Epigenetics and bipolar disorder: New opportunities and challenges
Abstract
Despite significant effort, understanding of the molecular causes and mechanisms of bipolar disorder (BD) remains a major challenge. Numerous molecular genetic linkage and association studies have been conducted over the last two decades; however, the data are quite inconsistent or even controversial. This article develops an argument that molecular studies of BD would benefit significantly from adding an epigenetic (epiG) perspective. EpiG factors refer to modifications of DNA and chromatin that “orchestrate” the activity of the genome, including regulation of gene expression. EpiG mechanisms are consistent with various non-Mendelian features of BD such as the relatively high degree of discordance in monozygotic (MZ) twins, the critical age group for susceptibility to the disease, clinical differences in males and females, and fluctuation of the disease course, including interchanges of manic and depressive phases, among others. Apart from the phenomenological consistency, molecular epiG peculiarities may shed new light on the understanding of controversial molecular genetic findings. The relevance of epigenetics for the molecular studies of BD is demonstrated using the examples of genetic studies of BD on chromosome 11p and the X chromosome. A spectrum of epiG mechanisms such as genomic imprinting, tissue-specific effects, paramutagenesis, and epiG polymorphism, as well as epiG regulation of X chromosome inactivation, is introduced. All this serves the goal of demonstrating that epiG factors cannot be ignored anymore in complex phenotypes such as BD, and systematic large-scale epiG studies of BD have to be initiated. © 2003 Wiley-Liss, Inc.
INTRODUCTION
Epigenetics is a new concept in studies of bipolar disorder (BD) (manic depression) and in psychiatric research in general. The Public MedLine search reveals a single reference for key words “epigenetic” plus “manic depression” (for comparison, the number of publications using “genetic” plus “manic depression” exceeds 1,100). This paper describes epigenetic (epiG) phenomena and explains why epigenetics is relevant to BD as well as how epiG strategies may supplement the traditional DNA sequence-oriented research paradigm in the elucidation of the molecular mechanisms of BD.
Depending on the period of time, epigenetics has been understood in quite different ways. Coining the term epigenetics, C.H. Waddington referred to epigenetics as the developmental processes that “connect” the genotype to phenotype, or the processes by which genotype gives rise to phenotype
Coining the term epigenetics, C.H. Waddington referred to epigenetics as the developmental processes that “connect” the genotype to phenotype, or the processes by which genotype gives rise to phenotype.
[Waddington, 1957]. These ideas originated long before the advent of molecular biology and for several decades represented predominantly a theoretical development. It is interesting to note that Waddington's work was noticed by psychiatrists and psychologists and applied to human behavior. For example, Singer and Wynne [1965] stated that “the individual's biological capacities for focusing attention and for perceiving, thinking, and communicating gradually are shaped and modified by interchange with the environment during development. This viewpoint is epigenetic.” Gottesman [1974] introduced the idea of epigenesis/epigenetics in psychiatric genetics in his papers and in his books with J. Shields [Gottesman and Shields, 1972], with one of them including “epigenetic” in the title [Gottesman and Shields, 1982].
Over the last decades epigenetics has undergone a significant metamorphosis from an abstract theory of developmental gene-environment interaction to a very dynamic and rapidly developing branch of molecular science. Epigenetics now investigates DNA and chromatin modifications that play critical roles in numerous important cellular processes
Over the last decades epigenetics has undergone a significant metamorphosis from an abstract theory of developmental gene-environment interaction to a very dynamic and rapidly developing branch of molecular science. Epigenetics now investigates DNA and chromatin modifications that play critical roles in numerous important cellular processes.
such as regulation of gene activity, including tissue-specific and age-dependent changes in gene activity, development of multicellular organisms, X chromosome inactivation, DNA mutagenesis, and meiotic recombination, among numerous others. The number of publications dedicated to epiG developments is visibly increasing every year. A number of journal issues have been dedicated to providing an overview of the progress and trends in epigenetics (e.g., Trends in Genetics 1997;13:8, Cell 1998;93:3, Science 2001;293:5532, and Nature 2003;421:686). The journal Science predicted that in the near future epigenetics may become a field of primary scientific interest, along with Alzheimers research, X-ray astronomy, river restoration, nanocomputers, and polio eradication [Anonymous, 1999]. It is also necessary to admit that epigenetics is still mapping the epiG territories of the molecular world by uncovering new epiG mechanisms and new epiG roles in the cell. The relative youth of the field makes it difficult to come up with a consensus of what epigenetics is. Some have called this situation in epigenetics a “semantic morass” [Lederberg, 2001]. According to one of the most popular definitions (which is also very relevant to the subject of this article), epigenetics refers to regulation of gene expression that is controlled by heritable but potentially reversible changes in DNA methylation and/or chromatin structure [Henikoff and Matzke, 1997].
Despite the uncertainties of the scope and essential features of epigenetics, there is sufficient experimental and theoretical background to consider epiG factors in etiopathogenesis of human diseases. The next sections are dedicated to the brief description of epiG mechanisms and explanation of why epiG dysregulation of genes can be a primary disease mechanism.
BASIC PRINCIPLES OF epiG REGULATION
In the mammalian genome, cytosines can have two functional states: unmodified cytosines (C) and cytosines that are methylated at the 5-position of the pyrimidine ring (metC). DNA methylation is an enzymatic reaction performed by several types of DNA-methyltransferases [reviewed in Bestor, 2000]. A large number of genes exhibit an inverse correlation between the degree of methylation and gene expression, and there is an increasing body of experimental evidence suggesting that epiG modification is intimately involved in the regulation of expression of genes [Holliday, 1996; Holliday et al., 1996; Razin and Shemer, 1999; Yeivin and Razin, 1993]. One of the mechanisms of epiG regulation of genes is related to methylation of the binding sites for transcription factors that changes the affinity of such factors to regulatory sequences of specific genes [Riggs and Porter, 1996; Riggs et al., 1998]. In addition to positional effects of metC, the density of metC in a gene regulatory region also contributes to gene activity. This type of regulation is linked to another layer of epiG regulation, namely, histone modification. Histones represent a key component of chromatin. metC binding protein (MeCP2) binds to methylated DNA and attracts histone deacetylases that hypoacetylate histones [Nan et al., 1997; Jones et al., 1998]. Transcriptionally competent chromatin is normally enriched with acetylated histones, while transcriptionally silent chromatin is deacetylated [reviewed in Robertson and Wolffe, 2000]. The interaction of DNA methylation and histone acetylation demonstrates that the two types epiG regulation act in concert. EpiG mechanisms of transcriptional repression are of primary importance for X chromosome inactivation, genomic imprinting, and the transcriptional inactivation of endogenous retroviruses and other parasitic sequences [reviewed in Wolffe and Matzke, 1999]. Over the last several years, several new types of histone modifications were identified that have provided the basis for the concept of the histone code [Jenuwein and Allis, 2001].
During embryonic development, the genome is subjected to major changes in epiG regulation [Monk et al., 1987; Mayer et al., 2000]. Although the mechanisms of such epiG dynamics are yet to be understood, epiG changes are consistent with cell differentiation and embryonic development. After tissues are formed, genomes of somatic cells are locked into tissue-specific patterns of gene expression, and epiG phenotypes of the somatic cells are inherited during mitotic divisions of cells [Maynard Smith, 1990]. In parallel with DNA sequence-based inheritance, epiG modification of the genome is called an epiG inheritance system [Maynard Smith, 1990]. While DNA sequences provide the order in which nucleotides have to be assembled in the mRNA molecule, the epiG inheritance system controls the quantitative aspects of mRNA synthesis. Cells operate normally only if both DNA sequence and epiG components of the genome function properly.
DNA and chromatin modifications do react to the events outside the cell, i.e., extracellular environment. In addition, quite significant epiG changes may occur even in the absence of evident environmental differences, i.e., due to stochastic reasons. After mitotic division, the daughter chromosomes do not necessarily carry identical epiG patterns in comparison to the parental chromosomes. Over time, substantial epiG differences may be accumulated across the cells of the same cell line or the same tissue. EpiG regulation of a gene is a dynamic process, and the epiG status of a gene can be subject to changes. In tissue culture, fidelity of maintenance methylation in mammalian cells was detected to be between 97% and 99.9%, and de novo methylation activity was as high as 3–5% per mitosis [Pfeifer et al., 1990; Riggs et al., 1998]. It is important to note that epiG patterns are not established chaotically; there is a significant continuity of epiG patterns during mitotic divisions, but there is also quite significant epiG variation across cells, and cells belonging to the same organism and performing similar functions may exhibit quite different DNA-based and chromatin modifications. EpiG metastability also applies to the germline. During the maturation of the germ line, gametes seem to be reprogramming their epiG status; i.e., epiG signals are erased and a new epiG profile is established [Li, 2002]. Interestingly, not all epiG signals are erased during the gametogenesis, and such epigenetically determined features can be transmitted from one generation to another [Rakyan et al., 2001], which blurs the demarcation between epiG- and DNA sequence-based traits. Why only some epiG signals can be transmitted from one generation to another, while others are erased during gametogenesis, is unclear.
EPIGENETICS AND NON-MENDELIAN FEATURES OF BD
The primary role of epiG factors in regulation of gene activity represents one of the two critical aspects that make epiGs very relevant to human morbid biology. Traditionally, DNA sequence variation is thought to be the main endogenous disease factor in human diseases that exhibit a heritable component. It is important to understand, however, that impeccable DNA sequence is not a sufficient condition for the normal functioning of the genome. DNA sequences have to be organized in a way that provides the appropriate expression of the required genes in the specific type of cells. Dysregulation of gene activity and deviations from the normal expression pattern can be as detrimental to a cell as mutant DNA sequences resulting in dysfunctional proteins.
Dysregulation of gene activity and deviations from the normal expression pattern can be as detrimental to a cell as mutant DNA sequences resulting in dysfunctional proteins.
Insufficient amount of a structurally perfect protein may cause pathological events that are indistinguishable from those initiated by a mutant protein. An example of an extreme scenario is a complete epiG inactivation of the gene and DNA sequence mutation generating a stop codon. In the first case, there is no protein, and in the second case, the protein is truncated and cannot perform the required function in the cell. In a similar way, hyperactivity of genes would result in overproduction of some receptors and enzymes and also have a negative impact on a cell and an organism.
Another epiG aspect that makes epigenetics so relevant to the understanding of molecular mechanisms of various non-Mendelian features of complex diseases is the partial epiG stability, or metastability. As it has been discussed in the above section, epiG regulation of genes is in constant flux, unlike the DNA sequence (especially the coding genome), which basically does not change during the life of an individual (with rare cases of somatic mutations, mitotic recombination).
Shifting some of the emphasis from DNA sequence variation as the main endogenous cause of a complex disease to the putative epiG dysregulation provides a new platform for understanding a series of features of complex diseases that cannot be explained by the DNA sequence variation [Petronis, 2001]1. The most straightforward example is monozygotic (MZ) twin discordance, one of the major mysteries in complex traits. The rate of MZ twin discordance in BD across different twin studies varies between 30% and 70% [Bertelsen et al., 1977; Jones et al., 2002]. How can two genetically identical individuals who live in a similar environment exhibit major phenotypic differences? Although the phenomenon of twin discordance has been known for nearly a century, understanding of its molecular origin has not progressed very far [Martin et al., 1997]. From the epiG point of view, the molecular cause of MZ twin discordance lies in the significant epiG differences accumulated over numerous cell divisions during development and aging [Petronis et al., 2003]. Such epiG differences may cause differential epiG regulation of disease-relevant genes that would result in reaching the threshold of clinical symptoms in only one of two identical twins.
Age-dependent epiG changes may shed a new light on the relatively late age of onset in BD. Unlike most Mendelian disorders, which present with clinical symptoms before puberty, the age of onset in most BD cases is in the twenties [Goodwin and Jamison, 1990]. According to the epiG theory, epiG status of a disease gene(s) may be misregulated to some extent from birth that has no major pathological impact on the functioning of a cell for years or decades. The disease occurs when the “epiG” clock of a gene reaches the critical level of epiG misregulation (epimutation) that results in clinical symptoms and syndromes. It is also interesting to note that BD exhibits incidence peaks at ages 20–30 and 45–50 [Hamilton, 1989; Goodwin and Jamison, 1990, p. 132]. The ages of the increased incidence seem to coincide (or follow) major hormonal rearrangements in the organism. In this respect, changes in the hormonal milieu during maturation and involution can substantially affect regulation of genes, at least to some extent this is achieved via their epiG modifications. Furthermore, males and females exhibit differences in terms of susceptibility to and course of BD [Goodwin and Jamison, 1990, p. 139–140, 164–168; Leibenluft, 1996; Seeman, 1997]. One of the working hypotheses to explain this phenomenon could be that sex hormone-mediated changes in epiG regulation result in differential epiG modification of critical genes.
EpiG fluctuation in regulation of genes' activity may explain clinical remissions and relapses, and even occasional recoveries in BD.
EpiG fluctuation in regulation of genes' activity may explain clinical remissions and relapses, and even occasional recoveries in BD.
Particularly interesting is the presence of two opposite disease phenotypes, mania and depression, one of the key clinical attributes of BD that paradoxically has been rarely investigated in molecular studies of BD. Depressive and manic cycles, or phases, may last up to a year, and between acute phases the patients make a nearly complete recovery [Hamilton, 1989]. The fluctuating course and sometimes recovery are consistent with dynamic changes in the epiG regulation of genes, much more so than the stable DNA sequences.
Parent-of-origin effects present with differential risk to the offspring to be affected with a disease depending on which of the two parents is affected with that specific disease; this has also been documented in BD [Grigoroiu-Serbanescu et al., 1995; McMahon et al., 1995]. In molecular genetic studies, parental effects present with co-segregation of the disease with either maternal or paternal alleles and/or differences in the lod scores in the families with maternal vs. paternal transmission of the disease [Nothen et al., 1999; reviewed in Petronis, 2000; Schulze et al., 2003]. Parental origin effect is one of the classical epiG mechanisms of differential regulation of gene activity, called genomic imprinting.
EpiG metastability may shed a new light on the origin of sporadic and familial cases of complex diseases like BD. As a rule, the clinical phenotypes of sporadic cases are indistinguishable from familial ones. Meiotically stable epimutations can segregate like typical Mendelian factors. The epiG metastability may represent familial cases of a disease, while epimutations that are erased during gametogenesis may be the cause of sporadic cases. In summary, the epiG theory puts the emphasis on epiG changes (or absence of such changes) in specific genomic loci during the meiosis, while the traditional DNA sequence-oriented paradigm infers that different genes are operating in familial and sporadic cases (major genes and minor genes, respectively).
In summary, the dynamic epiG inheritance system may provide a new explanation for a number of features of BD, and epiG mechanisms can be a common denominator for the wide variety of epidemiological and clinical findings of this disease.
The dynamic epiG inheritance system may provide a new explanation for a number of features of BD, and epiG mechanisms can be a common denominator for the wide variety of epidemiological and clinical findings of this disease.
In addition, molecular epiG studies may shed a new light on the understanding of various controversial issues in the traditional genetic studies of BD. Two cases in which the potential role of epiG is a factor are discussed below using the examples of genetic studies of BD on chromosomes 11p and X.
BD AND CHROMOSOME 11p
The short arm of chromosome 11 has been one of the primary targets in genetic linkage and association studies of BD. Chromosome 11p was one of the first chromosomes subjected to the genetic linkage analysis of BD using molecular genetic markers. In 1987, strong evidence for linkage (lod score of ∼5) of BD to two genes on 11p15, namely, Harvey-RAS (HRAS) and insulin (INS) in the Old Order Amish kindred, was reported [Egeland et al., 1987]. Inclusion of two additional branches of the pedigree and update of the diagnoses, however, resulted in a significant decrease of the lod score that was recalculated to be below the critical value of linkage [Kelsoe et al., 1989]. The most dramatic effect on the lod scores in the core pedigree resulted from the updated diagnostic information on two family members who became ill after the original study (IV-20 and III-21). Altogether, these changes lowered the lod scores to 1.03 (theta = 0.17) and 1.75 (theta = 0.14) for HRAS and INS, respectively.
The initial controversial finding sparked a series of studies using parametric and nonparametric genetic linkage as well as association for candidate genes, and more recently applying the transmission disequilibrium approach. Numerous groups tested BD families from different ethnic backgrounds and concluded that chromosome 11p is not linked to BD [Neiswanger et al., 1990; Detera-Wadleigh et al., 1994; De bruyn et al., 1994a; Maziade et al., 2001]. Some findings, however, were somewhat promising. For example, BD families from Iceland and the UK showed moderate evidence for linkage to tyrosine hydroxylase gene (TH) in which a two-locus admixture analysis with chromosome 21 markers revealed an overall lod score of 3.87 [Smyth et al., 1997]. Two large Costa Rican pedigrees generated lod scores of ∼2.00 for 11p markers (D11S929, D11S1392, D11S1312), although located at 11p13-p14, tens of megabases centromeric to 11p15 [McInnes et al., 1996].
In terms of genetic association studies, two candidate genes, namely, the gene for tyrosine hydroxylase (TH) and the gene for dopamine D4 receptor (DRD4), have been subjected to a thorough investigation in BD. Tyrosine hydroxylase is a rate-limiting enzyme that converts phenylalanine to dopamine, and this provides excellent rationale for genetic studies of TH in neuropsychiatric disorders. The tetranucleotide TCAT repeat variants in the first intron exhibits a quantitative silencing effect on the regulation of TH expression [Meloni et al., 1998; Albanese et al., 2001]. Although several association studies of TH and BD looked promising [Leboyer et al., 1990; Serretti et al., 1998], combined data sets and meta-analyses that included hundreds of matched case-control pairs were negative in both the total sample and the geographical subgroups [Souery et al., 1999]. Another combined study also detected no association of BD with two TH polymorphisms (intron 1 tetranucleotide and PstI); however, weak association (P = 0.047) was observed in the unipolar sample with the TH-PstI polymorphism [Furlong et al., 1999].
Genetic association studies of DRD4, another player of the catecholamine system, with BD are also quite controversial. The 48-bp repeat polymorphism in the third exon of DRD4 showed a trend toward an excess of the two-repeat allele in the individuals affected with major depression compared to controls [Serretti et al., 2001]. No association of DRD4 with affective disorders, however, was detected in the case-control study using a sample from the relatively genetically homogenous Chinese population [Li et al., 1999]. In family-based association studies, no evidence for DRD4 transmission disequilibrium in BD families was detected by Italian researchers [Bocchetta et al., 1999], but our group found a significant excess of DRD4 four-repeat allele transmissions to the BD probands [Muglia et al., 2002]. Particularly relevant to this article was the observation that the biased transmission derived almost exclusively from the maternal meioses (P = 0.009 for maternal alleles, P = 0.46 for paternal alleles).
As has typically been found in psychiatric genetics, both linkage and association studies of chromosome 11p markers and BD led to controversial findings, and despite significant effort, the role of chromosome 11p genes in etiopathogenesis of BD remains unclear.
As has typically been found in psychiatric genetics, both linkage and association studies of chromosome 11p markers and BD led to controversial findings, and despite significant effort, the role of chromosome 11p genes in etiopathogenesis of BD remains unclear.
At least part of such inconsistencies may be related to complex epiG regulation on chromosome 11p. The tip of chromosome 11p is one of the most interesting regions of the genome from the epiG point of view. Various epiG mechanisms that may have an impact on chromosome 11p genes and be involved in BD are described below using the studies of INS analysis in type 1 diabetes as an example.
EPIGENETICS OF CHROMOSOME 11p: GENOMIC IMPRINTING, TISSUE SPECIFICITY, epiG HETEROGENEITY OF GENETIC ALLELES, AND PARAMUTAGENESIS
Over the last decade the gene encoding insulin (INS) has been one of the primary targets in the genetic studies of type 1 diabetes. There is no DNA sequence variation in the coding region of INS, and the source of interest derives from the promoter region containing embedded variable-number tandem repeats (VNTRs) that is located 600-bp upstream from the INS transcription start site [Bell et al., 1982]. It has been consistently detected that VNTR class I alleles (smaller number of repeats) are associated with type 1 diabetes, while class III alleles (larger number of repeats) are protective. The class I alleles represent a common allele among Caucasians (allele frequency of 0.71) [Bennett et al., 1996], which means that about 50% of Caucasians are homozygous for class I alleles; however, only decimal percent points develop type 1 diabetes. Parental origin effects at the INS locus are possible since family-based studies showed that probands inherited their class I VNTR alleles predominantly from the fathers [Julier et al., 1991]. Further studies, however, led to quite controversial results demonstrating both paternal [e.g., Polychronakos et al., 1995] and maternal [e.g., Bennett et al., 1996] effects. The class I/I homozygous genotype is associated with a 2- to 5-fold increased risk of developing type 1 diabetes, while class III alleles are dominantly protective [Bell et al., 1984; Raffel et al., 1992; Bennett et al., 1995]. Several studies detected that class III VNTR alleles were associated with 2- to 3-fold higher INS mRNA levels than class I human in embryonic and postnatal thymi [Pugliese et al., 1997; Vafiadis et al., 1997], which seems to be critical for the efficient eradication of insulin-aggressive T cells. The situation is the opposite in the pancreas: carriers of class I alleles demonstrated higher level of steady-state INS mRNA levels in comparison to carriers of class III alleles [Bennett et al., 1996; Vafiadis et al., 1996], and class III alleles are a risk factor to type 2 diabetes [Huxtable et al., 2000]. Subdivision into risk and protective alleles in type 1 diabetes, however, is not universal, and some class I alleles are protecting [Bennett et al., 1995, 1997]. In a similar way, some class III alleles showed no INS expression in the embryonic thymus [Pugliese et al., 1997; Vafiadis et al., 1997], and therefore such alleles should be predisposing to type 1 diabetes. The reasons of the intraclass heterogeneity of class III alleles cannot be explained by DNA sequence variation [Vafiadis et al., 2001]. In terms of class I exceptions, it was detected that class I alleles do not predispose to the disease when paternally inherited if the father's untransmitted allele belongs to class III [Bennett et al., 1997].
The epiG Perspective
A number of the above observations can be related to epigenetics. Parent effects of the INS VNTR transmission to type 1 diabetes patients may be related to genomic imprinting, an epiG phenomenon whereby genes are expressed or not depending on their parental origin [for reviews, see Hall, 1990; Barlow, 1995; Horsthemke et al., 1999]. INS is situated in the cluster of imprinted genes on chromosome 11p15 and surrounded by differently imprinted genes [Maher and Reik, 2000]. Evidence for imprinting of INS in the extraembryonic tissues was recently documented [Moore et al., 2001]. In terms of why the data on parental origin effects at INS are controversial, polymorphic genomic imprinting, i.e., imprinting presented in only a proportion of individuals, may be one of the possible answers. Examples of such imprinting are the genes encoding insulin-like growth factor II (IGF2) [Xu et al., 1993] and Wilms tumor (WT1) [Jinno et al., 1994], both located on the 11p15 imprinting cluster. In a similar way to INS, polymorphic imprinting may include DRD4 and other genes in BD, and such parental origin effect studies across different populations may not necessarily be consistent. It is interesting to note that an insulator element was identified between oppositely imprinted genes in the mouse studies [Bell and Felsenfeld, 2000]. Mutations in the insulator region caused disruption of the normal imprinting pattern [Kanduri et al., 2000]. If such insulators do not perform their function efficiently, the wave of epiG modification from the neighboring imprinted genes may spread to the surrounding regions that would result in epiG dysregulation.
It has to be admitted, however, that there is no full consensus in the scientific community on the origin and biological role of genomic imprinting. An innovative hypothesis has been developed by Sapienza's group suggesting that differential expression of several dozen genes is just a visible tip of the imprinting iceberg. The hidden part of such an iceberg represents genome-wide epiG differences in maternal and paternal chromosomes that ensure proper pairing and recombination of homologous chromosomes during meiosis [Pardo-Manuel de Villena et al., 2000]. This theory suggests that parental origin effects may result from nonrandom segregation of homologous chromosomes, i.e., transmission ratio distortion [Pardo-Manuel de Villena et al., 2000]. Failure to detect parent-of-origin specific monoallelic expression on 11p15 in BD does not necessarily mean that findings of parental differences in genetic studies [e.g., Muglia et al., 2002] were type 1 error. The next experimental step would be to investigate the rates of meiotic recombination at the specific locus, e.g., 11p15, in the affected individuals vs. unaffected family members. Changes in meiotic recombination have been rarely analyzed in human morbid biology [Petronis, 1999].
Tissue-specific differences (thymus vs. pancreas) of INS VNTR allele expression is also likely to be of epiG origin in that expression of the same gene across tissues depends on differential modification of DNA and chromatin. Although it is not completely clear what determines tissue-specific gene expression patterns, epiG differences across tissues suggest that epiG mechanisms are operating in these processes (see the “Basic Principles of epiG Regulation” section). It is important to note that INS class I alleles are a risk factor for type 1 diabetes, while class III alleles increase susceptibility to type 2 diabetes. In a similar way, DRD4, TH, or any other genes may exhibit epiG differential expression in different brain regions, and some alleles of such genes may be predisposing to one subtype of BD (e.g., early onset), while other alleles of the same genes may increase the risk to another subtype of BD (e.g., late onset). It is evident that if genetic studies are performed in an undifferentiated sample of BD patients, and depending on which subtype is predominant, evidence for association can be seen for different alleles.
The observation that the same class INS alleles may behave differently (act as both risk and protective factors) is consistent with epiG variation within the group of genetic alleles. In our studies of the dopamine D2 receptor gene (DRD2), we detected that DRD2 methylation patterns exhibit numerous epiG differences across individuals [Popendikyte et al., 1999]. Although only a few human genes have been subjected to a detailed epiG study, all the logic of epiG metastability during gametogenesis and pre- as well as postnatal development suggests that the overwhelming majority of human genes should exhibit unique patterns of epiG regulation. In this light, division of genes based on their DNA sequence variation (e.g., several DNA sequence types of TH or DRD4 alleles) represents an incomplete characteristic of the gene. It is very possible that there is a wide epiG variety for each gene on chromosome 11p (and any other genomic region) and only specific epialleles are the real players in BD. In summary, in addition to DNA sequence polymorphisms, variation of epiG patterns—epiG polymorphisms—should be taken into account and epiG peculiarities must be investigated along with the DNA sequence variation of the genes.
The molecular mechanism of why some class I alleles do not predispose to the disease when paternally inherited if the father's untransmitted allele belongs to class III [Bennett et al., 1997] may also be of epiG origin. More specifically, this finding points to the so-called paramutagenic allelic interaction during meiosis, similar to that detected in plants. Paramutation refers to a directed, heritable alteration of one allele caused by its interaction with a homologous allele of the same gene, and this phenomenon has been predominantly investigated in plants [Patterson et al., 1993]. An example could be the b gene of maize that encodes a transcriptional activator of anthocyanin pigment biosynthetic genes [Patterson et al., 1993]. Some b alleles may be subjected to paramutagenesis, which converts B-I (intensely pigmented plant) into B′ (weakly pigmented plant) in the B′/B-I heterozygote, such that all progeny receive the B′allele. The new B′ (which was B-I in the previous generation) is weakly pigmented and fully capable of changing another B-I allele into B′. B′ allele acts in trans to suppress the transcription of B-I, with transcription remaining low in subsequent generations, even when the original B′ allele segregates away. Despite dramatic differences in phenotype and transcription of B′ and B-I, there is no evidence for DNA rearrangements or any other changes in sequence of the two alleles. There is increasing evidence that paramutagenesis is an epiG phenomenon [Hollick et al., 1997].
What if paramutagenetic effects are not limited to the INS region? Paramutable alleles of other human genes would exhibit phenotypes similar to the ones caused by their paramutagenic counterparts. The idea that two different genetic alleles may interact during the meiosis during which one of them is epigenetically modified according to the epiG mold of the other compromises the very essence of Mendelian segregation. Both alternative alleles will determine similar phenotypes. Although still speculative, this epiG mechanism may be considered in the key individuals whose genotypes determine the outcome of linkage analysis. As it has been mentioned above, the most dramatic effect on the lod scores in the Old Order Amish core pedigree resulted from the updated diagnostic information on two family members (IV-20 and III-21) who became ill after the original study. Although the two members did not inherit the putative risk genetic alleles in the traditional sense and therefore were not supposed to get affected with BD, the above-described paramutagenic mechanisms operating at chromosome 11p15 may account for the inconsistent findings.
BD AND X CHROMOSOME
The idea that there is an X-linked subtype of BD dates back to at least the 1930s (the disease was then called manic-depressive illness), and this originated from the observation of an excess of affected females and a deficiency of male-to-male transmission. The finding was supported by evidence for linkage between manic-depressive illness and color blindness (protanopia and deuteranopia) (Xq28) [Baron, 1977; Mendlewicz et al., 1979], although the evidence for linkage was not universal [Gershon et al., 1979, 1980]. Eventually glucose-6-phosphate dehydrogenase (G6PD), a biochemical marker encoded by a gene on Xq28, was subjected to linkage studies and revealed a lod score of 4.32 in a large family of Persian Sephardic Jewish origin segregating both BD and G6PD deficiency [Mendlewicz et al., 1980]. The interest of a role for the X chromosome in BD peaked when a maximum lod score ranging from 7.52–9.17 (depending upon the linkage parameters) was found in BD linkage studies with color blindness and G6PD [Baron et al., 1987]. But again, like the chromosome 11p15 case in the Old Order Amish kindred, extension and reevaluation of pedigree data, including new individuals, diagnostic follow-up, and analysis using more informative DNA markers, demonstrated greatly diminished support for linkage to Xq28 [Baron et al., 1993]. Since then, the priorities in the search for BD genes shifted to other chromosomes [Berrettini, 2001; Craddock and Jones, 2001]; however, occasional reports of putative evidence of linkage to Xq26-28 markers to BD were presented [De bruyn et al., 1994b; Stine et al., 1997]. Among the more prominent findings of the last decade, there is evidence for linkage with a maximum lod score of 3.54 at the marker DXS994 (Xq26) in a large bipolar Finnish kindred [Pekkarinen et al., 1995].
The results of the above studies suggest that a gene or genes predisposing to BD may be localized on the X chromosome in a subgroup of BD cases; however, the findings thus far are quite controversial and elucidation of concrete BD genes in the near future does not look very realistic. It is somewhat surprising that epiG regulation of the X chromosome has been ignored in the studies of human genetic diseases, including BD. Two specific epiG aspects of X chromosome that may shed some new light on the traditional genetic studies will be discussed below: skewed X inactivation and heterogeneity of expressed genes on the inactivated X chromosome.
EPIGENETICS OF X CHROMOSOME: SKEWED X INACTIVATION AND VARIABLE EXPRESSION OF GENES ON THE INACTIVATED X
Skewed X Inactivation
X chromosome inactivation occurs in the late blastocyst stage of embryogenesis at about the 32- to 64-cell stage of female mammalian development to transcriptionally silence one of the two X chromosomes, thereby achieving dosage compensation with karyotypically normal males [Willard, 2000]. X inactivation requires several coupled processes—initiation, spreading, and maintenance of inactivation [Willard, 2000]. The master regulatory switch controlling X inactivation in humans is the XIST gene (Xist in mice), which is expressed exclusively only on the inactive X chromosome and which accumulates over the entire length of the X chromosome [Brockdorff, 2002]. XIST transcripts start a cascade of events that invariably silence the X chromosome from which such transcripts are synthesized. X inactivation to a significant extent is an epiG phenomenon. There is evidence from mice studies that Xist RNA recruits histone methyltransferases and deacetylases that change chromatin structure from active to inactive [Li, 2002]. In maintenance of X inactivation, Xist RNA is not necessary; however, DNA methylation and histone deacetylation seem to be playing a critical role [Brockdorff, 2002].
The mechanisms determining which of the two (in normal cases) female X chromosomes is to be inactivated are not completely clear. Although traditionally it was thought that such selection is random, there is increasing evidence that the process at least in some cases is predetermined. Naumova et al. [1996] reported a family in which there is heritable skewing of X inactivation [Naumova et al., 1996; Plenge et al., 1997]. Apart from genetic determination of X inactivation patterns, there are numerous examples of skewing of X chromosome that results from selection against cells with either imbalanced gene expression or mutations that affect cell growth [Belmont, 1996; Puck and Willard, 1998; Willard, 2000]. The relative ratio of the two cell populations in a given female is frequently referred to as the X inactivation pattern. It was detected that approximately 50% of female carriers of X-linked mental retardation exhibited skewed X inactivation (the ratio between two X chromosomes in the active state was 80:20% or higher) [Plenge et al., 2002]. Nonrandom X chromosome inactivation has also been observed for a number of disorders, including several X-linked immunodeficiencies, Lesch-Nyhan disease, incontinentia pigmenti, focal dermal hypoplasia, and adrenoleukodystrophy [Willard, 2000]. Skewed X inactivation may cause discordance for X-linked recessive diseases in female MZ twins [Van den Veyver, 2001]. In cancers, skewed X inactivation has been used as a tool to examine the clonal origin of neoplasias in females; i.e., if a tumor arose from the single cell after the time of X chromosome inactivation, then it will have the same X chromosome active in all cells [Brown, 1999]. In the normal population of females without a family history of X-linked disorders, 5–20% of apparently normal women have constitutional skewing of X inactivation [Belmont, 1996]. According to other authors, 30–40% of females exhibit ratios of 60:40% or more, and 10% of normal females demonstrate even more extreme ratios (80:20% or greater) [Willard, 2000].
Genes That Escape X Chromosome Inactivation
To further complicate the issue, not all genes on the inactive X chromosome are subject to inactivation. An analysis of ∼200 X-linked transcribed sequences showed that over 10% of corresponding genes escape X inactivation and exhibit biallelic expression [Carrel et al., 1999]. As more studies are being performed, the number of gene escapees is increasing [L. Carrel, personal communication]. Such genes are predominantly but not exclusively located on Xp. The number of clear-cut cases, where all analyzed cell lines would exhibit an unequivocal pattern of biallelic expression of a specific gene, is smaller on the long arm of the X chromosome. Xq genes, however, exhibit a higher degree of variation in terms of mono- vs. biallelic expression of X chromosome genes across different cells lines, which points at putative interindividual variation for the escape from inactivation [Carrel and Willard, 1999]. Variegation of active and inactive domains on the otherwise inactivated X chromosome most likely are of epiG origin.
epiG Regulation of X Chromosome and BD
Both skewed X inactivation and differences in the genes escaping X chromosome inactivation are immediately relevant to genetic studies of BD (as well as numerous other complex diseases).
Skewed X inactivation indicates that since the ratio of expressed genes on parental X chromosomes deviates from the expected 50–50%, the sets of expressing X chromosome genes in such individuals can be quite different. Results of linkage and association studies will significantly depend on the proportion of females with skewed X inactivation patterns. For example, two sisters inherited the same risk factor on the X chromosome for BD, but only one of them will become affected with the disease because their X inactivation patterns are different. Since skewed X inactivation is quite a common phenomenon among females, it is evident that epiG analysis of X inactivation patterns is absolutely necessary in order to unequivocally address the issue of the role of X chromosome in a disease. In addition, comparative analysis of X inactivation in affected individuals and control population may also be provide some useful information of the role of X chromosome. Differences in the rate of X skewing in affected individuals vs. controls would indirectly argue that this chromosome is involved in the etiology of a disease. Such analysis may be of value if linkage and association studies are not sensitive enough to identify genetic risk factors of medium- to small-size effects. Furthermore, exploratory linkage and association studies using the sample stratified into subgroups of skewed X (assumes presence of chromosome X-related disease factors) and non-skewed X (assumes absence of chromosome X-related disease factors) may be helpful in addressing the issue of genetic heterogeneity and increasing the power of genetic analyses.
Genes escaping X inactivation add another level of complexity on top of skewed X inactivation in genetic linkage and association studies. This is especially important in the cases when such X chromosome genes exhibit varying patterns of expression/nonexpression across different individuals.
epiG STUDIES OF BD: NEW COMPLEXITIES
Experimental epiG studies of complex diseases, however, are making the first steps (with the exception of cancer), and there are numerous logistic, methodological, and technical issues that will have to be addressed. First of all, for epiG studies, a tissue where the disease process originates is absolutely necessary. This means that in the case of BD, brain tissues should be investigated. Even if brain tissues are available, the next immediate complexity is to decide on what specific region of the very complex organ should be investigated, or even maybe selection of a brain region is not sufficient, and it is necessary to focus on population of some specific cells. The second group of problems is related to technical complexities. Chromatin modification (histone acetylation, methylation) is not stable in human tissues with a quite long postmortem interval, and therefore, epiG studies of human brain are limited to DNA modification analyses. Mapping of metC using the more traditional methylation-sensitive restriction enzyme-based methods are of low informational value, while the currently gold standard technique of bisulfite modification [Frommer et al., 1992; Clark et al., 1994] is labor intensive. In order to generate a reliable DNA modification profile, it is necessary to sequence a large number of clones representing each DNA sample. Most of the epiG analyses performed thus far using the bisulfite modification approach have been limited to several hundred base pairs of a gene of interest. The third complexity deals with formal analysis of epiG profiles and comparison of such profiles across individuals. At the moment, it is not a problem to differentiate all-or-nothing situations in epiG modifications, e.g., gene pairs where one of the genes is genomically imprinted or cases of epiG shutdown of tumor suppressor genes in cancer. BD, as well as other complex diseases, may present with less straightforward, “many shades of gray” epiG changes. Despite these complexities, the epiG aspect of BD cannot be ignored anymore. All of the examples delineated in this article provide strong evidence that epiG aspects should be seriously considered when dealing with such a complex phenotype as BD. EpiG hypotheses must be formulated and tested experimentally. Some genes of interest in psychiatric diseases have already been investigated from the epiG point of view [Petronis, 1999; Chen et al., 2002; Tremolizzo et al., 2002; Petronis et al., 2003].
Acknowledgements
I thank Dr. Irv Gottesman (University of Minnesota, Minneapolis), Dr. M. Seeman (Centre for Addiction and Mental Health, Toronto, Canada), Dr. L. Carrel (Penn State University, Hershey, Pennsylvania), Dr. C. Brown (University of British Columbia, Vancouver, Canada), and Mr. Z.A. Kaminsky (Centre for Addiction and Mental Health, Toronto, Canada) for their valuable comments and advice.
