable. For example, it will be interesting to elucidate whether there is a structural difference between the human and mouse REST proteins.

Evolutionary rewiring plays a critical role in the transcriptional regulatory networks controlling brain development, but cross-species differences in gene regulation have long been recognized as major contributors to speciation and adaptation, especially between closely related species.^5,22 With the advent of new deep sequencing technologies, researchers have been able to integrate diverse data to study regulatory mechanisms at an unprecedented resolution and scale. ChIP-seq based transcription factor binding maps show that the divergence of transcription factor binding is common in metazoans. TF binding conservation in vertebrates, nevertheless, varies significantly, depending on the specific TFs and tissues, which can be as high as 60% and as low as 2%.²³ Some transcription factors, such as NF-κB, bind to their DNA motifs and can recruit cofactors in a context specific manner.²⁴ Thus, for a transcription factor like REST, whose regulatory outcome is largely mediated by its cofactors, the conservation of cofactor occupancy is also important to consider when studying the functional consequence of regulatory divergence.¹⁵

One important question then is how differential gene regulation emerges. Mutations in TF binding sites, including point mutations and indels, can explain a substantial portion but not all of the binding divergence.^25-27 Despite the dramatic turnover of individual TF binding, Ballester et al. found that about 2-thirds of TF binding sites were located within cis-regulatory modules (CRMs), although less than half of the CRMs in humans were also CRMs in other species,²⁸ suggesting some degree of co-evolution of different TF binding.

Repeat elements are another known factor contributing critically to regulatory network rewiring. Repeats such as endogenous retroviral sequences (ERVs) and SINEs are involved in the recent evolution of promoters and enhancers.²⁹ Retrotransposition followed by point mutations or indels could lead to the emergence of novel CpG sites, providing potential regulatory elements for TF binding and DNA methylation.³⁰ As discussed above, many human specific REST-bound RE1 motifs are derived from LINEs/SINEs. The expansion of transposable elements has also given rise to novel binding sites for other TFs, for example, ERV1 repeats for p53,³¹ OCT4 and NANOG,³² and SINEs for CTCF.³³

In comparison to ChIP-seq studies focused on particular TFs, mapping open chromatin by DNase I footprinting sequencing (DNase-seq) or ATAC-seq³⁴ potentially can identify “all” TF binding sites in a sample. Its application to multiple human and mouse cells and tissue types has also demonstrated the dramatic turnover of TF bindings: ∼50% of mouse TF-DNA interactions are maintained in humans despite the finding that the trans-acting circuitry in mice is highly similar to that in humans.³⁵ The prevalence of TF network rewiring is also supported by the cross-species comparison of DNA methylation, a key epigenetic regulator of gene expression.³⁰

At the DNA sequence level, regulatory elements such as promoters and enhancers also seem to have experienced various degrees of evolutionary divergence. Young et. al. reported that promoter turnover has occurred at ∼56% of protein-coding genes since humans and mice diverged.³⁶ In a study of H3K4me3-marked active promoters in the prefrontal cortices of human, chimpanzee and macaque, ∼400 regions were found to be enriched for H3K4me3 in human specifically, including some proximal to DPP10, CNTN4, and CHL1, which are strongly linked to autism, schizophrenia and related disorders.³⁷ Forebrain enhancers have also diverged significantly, with only 58% of active enhancers bound by the p300/CBP in human forebrains exhibiting conserved binding in mice.³⁸ More comprehensive investigations have been performed for active promoters and enhancers in livers across 20 mammalian species using H3K27ac and H3K4me3 ChIP-seq analysis;²⁹ Villar et al. found that enhancers have evolved more rapidly than proximal promoters, but more slowly than TF binding (e.g., CEBPA).²⁹ They also showed that the recently evolved enhancers are frequently associated with genes under positive selection.

Although considered to be less extensive, evolutionary changes in protein coding regions of the human genome also play a role in the emergence of human brain specific features and its regulation. Positive selection has undoubtedly played a critical role in human evolution. Most (if not all) of the phenotypic traits that define our species: advanced cognitive abilities, complex vocal organs, bipedalism, opposable thumbs, and expanded brain size, are likely to be the product of strong positive selection.^39-41 Indeed, positive selection has been found to occur at the promoters of many genes (e.g., SCN1A and PRSS12) involved in neural development and function.⁴² Moreover, experimental evidence has established the neural functions of 2 human specific genes. Two independent research groups have reported that SRGAP2, which regulates neuronal migration and morphology, has undergone 2 rounds of human-specific partial duplications, leading to the production of an incomplete protein that antagonizes the full-length protein and contributes to the evolution of the human neocortex.^39,40 Likewise, forced expression of the human-specific gene ARHGAP11B in embryonic mouse neocortex has been shown to increase basal progenitor proliferation and induce cortical folding, strongly supporting its role in the evolutionary expansion of the human neocortex.⁴³

Human accelerated regions (HARs) often encode developmental enhancer RNAs (eRNAs) and thus are also important for human brain evolution. For example, HAR1 eRNAs are specifically expressed in the developing human neocortex,⁴² while HARE5 enhances FZD8 expression, a receptor of the Wnt pathway implicated in brain development and size.⁴⁴ The HARE5 enhancer has experienced fast evolution in primates⁴⁵ and has species-specific functions in brain development. Specifically, HARE5 has been shown to promote faster progenitor cell cycles and increase neocortical size when introduced to transgenic mice.⁴⁴ Interestingly, ARHGAP11B was among a list of “young” (recently evolved) genes that are expressed in the early (fetal or infant) developing human brain but not in the mouse brain.^46,47 In the future, it will be valuable to identify TFs that bind to these eRNAs and specify their brain expression.

Perspective

The current work on regulatory network comparison is largely focused on regulatory elements that are identified by one particular type of experimental or computational approach, such as by the identification of genomic regions occupied by transcription factors or associated with specific epigenetic modifications. Not all those TF-bound sites are functionally important. In cells, multiple layers of regulation function synergistically, including post-transcriptional regulation by miRNA networks, which by themselves have important and unique roles in regulating cell and region specific gene expression in brains.⁴⁸ Therefore, it will be important to carry out simultaneous comparisons of multiple layers of regulatory networks in order to understand their common and distinct evolution. Further understanding of how these networks co-evolve to affect gene expression and contribute to phenotypic consequence will be essential. In addition, the spatiotemporal and dynamic aspects of regulatory networks are certainly important factors to be considered in order to fully appreciate the network divergence among species. For example, given the complex and diverse roles of REST in different stages of neural and brain development, it will be crucial to understand how REST regulatory networks change as neurons develop and mature in different species. Last but not the least, it will be important to find out how much of the network rewiring actually contributes to regulatory robustness and how much to functional innovation.

Abbreviations

aRG

apical radial glia

ATAC-seq

assay for transposases-accessible chromatin using sequencing

bIPs

basal intermediate progenitors

BP

basal progenitor

bRG

basal radial glia

ChIP-seq

chromatin immunoprecipitation coupled with deep sequencing

CRM

cis-regulatory module

DNase-seq

DNase I footprinting sequencing

ERV

endogenous retroviral sequence

eRNA

enhancer RNA

ESC

embryonic stem cell

HAR

human accelerated region

LINE

long interspersed nuclear element

NRSF

neuron restrictive silencer factor

OSVZ

outer subventricular zone

REST

repressor element 1 silencing transcription factor

RE1

repressor element 1

SINE

short interspersed nuclear element

SVZ

subventricular zone

TF

transcription factor

Disclosure of potential conflicts of interest

No potential conflicts of interest were disclosed.

Funding

This work was supported by the National Institutes of Health (MH099452 to DZ and partially by MH099427 and HL129807).