Human monozygotic twins and other genetically identical organisms are almost always strikingly similar in appearance, yet they are often discordant for important phenotypes including complex diseases. Such variation among organisms with virtually identical chromosomal DNA sequences has largely been attributed to the effects of environment. Environmental factors can have a strong effect on some phenotypes, but evidence from both animal and human experiments suggests that the impact of environment has been overstated and that our views on the causes of phenotypic differences in genetically identical organisms require revision. New theoretical and experimental opportunities arise if epigenetic factors are considered as part of the molecular control of phenotype. Epigenetic mechanisms may explain paradoxical findings in twin and inbred animal studies when phenotypic differences occur in the absence of observable environmental differences and also when environmental differences do not significantly increase the degree of phenotypic variation.
Identical human twins have been a source of superstition and fascination throughout human history, from Romulus and Remus, the mythical founders of Rome, to movies such as Cronenberg's ‘Dead Ringers’, to the twin paradox in the theory of special relativity (1). For biologists and psychologists, twins have been an important resource for exploring the etiology of disease and for understanding the role of genetic and environmental factors in determining phenotype, and this fundamental question was first enunciated in its alliterated form by Galton in the 19th century (2). The relative contribution of nature versus nurture can be estimated by comparing the degree of phenotypic similarities in monozygotic (MZ) versus dizygotic (DZ) twins. MZ twins arise from the same zygote, whereas dizygotic twins arise from a pair of separate eggs, fertilized by two different sperm. As a result, MZ twins have the same chromosomal DNA sequence, except for very small errors of DNA replication after the four to eight cell zygote stage. MZ twins share all of their nuclear DNA, whereas DZ twins share only 50% of DNA sequence variation, on average. Therefore, the degree of genetic contribution to a given phenotype can be estimated from the comparison of MZ to DZ concordance rates or intra-class correlation coefficients. Traits that show higher MZ versus DZ similarity are assumed to have a genetic component because the degree of genetic sharing and the degree of phenotypic similarity are correlated. The amount of genetic contribution can be expressed as the ‘heritability’ (h2), which is calculated in various ways (e.g. as twice the difference between the MZ and DZ concordance rates) (3,4).
Most (if not all) common human diseases show a significant heritability by this definition; however, while MZ twins appear virtually identical, they are often discordant for disease. For example, the heritability of schizophrenia is variously reported to be in the range of 0.70–0.84, on the basis of an MZ twin concordance of ∼50% and a DZ twin concordance of 10–15% (5–7). A heritability figure of 0.8 (or 80%) seems to imply that the genetic contribution to disease susceptibility is the main component of risk. Yet, half of MZ twin pairs, in the case of schizophrenia, do not share the disease. The situation is similar for virtually all complex non-Mendelian diseases in which there is clearly some appreciable degree of heritable risk, yet a significant proportion of MZ twin pairs is discordant for the disease. Figure 1 shows the MZ and DZ concordance rates for some common behavioral and medical disorders.
Although heredity clearly influences disease risk, the substantial discordance between MZ twins indicates that chromosomal DNA sequence alone cannot completely determine susceptibility (8). The imperfect disease concordance in MZ twins is an example of a more general phenomenon: (i.e. phenotypic differences between or among genetically identical organisms). These differences have usually been attributed to the effects of environment (the ‘non-shared environment’ in the case of MZ twins) (9), as a default explanation for variation that remains after genetic effects are accounted for. It is not easy to measure empirically the amount of non-genetic variation that is due to environmental factors. There are examples of significant environmental effects on disease risk, such as smoking and lung cancer (10), but direct evidence of other measured environmental effects on phenotype is rare. In addition, there is an increasing body of experimental evidence suggesting that the generally accepted assumption—variation not attributable to genetic factors must therefore be environmental—may require revision. This article will review human twin and animal data that highlight paradoxical findings regarding the contribution of heredity and environment to phenotype, followed by a reinterpretation of these experiments that incorporates epigenetic factors.
STUDIES OF MZ TWINS RAISED APART OR TOGETHER: THE EPHEMERAL ROLE OF ENVIRONMENT
One of the landmark studies in human twin research that challenges the received importance of environment is the Minnesota Study of Twins Reared Apart, in which detailed physical and psychological assessments were conducted longitudinally in over 100 MZ and DZ pairs of twins who had been reared apart since early childhood (11,12). In a variation of the traditional twin study comparing MZ twins reared together (MZT) with DZ twins reared together, the Minnesota study compares MZT with MZ twins raised apart (MZA). This study design allows comparisons between genetically identical MZ twin pairs who have been raised in a shared environment, at least as similar as for any two siblings, and those who have been raised in different homes, cities and states. Thus, the degree of dissimilarity between the MZT and the MZA pairs can be assumed to be the result of different environments (13). A series of tests were administered simultaneously to each pair of MZA and MZT twins, and the correlations of their scores on each scale were calculated and compared with test–retest correlations as a measure of the reliability of each scale. The intra-class correlation (R) within pairs of MZA (RMZA) and MZT (RMZT) was then expressed as a ratio (RMZA/RMZT). Surprisingly, the correlations within MZT and MZA twin pairs on personality measurements were almost identical (e.g. RMZA=0.50 and RMZT=0.49 on the Multidimensional Personality Questionnaire—MPQ). The RMZA/RMZT ratio for the MPQ was 1.02, compared with 1.01 for fingerprint ridge counts. Out of 22 measurements for which the RMZA/RMZT ratio was reported, 15 had a value over 0.9. In addition to the traits mentioned previously, these included: electroencephalographic patterns; systolic blood pressure; heart rate; electrodermal response (EDR) amplitude in males and number of EDR trials to reach habituation; the performance scale on the WAIS-IQ; the Raven Mill-Hill IQ test; the California Psychological Inventory; social attitudes on religious and non-religious scales and various scales of MPQ (11,12).
The findings of the Minnesota study are generally consistent with other studies of MZA twins. For example, a recent effort to look at the etiology of migraine headaches gathered data from the Swedish Twin Registry and found that susceptibility to migraine was mostly inherited and that the twins separated earlier had even greater similarity in migraine status. For women in particular, migraine profiles were very similar in MZA compared to the MZT group, RMZA=0.58 and RMZT=0.46 (RMZA/RMZT=1.26), whereas for men, the migraine incidence was too low and the confidence intervals for correlations were too wide to be able to draw firm conclusions (14). The Swedish Twin sample was also utilized to explore the factors affecting regular tobacco smoking, and again the results were similar for the two groups of MZ male twins, RMZA=0.84 and RMZT=0.83 (RMZA/RMZT=1.01). For women, the data were more difficult to interpret, RMZA=0.44 and RMZT=0.68 (RMZA/RMZT=0.65), but when more recent cohorts were examined, the rates of smoking in women were similar to men and the heritability was much the same as for men. Smoking rates in women in the early 1900s (the oldest cohort in this sample) were very low and social factors inhibiting smoking in women could result in there being strong local environmental effects, whereas more modern cohorts, with less restrictions on acceptable female behavior, were free to smoke for the same reasons (genetic) as men (15).
Peptic ulcer, a disease that contains an evident environmental component (i.e. exposure to Helicobacter pylori), has also been subjected to an MZT/MZA study design. The results are another example of paradoxical findings in which the contribution of environmental and genetic factors is unclear. Again, using the Swedish Twin Registry, MZ and DZ concordance rates in twins raised together for peptic ulcer were reported to be 0.65 and 0.35, respectively, suggesting that genetic factors are indeed important in vulnerability (heritability 0.62). However, comparisons between MZA or MZT showed RMZA=0.67 and RMZT=0.65 (RMZA/RMZT=1.03), suggesting that the common home environment has little effect on susceptibility to peptic ulcer (16). So, the question that remains after analyzing these data is: what can account for the discordance in MZ twin pairs? If environment were a significant factor, then why is the RMZA/RMZT almost 1 (1.03)?
All the aforementioned examples of the MZT/MZA comparisons, from anthropometric data to psychological traits to pathogen susceptibility, lead to a paradoxical conclusion regarding the role of environmental factors. The fact that the MZ twin concordance (correlation) is well below 100% argues that environment is important. However, different environments in MZA do not result in a higher degree of phenotypic discordance when compared with MZT. These MZT/MZA studies clearly illustrate the core problem that we wish to explore in this paper; though some studies implicate non-genetic factors in susceptibility to disease or other phenotypes, the available data do not support the interpretation that the remaining variation in phenotype is due to environmental factors. Similar inconsistencies regarding the impact of environmental effects have also been detected in the studies of experimental animals.
PHENOTYPIC VARIATION IN GENETICALLY IDENTICAL ANIMALS: ‘THE THIRD COMPONENT’
Some of the questions raised by human twin studies can be re-examined by experimental manipulations of laboratory animals. Animal strains that have been inbred for many generations have almost identical genomes, that is, they are virtually isogenic. True MZ twins can also be generated through in vitro embryo manipulations that provide an opportunity to directly separate the effects of genes from pre-natal environment. Although environmental ‘twins’ do not exist in humans, a close approximation can be created by strictly controlling the environment of laboratory animals in a way that is impossible with humans. At the very least, the effects of constrained versus diverse environments can be quantified to determine the relative contribution of specific environmental factors to phenotypic variation.
In an elegant series of experiments designed to explore the relative contributions of genes, environment and other factors to laboratory animal phenotype, Gartner (17) was able to demonstrate that the majority of random non-genetic variability was not due to the environment. Genetic sources of variation were minimized by using inbred animals, but reduction of genetic variation did not substantially reduce the amount of observed variation in phenotypes such as body weight or kidney size. Strict standardization of the environment within a laboratory did not have a major effect on inter-individual variability when compared with tremendous environmental variability in a natural setting. Only 20–30% of the variability could be attributed to environmental factors, with the remaining 70–80% of non-genetic variation due to a ‘third component…effective at or before fertilization’ (17).
To directly segregate genetic from pre- and post-natal effects, in vitro embryo manipulations in isogenic animals can be performed. In two mouse strains and in Friesian cattle, Gartner artificially created MZ and DZ twins by transplanting divided and non-divided eight-cell embryos into pseudopregnant surrogates. The effect of different uterine environments was tested by transplanting pairs of MZ or DZ embryos into the same or into two different surrogate dams. Pre-natal and post-natal environments were tightly controlled and, most importantly, were equally variable between the isogenic DZ and MZ twin pairs. The variance (s2) of mean body weights and time to reach certain developmental milestones like eye opening, between twin pairs (sb2) and within twin pairs (sw2) was calculated for both MZ and DZ groups. The F-test comparisons for variation between the MZ and DZ groups overall were not significantly different. However, variance within twin pairs (sw2) was significantly much greater for DZ twins than for MZ twins (ranging from P<0.01 to P<0.001 for individual weight measurements). Therefore, despite the fact that all mice were isogenic, and developed in identical pre- and post-natal environments, the MZ twin pairs showed a greater degree of phenotypic similarity among co-twins than did the DZ twin pairs, thus implicating non-DNA sequence—and non-environment—based influences on the zygote at or before the eight-cell stage as the main source of phenotypic variation. Gartner and Baunack (18) referred to this non-genetic influence as the ‘third component’, after genes and environment, the molecular basis of which remains unknown.
The cloning of mammals has recently been accomplished in a variety of species, and these experiments, technical feats in themselves, also present an opportunity to differentiate the effects of chromosomal DNA sequence from other factors that can influence phenotype. Although the offspring of these cloning experiments have the same genome as the donor animals, they exhibit a variety of phenotypic abnormalities that obviously cannot be attributed to genetic causes (19). In some cases, the phenotypes are pathological and represent disease states, whereas other abnormalities are more subtle, suggesting that these observations are relevant to understanding both susceptibility to human complex diseases and variation within a normal functional spectrum (20). The most famous of these cloning experiments was performed with sheep, but along with seemingly healthy lambs, many clone siblings died perinatally as a result of overgrowth, pulmonary hypertension and renal, hepatobiliary and body-wall defects (21). Some cloned mice are susceptible to obesity (22). In addition to higher overall weight, the cloned mice have the same inter-individual variability in weight as non-cloned control mice of the same genetic background (23). Clones in other species also show considerable variation in lifespan and disease phenotypes between genetically identical clones and non-cloned members of that species. This has been reported in pigs (24) and in cattle, where the main post-natal abnormality is musculoskeletal in origin (25).
This list is far from comprehensive and is meant to illustrate our point that significant phenotypic variation, including crossing a threshold to fatal disease, can emerge from animals that have an identical, cloned genetic background, and frequently-occurring differences in mitochondrial DNA cannot be a universal mechanism for a wide spectrum of phenotypic differences. These early examples of cloned animals were subjected to intense scrutiny in highly supervised and controlled environments, yet they still exhibit disease in an inconsistent fashion. If environment were the source of this phenotypic variation, then one would expect the same emergence of disease among non-cloned members of this species, in an even greater extent, because their environment is not usually so tightly constrained. More likely, there are other potential explanations for this variation.
The general conclusion drawn from the previously described experiments is that substantial phenotypic variation may occur in the absence of either genetic background differences or identifiable environmental variation. When genetic sources of variation are excluded, environmental factors are usually considered to be the source of the remaining variation. However, the previously described data do not support this hypothesis. It is easy to see how the environment is often blamed for this non-genetic variation in phenotype. It is difficult to prove that environmental factors are not affecting phenotype. Environmental sources of phenotypic variation can only be excluded by showing that variation persists in a zero-variation environment. Obviously it is difficult to design such an experiment in which environmental variation can be shown to be near-zero, but the studies described previously circumvent this problem. They did so by either directly controlling the degree of environmental variation (as in Gartner's experiments) or by using naturally occurring (human twins) or artificially induced (through in vitro embryo manipulations) controls as comparison groups. In all of these examples, there exists a component of phenotypic variation whose source remains unexplained.
THE EPIGENETIC PERSPECTIVE
Epigenetics refers to DNA and chromatin modifications that play a critical role in regulation of various genomic functions. Although the genotype of most cells of a given organism is the same (with the exception of gametes and the cells of the immune system), cellular phenotypes and functions differ radically, and this can be (at least to some extent) controlled by differential epigenetic regulation that is set up during cell differentiation and embryonic morphogenesis (26–28). Once the cellular phenotype is established, genomes of somatic cells are ‘locked’ in tissue-specific patterns of gene expression, generation after generation. This heritability of epigenetic information in somatic cells has been called an ‘epigenetic inheritance system’ (29). Even after the epigenomic profiles are established, a substantial degree of epigenetic variation can be generated during the mitotic divisions of a cell in the absence of any specific environmental factors. Such variation is most likely to be the outcome of stochastic events in the somatic inheritance of epigenetic profiles. One example of stochastic epigenetic event is a failure of DNA methyltransferase to identify a post-replicative hemimethylated DNA sequence, which would result in loss of methylation signal in the next round of DNA replication (reviewed in 30). In tissue culture experiments, the fidelity of maintenance DNA methylation in mammalian cells was detected to be between 97 and 99.9% (31). In addition, there was also de novo methylation activity, which reached 3–5% per mitosis (31). Thus, the epigenetic status of genes and genomes varies quite dynamically when compared with the relatively static DNA sequence. This partial epigenetic stability and the role of epigenetic regulation in orchestrating various genomic activities make epigenetics an attractive candidate molecular mechanism for phenotypic variation in genetically identical organisms.
From the epigenetic point of view, phenotypic differences in MZ twins could result, in part, from their epigenetic differences. Because of the partial stability of epigenetic regulation, a substantial degree of epigenetic dissimilarity can be accumulated over millions of mitotic divisions of cells in genetically identical organisms. This is consistent with experimental findings in MZ twins discordant for Beckwith–Wiedemann syndrome (BWS) (32). In skin fibroblasts from five MZ twin pairs discordant for BWS, the affected co-twins had an imprinting defect at the KCNQ1OT1 gene. The epigenetic defect is thought to arise from the unequal splitting of the inner cell mass (containing the DNA methylation enzymes) during twinning, which results in differential maintenance of imprinting at KCNQ1OT1. In another twin study, the bisulfite DNA modification-based mapping of methylated cytosines revealed numerous subtle inter-individual epigenetic differences, which are likely to be a genome-wide phenomenon (33). The finding that differences in MZT are similar to MZA, for a large number of traits, suggests that in such twins stochastic events may be a more important cause of phenotypic differences than specific environmental effects. If the emphasis is shifted from environment to stochasticity, it may become clear why MZ twins reared apart are not more different from each other than MZ twins reared together. It is possible that MZ twins are different for some traits, not because they are exposed to different environments but because those traits are determined by meta-stable epigenetic regulation on which environmental factors have only a modest impact.
It is not our intention to argue that environment has no effect in generating phenotypic differences in genetically identical organisms. Rather, we are suggesting that epigenetic studies of disease may help to understand the pathophysiology of, and susceptibility to, etiologically complex, common illnesses. The current method of studying most diseases includes molecular genetic approaches to identify gene-sequence variants that affect susceptibility and epidemiological efforts to identify environmental factors affecting either susceptibility or outcomes. However, epidemiological studies in humans are limited by a number of methodological issues. Obviously, it is unethical to deliberately expose people to putative disease-causing agents in a prospective randomized controlled trial and it is impossible to control human environments in a way that eliminates most sources of bias in epidemiological studies (34). Such designs may be possible in animal studies, but adequate animal models are available for only a small proportion of human conditions. In this situation, epigenetic studies may help identification of the molecular effects of the environmental factors. There is an increasing list of environmental events that result in epigenetic changes (35–41), including the recent finding of maternal behavior-induced epigenetic modification at the gene for the glucocorticoid receptor in animals (42). The advantage of the epigenetic perspective is that, especially in humans, identification of molecular epigenetic effects of environmental factors might be easier and more efficient than direct (but methodologically limited) epidemiological studies.
Epigenetic mechanisms can easily be integrated into a model of phenotypic variation in multicellular organisms, which can explain some of the phenotypic differences among genetically identical organisms. MZ twin discordance for complex, chronic, non-Mendelian disorders such as schizophrenia, multiple sclerosis or asthma could arise as a result of a chain of unfavorable epigenetic events in the affected twin. During embryogenesis, childhood and adolescence there is ample opportunity for multidirectional effects of tissue differentiation, stochastic factors, hormones and probably some external environmental factors (nutrition, medications, addictions, etc.) (39,43) to accumulate in only one of the two identical twins (Fig. 2) (30,33).
Variation in phenotype among isogenic animals can also be attributed to meta-stable epigenetic regulation. Gartner's experiments with controlled versus chaotic environmental conditions showed that non-environmental factors were responsible for the majority of phenotypic variation among inbred animals. Similar observations among cloned animals can be accounted for by epigenetic differences among these animals. The role of dysregulated epigenetic mechanisms in disease is also consistent with the experimental observations in cloned animals. The derivation of embryos from somatic cells, which contain quite different epigenetic profiles when compared with the germline, generates abnormalities of development that can arise from inadequate or inappropriate nuclear programming (22,44,45).
Evidence of epigenetic, non-environmentally mediated sources of variation in genetically identical organisms can be found in the examples of the mouse agouti and AxinFu loci (46,47). The agouti gene (A) is responsible for the coat color of wild-type mice and isogenic heterozygous c57BL/6 Avy/a mice have a range of coat colors from yellow to black (pseudoagouti). The darkness of the Avy/a mice was proportional to the amount of DNA methylation in the agouti locus, with complete methylation in black psuedoagouti mice and reduced methylation in yellow ones (46). Transplantation experiments of fertilized oocytes to surrogate dams demonstrated that color was influenced by the phenotype of the genetic dam, not the foster dam (46). Thus, an obvious phenotype of this isogenic mouse strain is controlled by epigenetic factors that are partially heritable.
Another example of the role of epigenetic mechanisms on phenotype is the murine axin-fused (AxinFu) allele, which in some cases produces a characteristically kinked tail. Like the agouti gene locus, the Axin gene contains an intracisternal-A particle (IAP) retrotransposon that is subjected to epigenetic modification. The methylation status of the long terminal repeat of the IAP in the AxinFu allele correlates with the degree of tail deformity. Furthermore, the presence of the deformity and associated methylation pattern in either sires or dams increases the probability of the same deformity in the offspring (47). These experiments demonstrate both stochastic and heritable features of epigenetic mechanisms on variability in isogenic animals.
The two epigenetic mouse studies described previously as well as experimental data from other species (48) suggest epigenetic signals can exhibit meiotic stability, i.e. epigenetic information can be transmitted from one generation to another. Traditionally, it has been thought that during the maturation of the germline, gametes re-program their epigenetic status by erasing the old and re-establishing a new epigenetic profile. Although the extent of meiotic epigenetic stability remains unknown, the implications are potentially dramatic, blurring the distinction between epigenetic and DNA sequence-based inheritance. The inheritance of epigenetic information and the potential for this to affect disease susceptibility also challenge the dominant paradigm of human morbid genetics, which is almost exclusively concentrated on DNA sequence variation (49). This partial heritability of epigenetic status of the germline may explain the molecular origin of Gartner's ‘third component’ (Fig. 3). The variance within twin pairs (sw2) was significantly lower for isogenic MZ twins than for isogenic DZ twins because MZ twins derived from the same zygote shared the same epigenomic background. DZ animals, however, originated from different zygotes that had different epigenetic backgrounds. This interpretation suggests that epigenetic meta-stability is not only limited to somatic cells but also applies to the germline and that germ cells of the same individuals may be carriers of different epigenomes despite their DNA sequence identity. Additionally, the inherited epigenetic signals have a significant impact on the phenotype despite numerous epigenetic changes that take place during embryogenesis (27,50).
Until recently, it has not been feasible to test these epigenetic interpretations of phenotypic differences directly among genetically identical organisms. Technologies for high-throughput, large scale epigenomic profiling have been developed (51–55), which along with more well-established techniques, such as focused fine-mapping of methylated cytosines using bisulfite modification or identification of histone modification status using chromatin immunoprecipitation, can evaluate epigenetic profiles in a target tissue and permit comparisons of epigenetic profile among different phenotypes. Methods such as these could be applied to genetically identical organisms to determine whether phenotypic differences are indeed correlated with differences in epigenetic profiles and where in the genome the crucial epigenetic signals may be located. The classical genetic phenomena of incomplete penetrance and variable expressivity may in part be explained by differences in epigenetic regulation of certain genes and their expression levels. We now have the experimental tools to test these hypotheses directly and characterize the extent to which epigenetic factors may influence the traditional dyad of genes and environment.
Vogel and Motulsky (56) wisely said that ‘human genetics is by no way a completed and closed complex of theory… [with] results that only need to be supplemented in a straightforward way and without major changes in conceptualization…anomalies and discrepancies may exist, but we often do not identify them because we share the ‘blind spots’ with other members of our paradigm group’ (56, p. 195). The source of phenotypic differences in genetically identical organisms may be one such blind spot among geneticists. Apart from human diseases, various concerns have been raised regarding the limitations of the DNA sequence-based paradigm, and the importance of epigenetic factors has been emphasized. Strohman (57) concluded that the Watson–Crick genetic code ‘which began as a narrowly defined and proper theory and paradigm of the gene, has mistakenly evolved into a theory and a paradigm of life’. In a similar way, Fedoroff et al. (58) stated that ‘our traditional genetic picture…which is concerned almost exclusively with the effect of nucleotide sequence changes on gene expression and function is substantially incomplete’, and that ‘epigenetic factors are significantly more important than it is generally thought’ (58). As seen in other fields of science (59), identification of the areas where inconsistencies or controversies lie may provide new opportunities for re-thinking fundamental laws, lead to new experimental designs, and may even result in major paradigmatic shifts.
We thank Dr Axel Schumacher for his help with drawing figures for this article. This research has been supported by the Special Initiative grant from the Ontario Mental Health Foundation and also by NARSAD, the Canadian Psychiatric Research Foundation, the Stanley Foundation, the Juvenile Diabetes Foundation International and the Crohn's and Colitis Foundation of Canada to A.P.