New subtypes of allele-specific epigenetic effects: implications for brain development, function and disease

Abstract

Typically, it is assumed that the maternal and paternal alleles for most genes are equally expressed. Known exceptions include canonical imprinted genes, random X-chromosome inactivation, olfactory receptors and clustered protocadherins. Here, we highlight recent studies showing that allele-specific expression is frequent in the genome and involves subtypes of epigenetic allelic effects that differ in terms of heritability, clonality and stability over time. Different forms of epigenetic allele regulation could have different roles in brain development, function, and disease. An emerging area involves understanding allelic effects in a cell-type and developmental stage-specific manner and determining how these effects influence the impact of genetic variants and mutations on the brain. A deeper understanding of epigenetics at the allele and cellular level in the brain could help clarify the mechanisms underlying phenotypic variance.

Introduction

Mental illnesses and age-related brain diseases are prevalent throughout the world and encompass many different conditions that can be variable both in phenotype and severity. Defining the factors that play important roles in determining phenotypes and disease severity is necessary for developing improved therapeutic interventions. Currently, we know the same genetic loci are often linked to multiple disorders, which clouds our ability to make genotype to phenotype predictions for most diseases [1–3]. While we do not fully understand the mechanisms that contribute to psychiatric disease risk and phenotypic variability, interactions between genetic and epigenetic factors likely play important roles. A major new area of focus emerging in the field involves understanding genetic and epigenetic risk factors at the allele and cellular levels in the brain. Genetic risk factors are frequently heterozygous, meaning one allele is impacted. However, less is known about allele-specific epigenetic effects that cause differences in the expression of one allele compared to the other. On the basis of new studies that reveal allelic differences in gene expression and epigenetic regulation in the brain and other tissues, we argue that this is an important new area for the neurosciences that will deepen our understanding of the mechanisms regulating brain cell development, function, plasticity, aging, and disease risks. Here, our central thesis is that many different subtypes of epigenetic allelic effects exist in the brain, which serve different functions, arise due to distinct mechanisms and have different potential effects on disease risks. We argue that the field needs to focus on more effectively elucidating subtypes of allele-specific gene regulation in vivo and at the cellular level in the brain. By achieving this, new molecular mechanisms controlling brain development, function and disease may be revealed.

Subtyping allele-specific epigenetic effects in the brain—heritability, clonality and stability

Epigenetic allele-specific expression (ASE), which causes one parental allele to be expressed differently from the other allele for a gene, can take on a variety of potential forms that can be subtyped based on heritability, clonality and stability (Figure 1). Understanding the mechanisms and expression events that fall into a particular subtype is important because it informs the functions, potential health effects, and mechanisms involved. Heritability refers to whether or not an allele-specific biochemical mark and expression event are heritable from parent to offspring and result in changes to the expression of one parental allele. Genomic imprinting is the prototypical example of this phenomenon. Imprinting effects in mice and humans have been profiled in multiple tissues and across different ages in males and females. Imprinted genes can be subtyped based on the nature of the expression effect. For example, canonical imprinting involves silencing of one parental allele at the tissue level.

Figure 1.

Decision tree schematic for determining allele-specific expression subtypes. First, RME can be a result of genetic or non-genetic mechanisms. Non-genetic mechanisms can be further delineated by heritability from maternal or paternal ancestral alleles, clonality from parental cell to daughter cells, and stability over time. Stable ASE lasting days could affect protein levels, while dynamic ASE due to transcriptional bursting might have limited effects on protein levels and predominantly impact RNA, though little is known.

However, RNASeq has revealed many imprinting effects that involve biases to express one parental allele at a higher level than the other [4–8]. We refer to these biases as noncanonical imprinting due to the expectation that noncanonical imprinting effects have different functions and mechanisms than canonical imprinting effects [4,9]. Noncanonical imprinting exists in wild-derived populations is especially of interest to neuroscience because it is relatively more prevalent in the brain compared to other tissues and more prevalent in the genome than canonical imprinting, at least in mice [4]. A recent study of >11 000 parent-of-origin (PofO) phased human transcriptomes and methylomes revealed imprinted regions that are polymorphic across individuals and that many imprinting effects involve biases to express one parental allele at a higher level than the other [10•], indicating that noncanonical imprinting exists in humans. These, and other studies of imprinting across different human tissues [7,11], have revealed the landscape of imprinting in the human genome. Finally, other paternally and maternally heritable epigenetic effects have been described and shown to affect offspring phenotypes and gene expression, though it is not yet known whether any of these heritable effects occur through allele-specific mechanisms [12].

Clonality refers to whether or not the allelic event is mitotically heritable from parent cell to daughter cell. Genomic imprinting and random X-chromosome inactivation (XCI) are classic examples, although, other clonal allelic events are being discovered. Widespread clonal random monoallelic gene expression was originally found for autosomal genes in human lymphoblastoid cell lines [13] and then in mouse cell lines [14]. More recently, clonal, mitotically inherited random monoallelic cis-elements were described in mouse embryonic stem cell and neural progenitor cell lines [15••]. However, the prevalence of clonal random monoallelic expression on the autosomes (aRME) is debated [16–19]. Data from single cell RNASeq experiments indicate that clonal aRME exists, but is very rare in vivo and predominantly related to genes that are lowly expressed [19,20]. In our view, more studies are warranted. In support of this, a new study directly followed the allelic expression of Bcl11b over time in mouse T cells using transgenic mice engineered to express different fluorescent proteins from each allele [21••]. The authors directly uncovered cases of clonal inheritance in which only a single allele was randomly expressed and daughter cells inherited the same monoallelic expression pattern as their parent cell. This study shows that stochastic epigenetic allelic expression of Bcl11b influences T cell fate timing and development, raising the question of whether this mechanism occurs for some genes in neural lineages. Recently, aRME was found to be prevalent in vivo and relatively enriched in the developing compared to adult mouse brain, though the study could not determine the stability or clonality of the allelic effects [22•]. Direct cell imaging experiments of allelic expression in brain cells is an important future direction to investigate stability and clonality and deepen our understanding of the phenomenon and the role of aRME in brain cell fate and development. In general, it is clear that most aRME events are mechanistically and manifestly different from random XCI in females and RME for clustered protocadherins and olfactory receptors.

The final factor, we consider is stability over time. ASE in a cell can be (i) dynamic over hours, (ii) permanent over the lifetime of the cell or (iii) stable over days, but remain flexible in response to developmental or environmental cues (Figure 1). Dynamic allelic expression due to transcriptional bursting is prevalent [20,23,24] and new studies of transcriptional bursting at the allele and cellular level in mouse fibroblasts indicate that an allele is expressed every 4 hours for the average gene and as rarely as once every 24 hours for a small number of genes [25••]. Little is known about transcriptional bursting in different brain cells. Permanent monoallelic states are known from canonical genomic imprinting, random XCI and random monoallelic expression of olfactory receptors and protocadherins. The mechanisms involved are becoming relatively well understood [26,27]. However, even these permanent allelic states can differ according to cell type and developmental stage [28,29]. Indeed, recent studies of imprinting in the brain revealed how imprinting can differ according to cell-type for well-studied imprinted genes, like Dlk1 [30–32] and Igf2 [33]. One intriguing new direction in the field involves testing the possibility that noncanonical imprinting effects (also called parental allele biases) in the brain can involve highly cell-type-specific imprinting [4]. Indeed, even genes that escape from random XCI in females have been revealed to be cell-type and developmental stage specific and more frequent than previously recognized [34]. In general, studies of allele-specific epigenetic events at the cellular level in the brain are in their infancy and the area is ripe for discovery.

Unlike dynamic and permanent allelic events, the potential for stable allelic effects that persist over days, but can change due to a specific developmental process or environmental cue are difficult to identify from RNASeq data and less understood. Indeed, few studies have followed allelic expression states in the same cell over time. However, evidence for developmentally regulated and stable aRME emerges from a combination of studies. First, in vitro studies found that aRME increased as embryonic stem cells differentiated into neural precursors, revealing developmental and cell lineage commitment effects [35,36]. The authors observed clonal inheritance of monoallelic expression among neural progenitors. More recently, in vitro studies of monoallelic expression in neural precursors differentiated into astrocytic versus neuronal lineages found that monoallelic expression is highly cell-type dependent. Most monoallelic and allele skewed expression events manifested after lineage commitment and most impacted genes are uniquely expressed in astrocytes or neurons [37]. These findings suggest roles for allele-level regulation in brain cell development and are consistent with the results of in vivo approaches to measure RME in the brain [22•], which found that aRME is more prevalent in the developing postnatal mouse brain compared to the adult brain. In the human brain, aRME is known to occur and impact disease-linked genes [22•,38] and was recently shown to be developmentally regulated in human induced pluripotent stem cell lines [39]. However, none of these studies directly examined the temporal stability of the allelic effects in individual cells.

To our knowledge, only one study has directly followed allelic expression at the cellular level over time and, by analyzing Bcl11b allelic expression in T cell lineages, the authors found that monoallelic (or biallelic) expression states can be stable for at least one hundred hours in the same cell, amounting to several days [21••]. In addition, the study shows how allelic states change as cells differentiate and finds that Notch signaling regulates stochastic epigenetic allelic states and T cell lineage commitment [21••]. An earlier study using a related approach found that ASE controls Nanog dosage in embryonic stem cells and a shift from monoallelic to biallelic expression occurs during differentiation [40]. Together, these studies provide direct evidence for aRME that is stable, but developmentally regulated, revealing an effect that is distinct from dynamic transcriptional bursting and permanent monoallelic or biallelic expression (Figure 1). Similar direct imaging approaches are now needed to study allelic expression for different genes in developing and adult brain cells in order to determine the nature of ASE in developing brain cells and begin to unravel the functions and mechanisms involved.

Finally, a potentially important observation we found in our recent study is that a subset of genes in the brain are distinguished by especially tight co-expression of both parental alleles, such that each parental allele is in lockstep with the other beyond what is typical for most genes [22•]. We refer to this phenomenon as allele co-expression and the functions and mechanisms involved are not known. Overall, we expect that different dynamic, permanent, and stable allelic states that include (i) canonical imprinting, (ii) noncanonical imprinting, (iii) stable random monoallelic expression, (iv) dynamic random monoallelic expression, (v) clonal random monoallelic expression, and (vi) allele co-expression, have unique functions and regulatory mechanisms, most of which are not known. Moreover, for most of these subtypes of ASE we do not know the identity of the genes regulated by these effects in different cell-types of the brain, the intrinsic and extrinsic signals that determine them or how they may influence disease risk, onset or severity.

Understanding subtypes of epigenetic allelic effects in brain development, function and disease

The emerging evidence for various subtypes of epigenetic allelic effects opens up many questions about their functions and how they might influence brain disease risk and progression. In general, ASE could function to (i) control gene dose, (ii) mediate gene expression through the resolution of allelic enhancer competitions or chromatin conformation incompatibilities, and (iii) promote molecular diversity among otherwise similar types of cells (Figure 2). Genomic imprinting, which is one of the best studied forms of ASE, arose in eutherians under selective pressures that are debated and, as it blocks parthenogenesis, imposes a requirement for sexual reproduction [41]. However, the function of imprinting in the brain is poorly understood. Recent reviews have comprehensively covered the known roles of imprinted genes on brain development, synaptic plasticity and different aspects of learned and innate behavior [28,29]. Although these studies show that imprinted genes play important roles in the brain, they do not address the motivation for studying imprinted genes compared to other epigenetic mechanisms in the brain. An enticing motivation is that imprinting in the brain helps reveals critical molecular and neural mechanisms for controlling important behavioral and physiological phenotypes. The Kinship Theory for the evolution of imprinting proposes that maternally expressed imprinted genes (MEGs) function antagonistically to paternally expressed genes (PEGs) due to an evolutionary conflict over traits that influence offspring demands on the mother [42]. This parental conflict arises over the dosage of specific genes, leading to silencing of one parental allele, and fits some cases of imprinting [43]. Co-expression networks of imprinted genes thought to reflect the evolutionary conflict between MEGs and PEGs have been described [7]. Moreover, an imprinted gene network of MEGs and PEGs has been found that regulates offspring growth [44,45]. If the Kinship Theory is an explanation for imprinting in the brain, then we expect to discover antagonistic molecular and neural mechanisms for controlling important brain functions and behaviors by studying imprinted genes. Indeed, a framework is emerging for MEGs and PEGs in the control of apoptosis in the brain [5,29], neurogenesis [29], maternal care [46], sleep [47] and monoaminergic neuron development and signaling [4,48–52]. Cell-type-specific imprinting effects are especially intriguing because they are potential markers for novel functional subpopulations of brain cells controlling a particular neural circuit, behavior pattern or internal state. Finally, imprinting causes parent-of-origin effects for inherited genetic variants such that the risk that an inherited pathogenic allele will cause disease depends upon whether it is maternally or paternally derived. Therefore, understanding parent-of-origin epigenetic effects and the neural mechanisms they regulate will help the field evaluate the potential effects of inherited genetic risk and de novo mutations in imprinted genes.

Figure 2.

Functions of epigenetic allele-specific expression subtypes. Epigenetic monoallelic and biallelic expression could function to (a) control gene dosage, (b) mediate gene expression through the resolution of allelic enhancer competitions or chromatin conformation incompatibilities, or (c) increase heterogeneity among otherwise homogenous cell populations and promote molecular diversity via allelic expression variation. (b) Enhancer competitions between two genes can occur when both genes need access to the same enhancer to be expressed. By allowing one gene to access the enhancer on one allele and the other to access it on the other parental allele, both genes can be expressed simultaneously, thereby resolving the conflict, which is known to occur for some imprinted genes. ASE can also potentially resolve chromatin conformation incompatibilities in which regulatory contacts for one gene interfere with those for a neighboring gene. For example, intragenic regulatory elements could cause transcriptional interference for the host gene (Gene 1) that is resolved by ASE to allow both genes to be expressed.

While our understanding of imprinting is growing, the functions of dynamic, stable, permanent and clonal aRME in the brain are largely unknown. Previous reviews have speculated on possible functions [9,17,53]. For example, based on studies of stochastic gene expression and pulsing in bacteria and mammalian cells [54], dynamic RME could function to diversify the molecular state of cells by randomizing gene dosage and the expression of heterozygous variants (Figure 2). This stochastic allelic diversification of a cell population would have the advantage of placing otherwise identical cell-types into different states, such that at least some cells are optimally poised to respond appropriately to unpredictable neural inputs and molecular signals (Figure 2c). This idea suggests that dynamic (and stable) RME could introduce molecular diversity into neural networks and that loss of this diversity would restrict and impair neural processing and behavioral responses. Randomly diversifying cells of a particular type would also be advantageous during development to help ensure some cells are in a state to respond correctly to unpredictable developmental cues in the environment, whereas those that do not respond correctly might be removed by apoptosis. Stable ASE that persists over days could potentially function to control gene dosage and combinatorial gene expression patterns in a cell to maintain a specific cellular state and/or resolve regulatory conflicts due to enhancer competition or incompatible chromatin states between neighboring genes (Figure 2) [9]. The advantages of clonally inheriting random monoallelic states compared to having each daughter cell independently initiate random allelic expression are not clear.

In the context of heterozygous mutations, random monoallelic expression can cause mosaics of brain cells that differentially express the mutated versus wildtype alleles for a gene [22•,55]. If the allelic states are permanent or stable, these effects could shape how heterozygous mutations and variants are expressed during critical developmental periods and in specific cell-types (Figure 3). Previously, the co-expression of multiple autism risk genes was found to converge in deep layer projection neurons during midfetal brain development, implicating this neuron population and developmental stage in autism risks [56,57]. RME and genomic imprinting could shape these types of co-expression effects at the allele and cellular level, thereby influencing disease risks and phenotypic variance (Figure 3). In addition, the number and location of brain cells monoallelically expressing a pathogenic allele or combination of pathogenic alleles could influence the phenotypic effects of the variant(s). Studies of epigenetic variance and ASE in cancer have shown that these are important driving forces in disease [58,59]. Therefore, understanding the genes, ages and brain cell-types impacted by different types of epigenetic allelic effects could help predict convergent effects from different genetic mutations in developing and adult brain cells and deepen our understanding the basis of phenotypic variance in mental illness and brain disease.

Figure 3.

Interactions between allele-specific epigenetic and genetic effects during brain development can influence the risks for brain disorders. (a) Inherited variants and de novo mutations influencing cell-type or brain-region-specific imprinted genes could have effects on neural circuits and pathways conferring disease risk that depend on the identity of the parental allele impacted (white star indicates mutated maternal allele) (b). Cell-type-dependent aRME could, depending on the number and location of monoallelic brain cells expressing a pathogenic allele (or combination thereof), influence the phenotypic effects of the variant(s). (c) Developmentally regulated monoallelic expression could interact with genetic variants to shape the phenotypes of brain cell populations. The schematic depicts three different cells expressing two different risk genes. The asterisk indicates the mutated allele for a heterozygous mutation in each gene. The image shows how the monoallelic expression of different combinations of pathogenic and wildtype alleles for the two risk genes at the cellular level during critical developmental periods could potentially influence brain cell health and contribute to disease risks. ASE, allele-specific expression.

What are the mechanisms that control different subtypes of allelic effects?

Several studies have focused on discovering imprinting and aRME effects in the mouse and human genome, but relatively few have investigated the mechanisms involved in different forms of ASE. The mechanisms regulating canonical genomic imprinting have been reviewed recently and are relatively well understood [60]. Less is known about the mechanisms involved in noncanonical genomic imprinting. Previous studies found that noncanonical imprinted genes have different chromatin modifications on the maternal versus paternal alleles [4] and that clusters of canonical imprinted genes can expand the imprinting effect to induce noncanonical imprinting on neighboring genes in developmental stage and cell-type-dependent manner [4–6]. Currently, we do not know the mechanisms involved. Some noncanonical imprinted genes are not associated with known imprinted gene clusters, and it is not known if the mechanisms involved at these loci are similar or different from clustered imprinted genes. An important first step is to determine the identity of the brain cells that exhibit imprinting and those that do not for each MEG and PEG. Comparisons of the epigenetic landscapes in the cell-types that exhibit imprinting versus those that do not will help reveal the mechanisms involved.

Several recent studies have expanded our understanding of the mechanisms underlying aRME, though few are focused on the brain. A recent major study of cellular allele-specific chromatin conformation using HiC and HiFISH in human fibroblast cell lines discovered that topologically associated domains are variable at the cellular level. Furthermore, individual alleles exhibited independent behavior, showing that chromatin conformation is highly variable between alleles at the cellular level [61•]. Currently, it is unknown which of these allele-specific chromatin conformations and allelic regulatory contacts are related to dynamic versus stable allelic expression states at the cellular level or how allele coexpression occurs. A recent study of dynamic RME in mouse fibroblasts found that allelic transcriptional burst size is controlled by promoter architecture, while allele expression frequency is controlled by enhancers [25••]. Moreover, a study of epigenetic variation in the human genome revealed genome-wide stochastic allele switching affecting as much of 8–10% of the autosomal genome [62••]. This study by Onuchic et al. indicates that stochastic switching, which the authors define as random transitions between fully methylated and unmethylated states within a sample, occurs at thousands of regulatory loci (epialleles) in a sequence-dependent manner. Interestingly, these epialleles are enriched near disease associated loci. Thus, one mechanism of aRME might be genomic sequence dependent, allowing for dynamic methylated and unmethylated states over time at specific genomic sites. The study observes enrichment of sequence-dependent allele-specific methylation within enhancers consistent with other reports [63] suggesting disruption of transcription factor binding within enhancers via point mutation is a possible mechanism of ASE [64]. Currently, it is unknown how long these epialleles maintain their methylated or unmethylated states, how they impact gene expression and what other epigenetic machinery facilitates these allelic effects. Finally, clonally inherited random monoallelic cis-elements, including enhancers and promoters, were recently uncovered in mouse embryonic stem cell and neural progenitor cell lines based on open chromatin using ATAC-Seq [15••]. Thus, the field is beginning to uncover candidate mechanisms and genomic elements for controlling different subtypes of epigenetic ASE. However, the field has not determined how different epigenetic mechanisms and cis-elements influence the stability, mitotic heritability or cell-type specificity of most allelic states, which is an important area for future study.

Conclusions

Epigenetic ASE in the brain has traditionally been a niche field of study focused on canonical imprinted genes, random XCI, olfactory receptors and clustered protocadherins. In our opinion, the latest studies on epigenetic ASE support a broadening of the field. They indicate that allele-specific epigenetic gene regulation is more prevalent in the genome and the brain than previously thought and takes on a variety of forms that are distinguished by differences in heritability, clonality and stability over time. Direct imaging studies in immune cells show the temporal stability of aRME states can be stable over days for single genes, which could affect protein levels and/or how heterozygous variants impact cell functions, disease risks and progression, and phenotypic variance. However, such studies have not yet been performed in brain cells. New studies of allele-specific DNA methylation and open chromatin have begun to uncover candidate cis-elements and mechanisms for controlling aRME. Recent studies have also revealed that epigenetic ASE is developmentally regulated and cell type specific, uncovering a potentially important new feature of gene regulation at the cellular level during brain development. Although the functions of epigenetic ASE are unclear in most cases, several testable hypotheses have been proposed. Overall, the primary future direction for noncanonical imprinting and aRME studies should focus on determining functions, which includes characterizing the following features: the cell-types and developmental stages impacted for specific genes, the temporal stability and functional effects of ASE at the protein level, and how these effects interact with genetic variants and mutations to influence phenotypes. Solving these problems will advance the field by determining how these newly discovered phenomena impact health and disease. Additional priorities involve determining the mechanisms involved in different subtypes of ASE, characterizing the identity of the genes and neural pathways impacted in different mouse and human tissues and determining how different ASE states change in response to aging, disease and other physiological and environmental factors.