Evolution of genomic imprinting as a coordinator of coadapted expression

Jason B. Wolf1

Department of Biology and Biochemistry, University of Bath, Bath BA2 7AY, United Kingdom

Edited by Marcus W. Feldman, Stanford University, Stanford, CA, and accepted by the Editorial Board February 12, 2013 (received for review April 3, 2012) Genomic imprinting is an epigenetic phenomenon in which the (21). Second, imprinted appear to modulate a limited expression of a gene copy inherited from the differs from number of types of traits, with most genes having effects on growth that of the copy inherited from the father. Many imprinted genes (especially in relation to the demand for maternal provisioning, appear to be highly interconnected through interactions mediated often via the ) and/or behaviors (17, 22), with behavioral by , RNA, and DNA. These kinds of interactions often favor effects being largely associated with parental and social behaviors the evolution of genetic coadaptation, where beneficially inter- (13, 23–27). Many of the theories for the evolution of genomic acting alleles evolve to become coinherited. Here I demonstrate imprinting arise from or are strongly tied to this apparently limited theoretically that the presence of gene interactions that favor range of influenced by most imprinted genes. For coadaptation can also favor the evolution of genomic imprinting. example, the fact that many imprinted genes appear to influence Selection favors genomic imprinting because it coordinates the prenatal growth and placental traits has been seen as support for coexpression of positively interacting alleles at different loci. Evo- a role of conflict (28, 29) or coadaptation in the origins of im- lution is expected to proceed through a scenario where selection printing (6), as has the role of imprinted genes in social behavior builds associations between beneficial combinations of alleles (e.g., refs. 25, 26) (although it is important to keep in mind that and, if one locus evolves to become imprinted, it leads to selection there are also many cases that do not clearly fit theoretical pre- for its interacting partners to match its pattern of imprinting. This dictions, e.g., ref. 30). Consequently, existing theories are gener- process should favor the evolution of physical linkage between ally consistent with the observation that there is a limited range of interacting genes and therefore may help explain why imprinted traits affected by imprinting (which is not surprising because they genes tend to be found in clusters. The model suggests that, have been motivated to explain these phenotypic effects). How- whereas some genes are expected to evolve their imprinting status ever, existing theories generally do not explain or involve the fi because selection directly favors a speci c pattern of parent-of- clustered nature of imprinted genes (31), although genetic conflict origin-dependent expression, other genes may evolve imprinting has been seen as a potential explanation for the clustering and as a coevolutionary response to match the expression pattern of interfering effects of at the callipyge locus (32). their interacting partners. As a result, some genes will show phe- Thefactthatimprintedgenesaffectalimitednumberoftrait notypic effects consistent with the predictions of models for the types provides ample opportunities for interactions between

fl EVOLUTION evolution of genomic imprinting (e.g., con ict models), but other imprinted genes. Indeed, imprinted genes appear to be highly in- genes may not, having simply evolved imprinting to follow the teractive; for example, a meta-analysis of microarray data (33) has lead of their interacting partners. identified an imprinted gene network (IGN) [also referred to as the Zac1–regulated IGN, where Zac1 is a zinc finger that reg- | recombination ulates apoptosis and cell cycle arrest (34)], where a set of coregu- lated imprinted genes appears to play an important role in enomic imprinting is an epigenetic phenomenon in which modulating embryonic growth. Likewise, a systems study of the Gthe expression of a gene depends on its parent of origin (1). human “interactome” suggests that imprinted genes are tightly Since its discovery, there has been a great deal of interest in un- connected through interactions in the protein interaction network, derstanding the evolutionary processes that could favor such a composing a functionally important “module” of the human peculiar pattern of monoallelic . Consequently, interactome that is particularly intolerant of errors (35). Fur- there is a diversity of potential explanations for the evolution thermore, noncoding RNAs appear to provide opportunities origins of imprinted expression, with most of the currently favored for interactions between loci within imprinted gene clusters “ ” theories relying on some asymmetry that generates selection (16), as do direct regulatory interactions within and between favoring silencing of either the paternally or the maternally inheri- chromosomes, which show a strong overrepresentation of imprin- ted alleles (2, 3). Such conditions favoring genomic imprinting ted domains (36). For example, locally acting trans interactions fl can arise most notably from con ict between the maternal and involving RNA are responsible for the callipyge phenomenon in paternal over maternal investment (4, 5), but can also (37). Interactions between imprinted genes located in dif- arise from selection favoring coadaptation of gene expression in ferent chromosomes have also been reported (38, 39). (or potentially fathers) and their offspring (6), patterns Imprinted genes are often coregulated (19) and tend to show of asymmetrical inheritance (7, 8), selection for parental similarity coordinated expression, which provides a strong opportunity for (9) or histocompatibility (10), and sexually antagonistic selection selection to favor genetic coadaptation, where combinations of (11). These theories have generated valuable hypotheses about the beneficially interacting alleles (i.e., alleles that “work well” functions of imprinted genes, which can provide important insights into the nature of their phenotypic effects, including the role of

imprinted genes in various pathologies (e.g., refs. 12, 13). Author contributions: J.B.W. designed research, performed research, and wrote Although imprinted genes are members of many different gene the paper. families (14, 15), produce a diversity of proteins and noncoding The author declares no conflict of interest. RNAs (16), and have effects on an assortment of biochemical This article is a PNAS Direct Submission. M.W.F. is a guest editor invited by the Editorial pathways (17), they tend to share two conspicuous patterns. Board. First, imprinted genes tend to have a clustered distribution in 1E-mail: [email protected]. mammalian (15, 18, 19) and plant genomes (20), with the mam- This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. malian clustering appearing to be conserved in vertebrate evolution 1073/pnas.1205686110/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1205686110 PNAS | March 26, 2013 | vol. 110 | no. 13 | 5085–5090 Downloaded by guest on September 24, 2021 together) are coinherited (40). That is, given the temporal and imprinting at the B locus has the value iB(D)y1 and is labeled IB. spatial coupling of imprinted gene expression there is ample op- I use this four-locus system to examine the conditions under portunity for imprinted gene products to interact in their effects which the imprinting alleles at the C and D loci are favored to on fitness, which could lead to genetic coadaptation. For example, understand how epistatic interactions between a pair of loci in- the IGN includes at least 16 imprinted genes that tend to be fluence the evolution of imprinting at those loci. coexpressed and show coordinated changes in response to dif- Selection in this model arises from the assumption that the A ferentiation (41), (e.g., knockout of leads to up- and B loci interact to affect fitness. I use the “classic” model of regulation of several members of the IGN, ref. 42), and in vitro additive-by-additive interactions (45, 46), which is a simple model manipulation (e.g., in vitro fertilization disrupts patterns of meth- of two-locus fitness effects that favors coadaptation, where certain ylation at specific sites within the H19-Igf2 (insulin-like growth allelic combinations at the interacting loci are favored and other factor 2) locus, and these disruptions appear to be compensated for combinations are disfavored. This form of interaction is particu- through the IGN, ref. 43). A subset of 11 of the genes in the IGN larly useful in examining the evolutionary dynamics of imprinting has also been identified as a coexpressed group that shows co- because it does not include any dominance effects. The presence ordinated down-regulation across multiple tissues through time of dominance effects can favor or disfavor the evolution of ge- (34). The fact that imprinted gene clusters are often coregulated nomic imprinting solely because monoallelic expression removes by imprinting control elements (44) also suggests the need for the opportunity for dominance effects to be expressed in heter- coordinated expression. This clustered distribution may also ozygotes (i.e., removes the physiological basis for dominance). reflect an evolutionary history where has As a result, imprinting can be favored solely because it leads built coadapted associations among interactors. to monoallelic expression, not because parent-of-origin-dependent Here I build a simple model in which alleles at a pair of loci expression is favored per se (47, 48). interact to affect fitness and imprinting evolves as a result of Fitness of a four-locus genotype in the absence of imprinting modifier loci. Using a classic model of gene interactions I dem- depends solely on the A-B two-locus genotype, but in the pres- onstrate that the evolution of genomic imprinting at one locus ence of imprinting these values are modified by the degree and can drive the imprinting status at other loci that interact with that direction of silencing of alleles and hence depend on the alleles imprinted gene. The genes that interact with imprinted loci are present at the C and D loci. Fitness as a function of the four- expected to evolve a pattern of imprinting that matches the im- locus genotype is denoted wijkl, where i, j, k, and l indicate the A-, printing pattern of their interacting partner(s). I emphasize that B-, C-, and D-locus genotype, respectively, where at each locus this model is not explicitly concerned with the evolutionary origins genotypes are numbered as X1X1 = 1, X1X2 = 2, X2X1 = 3, X2X2 = 4 of imprinting per se (i.e., why at least one gene becomes im- (where X indicates the locus in question, i.e., A, B, C, or D), with printedinthefirst place), but rather, with how the evolution of the first allele listed coming from the father. In the absence of (or presence of) imprinting at one locus affects the evolution of imprinting fitness is defined by wij••,wherethedot(“•”) subscript the imprinting status of that gene’s interacting partners. indicates the averaging over the genotypes at a locus. Fitness of an A-B two-locus genotype is determined by the average of the Model interaction effects of each of the possible combinations of alleles To examine the general consequences for interaction effects in at the two loci within individuals. For example, the double het- the evolution of genomic imprinting I build on a simple two- erozygote, A1A2B1B2 (w22••)hasafitness value determined by the locus system of gene interactions to include a pair of loci (loci C average of all possible combinations of interactions, A1 with B1, and D) that modify the imprinting status of these interacting loci A1 with B2, A2 with B1, and A2 with B2. When there is no im- (loci A and B). The “A” locus has two alleles, A1 and A2, which printing in the model, the expected fitnesses of the reciprocal have frequencies p1 and p2, respectively, and the “B” locus has heterozygotes are equal, such that w22•• = w23•• = w32•• = w33••, two alleles, B1 and B2, which have frequencies q1 and q2, re- but this symmetry does not hold when there is imprinting pres- spectively. The imprinting status of the A and B loci is modified ent. This fitness model can be conceptualized in various ways, by the C and D loci, respectively. The C and D loci are, there- including as a case where “matching” of allelic effects is favored fore, referred to as the “imprinting modifier loci,” whereas the (e.g., the rates of some processes governed by those loci are A and B loci are referred to as the “imprintable target loci.” coordinated by the matching) or the matching of some general The C locus has two alleles, C1 and C2, that have frequencies x1 allelic property is favored (like binding properties of two inter- and x2, respectively, with the C1 allele leading to imprinted ex- acting proteins, sequences, or RNA–target combinations) pression at the A locus (and hence referred to as an “imprinting (see SI Methods, Fitness Model for further consideration of the allele”), whereas the C2 allele leads to ordinary biallelic expression fitness model). (and is therefore the “nonimprinting allele”). The degree and I assume that the combinations of alleles at the A and B loci direction of imprinting (silencing) of the A locus by the C1 allele that share a subscript (A1 with B1 and A2 with B2) result in higher are measured by the imprinting parameter iA(C), where positive fitness whereas those that have different subscripts (A1 with B2 values indicate paternal expression (i.e., expression of the pa- and A2 with B1) lead to lower fitness. Baseline fitness is given the ternally inherited allele) and negative values indicate maternal ex- value 1. The A1B1 or A2B2 interactions (i.e., positive interaction) pression. The value of iC(A) is, therefore, bounded at +1 and –1, within a genotype have a fitness deviation of +s whereas the A1B2 where a value of jiA(C)j = 1 indicates complete silencing of one or A2B1 interactions (i.e., negative interaction) within a genotype allele. Likewise, the D locus has two alleles, D1 and D2, that have have a fitness deviation of –s. Genomic imprinting of either locus frequencies y1 and y2, respectively, with the D1 allele leading to alters fitness by changing the alleles that are expressed within imprinted expression at the B locus (and hence referred to as an an A-B two-locus genotype and thereby the average fitness of the imprinting allele), whereas the D2 allele leads to ordinary bial- possible allelic interactions. For example, the expected fitness of lelic expression. The pattern of imprinting caused by the D1 allele the double heterozygote A1A2B1B2, in which all four possible is measured by the imprinting parameter iB(D), with the pattern two-locus allelic interactions can occur, has the fitness value s − s − s + s = •• = + of effect on the expression of the B locus following that described w22•• 1 (i.e., w22 1 4 ) in the absence of imprinting, abovefortheinfluence of the C locus on expression at the A locus. but 1 + iA(C)iB(D)s if there is imprinting at the two loci (i.e., the The overall pattern of imprinting at the A locus (i.e., the weighted genotype is A1A2B1B2C1C1D1D1,sofitness is w2211). Therefore, if mean expression level) is denoted IA and is defined by the pat- both loci are paternally expressed [so iA(C) and iB(D) = 1] or both tern of effect of the imprinting allele, iA(C), weighted by its fre- are maternally expressed [so iA(C) and iB(D) = −1], fitness is 1 + s quency, x1; i.e., IA = iA(C)x1. Likewise, the overall mean pattern of because individuals expressed one of the positively interacting

5086 | www.pnas.org/cgi/doi/10.1073/pnas.1205686110 Wolf Downloaded by guest on September 24, 2021 combinations of alleles at the two loci (A1 and B1 or A2 and B2), (IA = −IB and the absolute value of each is 1), LD will have no depending on whether there is paternal or maternal expression, contribution to mean fitness. respectively. In contrast, if the two loci show opposite patterns of The fitness of the 16 haplotypes (Table S5) can be used to imprinting [i.e., either iA(C) = 1 and iB(D) = −1oriA(C) = −1 and calculate how selection changes the frequencies of alleles at the fi iB(D) = 1], then fitness has the value 1 – s because individuals imprinting modi er loci and builds associations between the A express a combination of the negative interaction alleles at the and B loci (see SI Methods, Evolution of Linkage Disequilibrium two loci. The full set of fitness values for the 256 ordered four- for more details and a consideration of other LD parameters). locus genotypes is given in Table S1. The expected fitness of each The evolutionary change in the association between alleles at the of the 16 ordered A-B two-locus genotypes under this fitness A and B loci (measured by dAB) across generations is calculated model is shown in Table S2. as the difference in the amount of LD before selection in gen- The frequencies of the 256 possible four-locus ordered geno- eration t (indicated by the subscript in Eq. 2) compared with the types (i.e., combinations of alleles at the A, B, C, and D loci in amount of LD before selection in generation t + 1, with the two which alleles are ordered by their parent of origin) are determined generations separated by a round of recombination that occurs by the frequencies of the 16 possible haplotypes under the as- after selection in generation t to produce the that form + sumption of random union of gametes (see Table S3 for the generation t 1 (49), definitions of haplotype frequencies). Haplotype and the corre- 2 3 sponding genotype frequencies depend on the frequencies of alleles p1p2q1q2 Δ ≈ 4 − − − − − 5− ; at each locus and the patterns of linkage disequilibria between dAB s dABðtÞ 1 r p1 p2 q1 q2 1 IAIB dABðtÞr + ð − Þð Þ alleles at different loci. Although linkage disequilibrium (LD) can 1 2r p1p2q1q2IAIB occur between alleles at any pair of loci, selection builds signif- [2] icant pairwise association only between alleles at the A and B loci (because they interact to affect fitness) and hence I focus primarily where r is the rate of recombination between the A and B loci. S4 on the LD parameter, dAB.SeeSI Methods, Evolution of Linkage A more exact solution is provided in SI Methods (Eq. ), but the Disequilibrium for a consideration of other LD parameters (see approximation is very close to the exact value under weak se- also Eq. S1 and Table S4 for a list of constraints on the possible lection in a population starting in linkage equilibrium (Fig. S1). The amount of LD generated by selection when there is no values of LD). Positive values of dAB indicate associations between the coadapted alleles (i.e., A with B and A with B ), assuming existing LD and there is free recombination is given by the top 1 1 2 2 2 the selection parameter, s, is positive. However, if s is negative, line in brackets in Eq. (which has the value sp1p2q1q2). The bottom line in brackets accounts for the evolution of LD when then the combinations of alleles with different subscripts (i.e., A1 there is no existing LD but there is restricted recombination be- with B2 and A2 with B1) are associated with higher fitness and, as tween the A and B loci (i.e., r < 1/2). That last term shows that, a result, a negative value of dAB would indicate an association between coadapted alleles. Thus, if s is positive, a positive value with restricted recombination and when both loci show the same pattern of imprinting (where IA and IB are the same sign), the of dAB would reflect coadaptation whereas, if s is negative, then fl evolution of LD will be enhanced, whereas, when the loci show anegativevalueofdAB would re ect coadaptation. EVOLUTION opposite patterns of imprinting (where IA and IB are of opposite All equations presented herein are approximations derived 2 under the assumptions of weak selection in a population that sign), the evolution of LD will be impeded. Eq. shows that the starts in linkage equilibrium, with a high rate of recombination sign of the LD generated by selection matches the sign of s. This between loci. Therefore, although some equations contain a re- occurs because the sign of s simply indicates which combination combination rate parameter, the approximations will be most of alleles is favored by selection. Because I have assumed that the imprinting modifier loci have accurate when there is free recombination. These weak selection additive effects on the degree of imprinting at the target loci, approximations describe the most important determinants of we can examine the conditions under which selection favors the evolutionary dynamics and outcomes, but the exact dynamics can evolution of genomic imprinting by deriving the patterns of se- be complicated by the intricacies of hitchhiking and interference lection on the imprinting alleles as the difference in the expected dynamics of multilocus evolution when selection is strong or re- fitness of the imprinting and nonimprinting alleles (i.e., the mean combination rates are low. See SI Methods for a further discussion fitness of the imprinting allele minus the mean fitness of the non- of the weak selection assumption. imprinting allele). This measures the additive effect of the im- Population mean fitness can be derived as the sum of the fi fi printing alleles on tness. For the C locus, the additive effect of products of the frequencies (Fijkl) and tness values (wijkl) of the the imprinting allele on fitness is simply four-locus genotypes (Eq. S1). Mean fitness can be approxi- mated as aC = sdABiAðCÞIB [3] ≈ + ð − Þð − Þ + ð + Þ [1] w 1 s p1 p2 q1 q2 2dABðtÞ 1 IAIB and for the D locus is S1 1 (see Eq. for an exact expression). It is clear from Eq. that aD = sdABiBðDÞIA: [4] the presence of matching imprinting patterns, where the two loci show the same sign of imprinting (i.e., IA and IB are the same sign, Eqs. 3 and 4 demonstrate that selection can favor imprinting so the two loci express the allele inherited from the same parent), at each locus, but only to the degree to which the other locus is increases the contribution of LD to mean fitness [captured in the already imprinted. For example, the effect of the imprinting al- ð + Þ term 2dAB 1 IAIB ] and can double the contribution of LD to lele at the C locus on fitness (aC,Eq.3) is dependent on the fi = =± mean tness when there is complete imprinting [iA(C) iB(D) 1] degree of imprinting at the B locus, captured in the parameter fi fi and the imprinting modi er loci are both xed for the imprinting IB. Imprinting at either locus is also favored only when there is alleles (x1 = y1 = 1), making IA = IB =±1. Likewise, opposing some level of coadaptation between the interacting A and B loci fi fl imprints (where the two imprinting modi er loci cause con ict- (i.e., dAB ≠ 0), which will generally occur whenever there is se- ing signs of imprinting and both are fixed for the imprinting lection favoring coadaptation (Eq. 2). Eqs. 3 and 4 also dem- allele) can reduce the contribution of LD to mean fitness and, onstrate that selection will favor the imprinting allele only when when there is complete opposing imprinting at the A and B loci the direction of imprinting at the two loci matches [i.e., iA(C) and

Wolf PNAS | March 26, 2013 | vol. 110 | no. 13 | 5087 Downloaded by guest on September 24, 2021 iB(D) are of the same sign]. In contrast, when the pattern of imprinted in mammals appears to predate the evolution of geno- imprinting at the two loci is opposing [i.e., iA(C) and iB(D) are mic imprinting (21). Therefore, it is possible that some genes in of opposite signs], then both additive effects will be negative and the cluster evolved imprinting through a process favoring a specific selection will favor the nonimprinting allele (i.e., selection favors pattern of parent-of-origin-dependent expression and selection the loss of imprinting). for coadaptation and then favored the “spread” of that imprint To understand how imprinting at the A and B loci evolves to other genes in the cluster. This notion mirrors the control of we can examine the evolutionary change in the frequency of the imprinting in clusters, where some genes appear to be the “pri- imprinting alleles at the C and D loci (Δx1 and Δy1, respectively). mary” target of silencing whereas others may be imprinted sec- If selection is weak, these expressions can be approximated as ondarily as the silencing event spreads locally (although these (see Eqs. S10 and S11 for exact expressions and Fig. S2 for secondarily imprinted genes may also be “innocent bystanders,” a comparison between the exact expressions and the approx- ref. 19). imations given below): Completely opposing imprints at the interacting loci, where the population is fixed for imprinting alleles that cause opposite Δ ≈ = = [5] x1 sdABðtÞx1x2y1iAðCÞiBðDÞ sdABðtÞx2IAIB x1x2aC patterns of imprinting on the A and B loci (with one paternally expressed and the other maternally expressed, which occurs when and IA = −IB), can potentially remove the beneficial effect that co- adaptation can have on mean fitness. This result is apparent in Δ ≈ = = : [6] y1 sdABðtÞx1y1y2iAðCÞiBðDÞ sdABðtÞy2IAIB y1y2aD the equation for mean fitness (Eq. 1), where the contribution of LD to mean fitness is 2sd ð Þð1 + I I Þ, which is zero when I = 5 6 AB t A B A Eqs. and demonstrate that the evolutionary change in the −I (and both have an absolute value of 1). Opposing patterns of fi B frequency of the imprinting modi er alleles (and hence the evo- imprinting remove the beneficial effects of LD on mean fitness lution of imprinting at the A and B loci) is simply a function of the fi because the two alleles that are inherited together are never additive effect of the imprinting allele on tness (aC and aD)and expressed together and hence the pattern of association between the amount of genetic variation at the locus (x1x2 or y1y2). alleles that are coinherited (i.e., the pattern of LD) is not “seen” Discussion phenotypically. Indeed, when both imprinting modifier loci cause opposing patterns of imprinting at the interacting loci [so i This simple model of gene interactions demonstrates that selection A(C) and iB(D) are of opposite sign], then selection favors the non- can favor the evolution of genomic imprinting because it coor- imprinting allele at both of the imprinting modifier loci (Eqs. 3 dinates the coexpression of positively interacting coinherited alleles and 4) and we would, therefore, expect the system to evolve to at different loci. Such a scenario, where some combinations of a nonimprinted state (Eqs. 5 and 6). Thus, selection can enhance alleles at different loci increase fitness (i.e., “positively interact”) coadaptation between interacting loci by favoring imprinting alleles and other combinations decrease fitness (i.e., “negatively interact”), when the imprinting modifier loci cause coordinated imprints, but is known to favor the evolution of genetic coadaptation, captured can also enhance coadaptation by favoring the loss of imprinting in the evolution of associations between positively interacting alleles when those alleles cause opposing patterns of imprinting alleles (i.e., LD, ref. 46). The model presented here demonstrates at a pair of interacting loci. that, as the two loci evolve coadaptation, selection favors the loci Perhaps the most important implication of this model is that it to show a matching pattern of imprinting. The outcome of this suggests that some loci may not reflect an evolutionary history of process is that, when the pair of loci interact and show the same fi pattern of imprinting (i.e., both show either maternal or paternal selection having speci cally favored either maternal or paternal expression), then the pair of alleles that are inherited together expression per se, but instead they have coevolved their expression (which are coadapted) are expressed together. However, if nei- pattern to follow the pattern of their interacting partner(s). In ther locus is imprinted, then selection does not favor imprinting such a scenario, some imprinted genes presumably play a direct at either locus. This situation is a sort of evolutionary stalemate, role in some process that generates selection for uniparental ex- where it would be adaptive to evolve coordinated imprinting at pression and, as a result, evolved to become imprinted, but other the two loci, but unless one starts to evolve imprinted expression loci that are not directly involved in that process may evolve to its interacting partner is not favored to be imprinted. This result become imprinted because they interact with an imprinted gene “ ” implies that the most likely scenario is that selection directly favors and hence have simply followed their lead. Given the complexity imprinting at one locus for whatever reason (e.g., as a result of of biological systems there are likely to be a diversity of scenarios conflict or coadaptation), and once one locus evolves imprinting, in which two loci interact but only one of the two is directly se- fl selection then favors its interacting partners to follow its lead and lected to have imprinted expression (which is presumably re ected evolve a matching form of imprinting. in the fact that, for instance, only some of the loci expressed in the The evolution of coadaptation is facilitated by the evolution of placenta are imprinted). For example, under a conflict scenario reduced recombination (presumably achieved by physical linkage) (e.g., ref. 29) one locus could influence some component of pla- between interacting genes (which maximizes the evolution of LD). cental development that influences extraction of maternal resources Reduced recombination is favored because recombination breaks through the placenta, whereas the second locus is involved in some up adaptive combinations of alleles, a phenomenon known as the component of placental development that does not influence ex- recombination load (50). Furthermore, when there is restricted traction of maternal resources. The two loci could interact during recombination, then coordinated genomic imprinting, where loci placental development and jointly influence placental function are expressed from the same parental allele, can facilitate the despite the fact that they do not both modulate parent–offspring evolution of genetic coadaptation (i.e., evolution of LD) (Eq. 2). conflict. Such scenarios could potentially explain why, although The same conditions given for the evolution of LD (Eq. 2) are, many genes appear to have gross phenotypic effects consistent therefore, also the conditions under which selection is expected to with one or more of the “adaptive” theories (51), many imprinted favor mechanisms that reduce the effective rate of recombination genes have phenotypic effects that appear inconsistent with the- and thereby reduce the recombination load. This scenario could oretical predictions (18, 52). Of course we probably do not fully potentially explain some of the clustering of imprinted genes in the understand the full patterns of phenotypic effects of imprinted genome, although it does not necessarily imply that the clustering genes and predictions from theories can be complex (e.g., refs. evolved before or after the evolution of the imprinting status. 25, 26), but it is important to consider the possibility that some Indeed, for at least some clusters, the clustering of genes that are genes have been “converted” to imprinting through the evolution

5088 | www.pnas.org/cgi/doi/10.1073/pnas.1205686110 Wolf Downloaded by guest on September 24, 2021 of coadaptation and, consequently, will not have effects that are important in the development of the conflict component of the not consistent with other theoretical predictions. kinship theory (51). Because this model does not make any assumptions about why Finally, this model makes several broad potentially testable at least one gene evolves to become imprinted at the start of the predictions, with the most notable prediction being that genes coadaptation process, it is of course possible that none of the that interact to affect fitness should show coordinated imprints. genes involved in an interaction currently play a role in any of the We would also expect variation at loci that interact to affect processes laid out by the various existing theories for the origin fitness to be in LD and, therefore, one could potentially examine of imprinting. For example, a gene may have evolved to become whether imprinted loci showing LD tend to show coordinated imprinted because of its role in conflict, but the basis for conflict patterns of imprinting. Because opposing imprints interfere with may no longer be present (e.g., the mating system may have the evolution of coadaptation and negate the potential fitness changed) or the specific gene function may have shifted. However, benefits of coadaptation through the evolution of LD, opposing once both genes in an interacting pair have evolved to be im- imprints between genes that interact to affect fitness would pro- printed, and the presence of imprinting enhances the coadaptation vide data that are in conflict with the pattern predicted by this between those genes, then selection can favor the maintenance model. This model also predicts the possibility that some of the of imprinted expression because of its contribution to coadapted imprinted genes that have effects that are inconsistent with adap- gene expression. It is, of course, also possible that selection never tive models for the evolutionary origin of imprinting may have favored parent-of-origin-dependent expression per se, but there evolved imprinting to match the pattern of their imprinted inter- “ ” may have been some sort of epigenetic drift where one locus acting partners. In such cases we would expect that the evolution randomly evolved some degree of imprinting and, as a result, se- of the imprinted status at the “leaders” in the process (which lection favored a matching pattern of imprinting at its inter- presumably evolved imprinting under one of the various adaptive acting partner, resulting in a reinforcing process where the pair theories) should predate and potentially be more taxonomically of loci evolved to a jointly imprinted state. As a result, none of widespread than is imprinting at the “followers” (because estab- the loci would necessarily have effects consistent with the pre- lishment of imprinting at the leader locus would have provoked dictions of the various models, but the coadaptation process the evolution of imprinting at the follower locus). The IGN, would maintain the imprints because they coordinate coadapted which includes many interacting imprinted genes (33), may be gene expression. a particularly valuable system to test these predictions, although Because most models for the evolutionary origins of genomic relevant data are lacking. For example, the fitness consequences imprinting represent broad scenarios and therefore rarely include specific mechanistic details, it is difficult to predict or generalize and patterns of natural allelic variation at the genes involved in about which processes are the most likely to produce the sorts of the IGN are unknown as is whether there is any relevant LD. evolutionary scenarios envisioned by this model. However, data Aside from the IGN there are presumably classes of genes, such on the IGN have demonstrated that imprinted genes interact to as those that show patterns of stringent allelic interactions (e.g., influence prenatal growth and development in mice (33), poten- oligomers or trans-acting antisense sequences), that are likely tially through interactions in the placenta (43), suggesting that to evolve coadaptation and could potentially provide important fl test cases.

theories such as con ict (29) and maternal-offspring coadaptation EVOLUTION models (6) (which are perhaps most relevant to prenatal growth ACKNOWLEDGMENTS. I thank Yaniv Brandvain, Laurence Hurst, and Hamish and development) may be particularly likely to initiate the evo- Spencer for insightful discussions that contributed to the development of lutionary process described by this model. Indeed, the IGN includes these ideas. I also thank an anonymous reviewer who provided critical analysis the imprinted H19 and Igf2 genes that have been especially of the model and important input that significantly improved this work.

1. Reik W, Walter J (2001) Genomic imprinting: Parental influence on the genome. Nat 17. Tycko B, Morison IM (2002) Physiological functions of imprinted genes. J Cell Physiol Rev Genet 2(1):21–32. 192(3):245–258. 2. Brandvain Y, Van Cleve J, Ubeda F, Wilkins JF (2011) Demography, kinship, and the 18. Morison IM, Ramsay JP, Spencer HG (2005) A census of mammalian imprinting. Trends evolving theory of genomic imprinting. Trends Genet 27(7):251–257. Genet 21(8):457–465. 3. Haig D (2002) Genomic Imprinting and Kinship (Rutgers Univ Press, New Brunswick, NJ). 19. Verona RI, Mann MRW, Bartolomei MS (2003) Genomic imprinting: intricacies of 4. Haig D, Moore T (1991) Genomic imprinting: A parental tug-of-war. Trends Genet 7: epigenetic regulation in clusters. Annu Rev Cell Dev Biol 19:237–259. 45–49. 20. Zhang M, et al. (2011) Extensive, clustered parental imprinting of protein-coding 5. Haig D (1997) Parental antagonism, relatedness asymmetries, and genomic imprint- and noncoding RNAs in developing maize . Proc Natl Acad Sci USA 108(50): – ing. Proc Biol Sci 264(1388):1657–1662. 20042 20047. 6. Wolf JB, Hager R (2006) A maternal-offspring coadaptation theory for the evolution 21. Dünzinger U, Nanda I, Schmid M, Haaf T, Zechner U (2005) Chicken orthologues of genomic imprinting. PLoS Biol 4(12):e380. of mammalian imprinted genes are clustered on macrochromosomes and replicate – 7. Spencer HG, Feldman MW, Clark AG, Weisstein AE (2004) The effect of genetic con- asynchronously. Trends Genet 21(9):488 492. flict on genomic imprinting and modification of expression at a -linked locus. 22. Costa LM, et al. (2012) Maternal control of nutrient allocation in plant seeds by genomic imprinting. Curr Biol 22:1–6. Genetics 166(1):565–579. 23. Swaney WT, Curley JP, Champagne FA, Keverne EB (2008) The paternally expressed 8. Wolf JB (2009) Cytonuclear interactions can favor the evolution of genomic imprinting. gene Peg3 regulates sexual experience-dependent preferences for estrous odors. Evolution 63(5):1364–1371. Behav Neurosci 122(5):963–973. 9. Spencer HG, Clark AG (2006) A chip off the old block: A model for the evolution of 24. Curley JP (2011) Is there a genomically imprinted social brain? Bioessays 33(9): genomic imprinting via selection for parental similarity. Genetics 174(2):931–935. 662–668. 10. Wolf JB, Hager R (2009) Selective abortion and the evolution of genomic imprinting. 25. Ubeda F, Gardner A (2010) A model for genomic imprinting in the social brain: J Evol Biol 22(12):2519–2523. Juveniles. Evolution 64(9):2587–2600, 10.1111/j.1558-5646.2010.01115.x. 11. Day T, Bonduriansky R (2004) Intralocus sexual conflict can drive the evolution of 26. Garfield AS, et al. (2011) Distinct physiological and behavioural functions for parental – genomic imprinting. Genetics 167(4):1537 1546. alleles of imprinted Grb10. Nature 469(7331):534–538. 12. Ubeda F (2008) Evolution of genomic imprinting with biparental care: Implications 27. Champagne FA, Curley JP, Swaney WT, Hasen NS, Keverne EB (2009) Paternal influence for Prader-Willi and Angelman syndromes. PLoS Biol 6(8):e208. on female behavior: The role of Peg3 in exploration, olfaction, and neuroendocrine 13. Badcock C, Crespi B (2006) Imbalanced genomic imprinting in brain development: regulation of maternal behavior of female mice. Behav Neurosci 123(3):469–480. – An evolutionary basis for the aetiology of . J Evol Biol 19(4):1007 1032. 28. Haig D (1996) Placental hormones, genomic imprinting and maternal-fetal commu- 14. Morison IM, Paton CJ, Cleverley SD (2001) The imprinted gene and parent-of-origin nication. J Evol Biol 9(3):357–380. effect database. Nucleic Acids Res 29(1):275–276. 29. Haig D (1993) Genetic conflicts in human pregnancy. Q Rev Biol 68(4):495–532. 15. Williamson CM et al. (2012) Mouse Imprinting Data and References (MRC Harwell, 30. Renfree M, et al. (2011) Genome sequence of an Australian kangaroo, Macropus eugenii, Oxfordshire). Available at http://www.har.mrc.ac.uk/research/genomic_imprinting/. provides insight into the evolution of mammalian reproduction and development. Accessed February 28, 2013. Genome Biol 12(8):R81. 16. Royo H, Cavaillé J (2008) Non-coding RNAs in imprinted gene clusters. Biol Cell 100(3): 31. Hurst LD (1997) Evolutionary theories of genomic imprinting. Genomic Imprinting, 149–166. eds Reik W, Surani A (Oxford Univ Press, Oxford).

Wolf PNAS | March 26, 2013 | vol. 110 | no. 13 | 5089 Downloaded by guest on September 24, 2021 32. Lewis A, Redrup L (2005) Genetic imprinting: Conflict at the Callipyge locus. Curr Biol 42. Gabory A, et al. (2009) H19 acts as a trans regulator of the imprinted gene network 15(8):R291–R294. controlling growth in mice. Development 136(20):3413–3421. 33. Varrault A, et al. (2006) Zac1 regulates an imprinted gene network critically involved 43. Fauque P, et al. (2010) Modulation of imprinted gene network in placenta results in normal in the control of embryonic growth. Dev Cell 11(5):711–722. development of in vitro manipulated mouse . Hum Mol Genet 19(9):1779–1790. 34. Lui JC, Finkielstain GP, Barnes KM, Baron J (2008) An imprinted gene network that 44. Spahn L, Barlow DP (2003) An ICE pattern crystallizes. Nat Genet 35(1):11–12. controls mammalian somatic growth is down-regulated during postnatal growth 45. Cheverud JM, Routman EJ (1996) Epistasis as a source of increased additive genetic deceleration in multiple organs. Am J Physiol Regul Integr Comp Physiol 295(1): variance at population bottlenecks. Evolution 50(3):1042–1051. – R189 R196. 46. Rice SH (2004) Evolutionary Theory: Mathematical and Conceptual Foundations 35. Sandhu KS (2010) Systems properties of proteins encoded by imprinted genes. Epi- (Sinauer, Sunderland, MA). genetics 5(7):627–636. 47. Brandvain Y (2010) Matrisibs, patrisibs, and the evolution of imprinting on 36. Zhao Z, et al. (2006) Circular chromosome conformation capture (4C) uncovers extensive and sex chromosomes. Am Nat 176(4):511–521. networks of epigenetically regulated intra- and interchromosomal interactions. Nat 48. Van Cleve J, Feldman MW (2007) Sex-specific viability, sex linkage and dominance in Genet 38(11):1341–1347. genomic imprinting. Genetics 176(2):1101–1118. 37. Davis E, et al. (2005) RNAi-mediated allelic trans-interaction at the imprinted Rtl1/ 49. Bürger R (2000) The Mathematical Theory of Selection, Recombination, and Mutation Peg11 locus. Curr Biol 15(8):743–749. 38. Cattanach BM, Beechey CV, Peters J (2004) Interactions between imprinting effects in (Wiley, Chichester, UK). the mouse. Genetics 168(1):397–413. 50. Barton NH, Charlesworth B (1998) Why sex and recombination? Science 281(5385): 39. Cattanach BM, Beechey CV, Peters J (2006) Interactions between imprinting effects: 1986–1990. Summary and review. Cytogenet Genome Res 113(1–4):17–23. 51. Haig D (2004) Genomic imprinting and kinship: How good is the evidence? Annu Rev 40. Hartl DL, Clark AG (1997) Principles of Population Genetics (Sinauer, Sunderland, MA). Genet 38:553–585. 41. Berg JS, et al. (2011) Imprinted genes that regulate early mammalian growth are 52. Hurst LD, McVean GT (1997) Growth effects of uniparental disomies and the conflict coexpressed in somatic stem cells. PLoS ONE 6(10):e26410. theory of genomic imprinting. Trends Genet 13(11):436–443.

5090 | www.pnas.org/cgi/doi/10.1073/pnas.1205686110 Wolf Downloaded by guest on September 24, 2021