A Supergene-Linked Estrogen Receptor Drives Alternative Phenotypes in a Polymorphic Songbird

A supergene-linked estrogen receptor drives alternative phenotypes in a polymorphic songbird

Jennifer R. Merritta,1, Kathleen E. Grogana, Wendy M. Zinzow-Kramera, Dan Sunb, Eric A. Ortlundc, Soojin V. Yib, and Donna L. Maneya

aDepartment of Psychology, Emory University, Atlanta, GA 30322; bSchool of Biological Sciences, Georgia Institute of Technology, Atlanta, GA 30332; and cDepartment of Biochemistry, Emory University, Atlanta, GA 30322

Edited by Gene E. Robinson, University of Illinois at Urbana–Champaign, Urbana, IL, and approved July 8, 2020 (received for review June 3, 2020) Behavioral evolution relies on genetic changes, yet few behaviors tan-striped (TS) morph are homozygous for the standard ar- can be traced to specific genetic sequences in vertebrates. Here we rangement, ZAL2 (13, 14) (Fig. 1A). The rearrangement is provide experimental evidence showing that differentiation of a maintained in the population because of the species’ unique dis- single gene has contributed to the evolution of divergent behav- assortative mating system; nearly every breeding pair consists of ioral phenotypes in the white-throated sparrow, a common back- one individual of each morph (15). Because almost all WS birds yard songbird. In this species, a series of chromosomal inversions are heterozygous for ZAL2m (15, 16), this mating system keeps has formed a supergene that segregates with an aggressive phe- ZAL2m in a near-constant state of heterozygosity (Fig. 1B), pro- notype. The supergene has captured ESR1, the gene that encodes foundly suppressing recombination and causing it to differentiate estrogen receptor α (ERα); as a result, this gene is accumulating from ZAL2 (15, 17). changes that now distinguish the supergene allele from the stan- The rearranged region of ZAL2m in white-throated sparrows dard allele. Our results show that in birds of the more aggressive can be considered a supergene because it harbors a discrete set phenotype, ERα knockdown caused a phenotypic change to that of coinherited, coevolving genes that influence a suite of traits. of the less aggressive phenotype. We next showed that in a free- This supergene dictates not only plumage coloration, but also living population, aggression is predicted by allelic imbalance fa- levels of territorial aggression. In both field and laboratory cis voring the supergene allele. Finally, we identified -regulatory studies, WS birds have been shown to be more aggressive than features, both genetic and epigenetic, that explain the allelic im-

TS birds. In free-living, breeding populations, for example, WS NEUROSCIENCE balance. This work provides a rare illustration of how genotypic birds of both sexes respond to simulated territorial intrusion divergence has led to behavioral phenotypic divergence in a (STI) with higher levels of aggression than their TS counterparts vertebrate. (18, 19). ZAL2m/ZAL2m homozygotes are quite rare (15, 20); those that have been described were extraordinarily aggressive chromosomal inversion | songbird | social behavior (16, 18). Thus, the ZAL2m rearrangement is associated with aggression in a dose-dependent manner. here is no doubt that many social behaviors have a genetic Territorial aggression in songbirds has been strongly linked to Tbasis. They are heritable, are acted on by natural selection, circulating levels of gonadal steroid hormones (21, 22). In white- and they evolve (1). Nevertheless, few genetic sequences have throated sparrows, WS birds have higher plasma levels of tes- been linked directly to social behaviors in vertebrates. Most tosterone and estradiol (E2) than TS birds (19); however, this behavioral phenotypes are polygenic, and social behavior itself is difference does not explain the morph difference in aggression. flexibly expressed depending on context. This complexity, together with the many levels of biological organization separating Significance a gene sequence from a social behavior, make it difficult to completely understand why and how natural genotypic variation contributes to behavioral phenotypes (2–5). The white-throated sparrow, a common North American song- The most promising animal models for identifying genetic bird, is a valuable model in behavioral genetics because of a targets of behavioral evolution are those with well-documented chromosomal rearrangement that segregates with a behavioral genetic variation linked to clear behavioral phenotypes. These phenotype. Birds with the rearrangement are more aggressive organisms include Microtus voles (6) and Peromyscus mice (7), in than those without it. Here we show that genetic differentiation which genetic variation in the vasopressin system has been of a single gene inside the rearrangement changes how that causally linked with variation in affiliative and parental behavior, gene is regulated, driving higher levels of expression in birds with the aggressive phenotype and altering behavior. By ex- respectively. Other promising models include organisms in which perimentally reducing the expression of this one gene, we were a behavioral phenotype is linked to large-scale changes in ge- able to change the phenotype of the more aggressive birds to nomic architecture. In fire ants (Solenopsis invicta), for example, the less aggressive phenotype. These results contribute to a the social structure of colonies segregates with a chromosomal greater understanding of how social behaviors can be encoded inversion inside which recombination is suppressed, leading in the genome and how they evolve. to the formation of a “supergene,” a group of genes that are “ ” locked together and inherited as a unit (8). Similarly, in ruffs Author contributions: J.R.M., K.E.G., W.M.Z.-K., E.A.O., S.V.Y., and D.L.M. designed re- (Philomachus pugnax), a chromosomal inversion appears to search; J.R.M., K.E.G., W.M.Z.-K., D.S., E.A.O., S.V.Y., and D.L.M. performed research; mediate a number of alternative reproductive strategies (9, 10), J.R.M., K.E.G., W.M.Z.-K., D.S., E.A.O., S.V.Y., and D.L.M. analyzed data; and J.R.M. and although the causal genes have not been identified. D.L.M. wrote the paper. Decades before the discovery of the chromosomal inversions The authors declare no competing interest. in fire ants and ruffs, Thorneycroft (11, 12) described a rear- This article is a PNAS Direct Submission. rangement of the second chromosome in white-throated sparrows Published under the PNAS license. (Zonotrichia albicollis), a common North American songbird. This 1To whom correspondence may be addressed. Email: [email protected]. m rearrangement, which has been called ZAL2 (“m” for meta- This article contains supporting information online at https://www.pnas.org/lookup/suppl/ centric) segregates with a plumage morph. White-striped (WS) doi:10.1073/pnas.2011347117/-/DCSupplemental. birds of both sexes have a copy of ZAL2m, whereas birds of the

www.pnas.org/cgi/doi/10.1073/pnas.2011347117 PNAS Latest Articles | 1of8 Downloaded by guest on September 29, 2021 A B WS birds, but not TS birds, are sensitive to the effects of E2 on aggression (23). To test whether this morph-specific sensitivity can be explained by the morph difference in ESR1 expression in TnA, we knocked down ESR1 expression in TnA using antisense oligonucleotides (Fig. 2A). Our rationale for this approach was that if we knocked down ESR1 expression in TnA to a TS-like level in WS birds, then the behavioral response to E2 in those WS birds would be TS-like (low response), and the morph difference in aggression would be abolished. This prediction was supported by our findings. In animals treated with scrambled (control) oligonucleotides, an oral bolus dose of E2 enhanced aggression toward a same-morph conspecific (Fig. 2 B, C, and F). This effect was observed in WS birds Fig. 1. Polymorphism in white-throated sparrows. (A) White-throated but not in TS birds (morph × treatment interaction, P < 0.01; SI sparrows occur in two morphs: a more aggressive white-striped (WS) Appendix, Table S2), consistent with previous findings (23, 24). morph and a less aggressive tan-striped (TS) morph. WS birds are hetero- m In other words, the morph with naturally higher expression of zygous for a rearrangement of chromosome 2, known as ZAL2 , which ESR1 in TnA responded to exogenous E2, but the morph with functions as a supergene. Note that we follow conventional nomenclature for avian chromosomes, numbering them from largest to smallest (13). low expression did not. When the morph difference in ESR1 Chromosome 2 in white-throated sparrows corresponds to chromosome 3 in expression in TnA was then abolished via administration of an- chickens (14). (B) Nearly all breeding pairs consist of one TS (ZAL2/ZAL2) bird tisense oligonucleotides (Fig. 2A), the E2-induced aggression in and one WS (ZAL2/ZAL2m) bird. As a result, approximately 50% of the off- WS birds was indistinguishable from that in TS birds (Fig. 2 C spring are ZAL2/ZAL2m heterozygotes and thus WS, and the rest are ZAL2/ and F). Thus, WS-typical levels of ESR1 expression in TnA were ZAL2 homozygotes and thus TS. (Photo courtesy of Jennifer Merritt.) necessary for E2 to facilitate aggression (SI Appendix, Tables S1–S3). To provide further support for the explanatory power of ESR1, Even when plasma testosterone or E2 is experimentally equal- we next tested for a correlation between the aggressive behavior ized, WS birds are still more aggressive (23, 24). This finding in the behavioral trials described above (Fig. 2B) and the level of suggests that WS birds are more sensitive than TS birds to the ESR1 expression in TnA. At 24 h after the second behavioral behavioral effects of these hormones, perhaps because of dif- trial, brains were collected and mRNA was extracted from TnA ferential expression of a steroid hormone receptor. One of the in each bird. Even when our analysis was limited to the control genes inside the supergene is ESR1 (14), which encodes estrogen animals and “misses”—in other words, animals not receiving receptor α (ERα). In rodents, manipulation of ESR1 expression knockdown in TnA (total n = 20)—ESR1 expression in TnA increases aggressive behavior and inhibits prosocial behavior predicted aggression independently of morph (Fig. 2 D and G). (25–27). In songbirds, including white-throated sparrows, ESR1 ESR2, a paralog of ESR1 that is not differentially expressed by expression in some brain regions is predictive of territorial ag- morph and is not on chromosome 2 (34), did not predict aggression (28, 29). Thus, it is particularly compelling that this gene gression (Fig. 2 E and H)orESR1 expression (SI Appendix, Fig. is locked inside the ZAL2/2m rearrangement. S1). These results from laboratory-housed birds replicate our ESR1 resides entirely within the rearrangement (14) and has findings in free-living populations (28) showing that ESR1 ex- accumulated nucleotide divergence between the ZAL2 and pression, in TnA specifically, predicts aggressive behavior. ZAL2m alleles. The resulting fixed nonsynonymous mutations are not likely to affect receptor function, however. Instead, vari- Allelic Imbalance in ESR1 Expression. We showed above that the ation in regulatory regions has been introduced that could alter aggressive phenotype of WS birds is mediated by their increased levels of expression (28). Several observations suggest that the ESR1 expression in TnA, compared with TS birds. We next hy- genetic differentiation between the ZAL2 and ZAL2m alleles of pothesized that this differential expression, which leads to dif- ESR1 could explain the behavioral polymorphism in this species. ferential behavior, is mediated by divergence of cis-regulatory First, ESR1 is differentially expressed between the morphs in regions of the ESR1 gene. To test for differential regulation of several brain regions associated with social behavior (28). Second, the ESR1 alleles, we quantified the allelic imbalance—the de- expression in some of these regions predicts aggressive behavior gree to which one allele is expressed more than the other—in better than does morph (28). Third, exogenous E2 facilitates ag- TnA (SI Appendix, Fig. S2). We conducted this study using tissue gression in WS birds, but not in TS birds (23, 24). Thus, we hy- from free-living WS birds (ZAL2m/ZAL2 heterozygotes) in pothesized that cis-regulatory variation in ESR1 leads to breeding condition, collected from our field site near Argyle, differential expression between the morphs, and that differential Maine. We quantified aggressive responses to an STI, during expression of ESR1 causes the aggressive phenotype. Support for which a caged male decoy was placed onto a resident pair’s these hypotheses would allow us to causally connect genotype to territory, conspecific male song was played through a speaker, phenotype. and the aggressive responses of the residents were observed for 10 min (19). Because ESR1 expression in TnA is strongly cor- Results related with the amount of singing in response to STI (28), we ESR1 Mediates an Aggressive Phenotype. ESR1 expression is several focused on that behavior in particular. Tissue was collected 24 h fold higher in WS birds than in TS birds in nucleus taeniae of the later to minimize the effect of the STI itself on gene expression, amygdala (TnA) (28, 30), also called the ventrolateral arcopal- and allelic imbalance was measured via qPCR. In the same birds, lium (31), which shares molecular markers, connectivity, and we looked for allelic imbalance in two other regions of the brain: function with the medial amygdala of mammals (31–33). In this the rostral portion of the medial preoptic area (POM) and the region, expression of ESR1 predicts territorial aggression (28); hypothalamus (HYP). Levels of ESR1 expression differ between therefore, we hypothesized that this morph difference in ESR1 the morphs in these regions, albeit to a much lesser extent than expression contributes to the behavioral phenotype. To test this in TnA (28, 30). hypothesis, we built on our previous finding that a bolus dose of We detected allelic imbalance in all three brain regions. The E2 enhances aggression compared with a control treatment in ZAL2 allele was overexpressed, relative to the ZAL2m allele, in WS birds but not in TS birds (24). In other words, we knew that HYP and POM (Fig. 3 A and D and SI Appendix,TablesS4andS5);

2of8 | www.pnas.org/cgi/doi/10.1073/pnas.2011347117 Merritt et al. Downloaded by guest on September 29, 2021 A B

CDE

FGH NEUROSCIENCE

Fig. 2. ESR1 expression mediates the morph difference in aggression. (A) ESR1 expression was reduced by ESR1 knockdown (ESR1-KD) in TnA. Data are shown for pooled left and right TnA for each animal. Horizontal bars represent means (n = 6 or 7 per group). (B) During behavioral testing, the cages of the focal bird and a subordinate were separated by a visual barrier (dark blue; cage and bird not drawn to scale). Ten minutes after oral administration of 17β- estradiol (E2) or a control treatment (CON), the visual barrier was removed for 10 min. Reprinted from ref. 24. Copyright (2018), with permission from Elsevier. (C–H) The y-axes depict the changes in behavior between the CON and E2 trials. ESR1 knockdown significantly reduced the degree to which E2 increased the number of attacks (C) and the time spent near the opponent (in the light blue area in B) in WS birds only (F). Both behaviors were correlated with the level of ESR1 expression (D and G) but not with the level of ESR2 expression (E and H)(n = 20). Residuals from a partial correlation controlling for morph are plotted in D, E, G, and H. Only birds receiving infusions of scrambled oligonucleotides (n = 13) and birds in which the cannulae missed TnA (n = 7) were included in analyses of ESR1 and ESR2 expression. All birds were laboratory-housed. More information is provided in SI Appendix, Tables S1–S3.*P < 0.05; horizontal bars represent means.

in contrast, the ZAL2m allele was more highly expressed than Cis-Regulation of ESR1 Expression. We showed above that ESR1 ZAL2 in TnA (Fig. 3G). This imbalance predicted the territorial expression in TnA is causal for an aggressive phenotype in WS response to STI (Fig. 3H). In contrast, allelic imbalance in HYP birds, and that this aggressive phenotype is predicted by ex- m and POM did not predict that response (Fig. 3 B and E). Thus, pression of the ZAL2 allele specifically. We next explored the the degree to which a bird engaged in territorial aggression, genetic and epigenetic mechanisms underlying allelic imbalance. which was markedly higher in the WS birds than in the TS birds First, we hypothesized that cis-regulatory divergence has led to (19, 24, 28), was predicted by the relative expression of the differential transcriptional activity of the two alleles, potentially ZAL2m allele. because of divergence of transcription factor binding sites. Sec- Engaging in territorial aggression can affect plasma levels of ond, we hypothesized that epigenetic regulation of the ESR1 steroid hormones (22) and presumably the expression of steroid- promoter regions differs between the alleles, resulting in differ- ESR1 related genes. Thus, it is possible that the correlation between ential expression. In this species, contains three transcription start sites, which are used in the brain in both morphs aggressive behavior and ESR1 allelic imbalance is caused by an (Fig. 4A) (28). The 2-kb regions immediately upstream of each effect of the behavior on expression. Therefore, we tested start site, which comprise the cis-regulatory elements (CREs) A, whether the pattern of allelic imbalance in adulthood is already B, and C, together contain 42 fixed differences between ZAL2 present early in development, before birds are engaging in ter- and ZAL2m (0.7% divergence) (35) (Fig. 4A and SI Appendix, ritorial aggression. Allelic imbalance was detected in nestlings in Table S6). To test whether these genetic differences are suffi- < all three regions (P 0.02 for all) (Fig. 3 C, F, and I and SI cient to cause allelic differences in transcription activity, we Appendix, Tables S4 and S5), and as was the case in adults, performed luciferase reporter assays in avian cells in culture. All m ZAL2 allele expression exceeded ZAL2 expression only in TnA three CREs were cloned upstream of the luciferase gene in a (Fig. 3I). Overall expression of ESR1 in TnA was higher in WS vector containing an SV40 enhancer such that expression of the nestlings than in TS nestlings at the same age (30), demon- luciferase gene was under transcriptional control of the CREs strating that the pattern of allelic imbalance and the morph (SI Appendix, Fig. S3A). These constructs were transfected into difference in ESR1 expression emerge early in development and DT40 cells, an avian cell line amenable to transfection (36, 37) thus are unlikely to be caused by engaging in territorial behavior. (SI Appendix, Table S7 and Fig. 4B). For all three CREs, the

Merritt et al. PNAS Latest Articles | 3of8 Downloaded by guest on September 29, 2021 differential methylation of the two alleles. To test this hypothesis, we sequenced bisulfite-converted DNA extracted from TnA ACB of free-living birds in breeding condition. This analysis revealed that the ZAL2 allele is in fact significantly more methylated than the ZAL2m allele (Fig. 4D and SI Appendix, Table S9), suggesting the morph difference in expression could be due in part to epigenetic control. The differential methylation could not be detected when we considered only shared sites, that is, CpG sites present on both alleles (Fig. 4D). Furthermore, methylation of DFE the ZAL2 allele was essentially equivalent in the WS and TS birds (Fig. 4D). Therefore, differential methylation contributing to morph differences in expression likely occurs predominantly at genetically differentiated CpG sites, not at shared sites. We next tested whether methylation of the CREs predicts expression for each allele. As a means of data reduction, we constructed correlation matrices of CpG methylation to identify GIH clusters of similarly methylated neighboring sites within each allele (Fig. 4E and SI Appendix, Fig. S5) following established protocols (39, 40). These clusters varied according to allele (Fig. 4E), suggesting that genetic differentiation has likely both contributed to and disrupted interactions between neighboring sites. Remarkably, of the three clusters that significantly predicted allele-specific expression, all contained at least one unshared CpG site and all were on ZAL2 rather than on ZAL2m. Thus, the predictive value of methylated CpG sites was markedly reduced for ZAL2m, which is missing the ZAL2-specific sites. Our findings show that this system represents a rare and inter- Fig. 3. Allelic imbalance in ESR1 expression. We quantified allelic imbalance esting example of cis-regulation that is attributable to a combi- in three brain regions in heterozygous (WS) adults (A, D, and G) and nest- nation of genetic and epigenetic forces (41). The regulatory lings (C, F, and I) sampled from a free-living population (adults, n = 15 to 18; variation in ESR1 predicts and likely causes differential expres- nestlings, n = 26 to 27). In the bar graphs, each column represents the rel- sion of this gene, which, as we showed above, can drive divergent ative expression of ZAL2 (blue) and ZAL2m (red) in a single bird. The dashed behavioral phenotypes. line represents a null ratio of 0.5. Behavioral responses of adult males to STI (Methods) were not predicted by the degree of allelic imbalance in HYP (B) Discussion or POM (E); however, they were significantly correlated with allelic imbalance in TnA (H)(n= 10) (SI Appendix, Tables S4 and S5). In this series of studies, we have demonstrated how genetic divergence in a single gene has contributed to the evolution of an aggressive phenotype in a naturally occurring, wild species of activity of the ZAL2m allele was higher than that of the ZAL2 vertebrate. The gene ESR1 has been captured by a chromosomal allele (Fig. 4B and SI Appendix, Table S7 and Fig. S3B). This rearrangement that comprises a supergene, a group of linked result suggests that the genetic differentiation between the alleles genes that are coinherited. Supergenes caused by inversions are is sufficient to cause differential expression even in the absence of associated with social behavior not only in white-throated spar- other factors (e.g., trans-regulatory elements or chromatin status) rows, but also in ruffs, Alpine silver ants (Formica selysi), and fire that might contribute to morph differences in expression. ants (8, 9, 42). The genes captured inside these inversions are in To explore the impact of allelic differentiation on transcrip- tight linkage disequilibrium, making it difficult to identify causal tion factor binding, we identified transcription factor binding alleles or to uncover the genetic and epigenetic mechanisms that sites that are allele-specific, meaning that a single-nucleotide affect their expression (43, 44). Here we show that identifying polymorphism (SNP) has either abolished the site or reduced its such alleles is possible when genomic resources are available, the affinity for a particular transcription factor on one allele compared model is amenable to mechanistic experimental approaches, and with the other. To identify such sites, we used sTRAP, a tool that there is already some knowledge of the physiological mecha- predicts the impact of SNPs on the affinity of transcription factors nisms underlying the behavior—in this case, steroid hormones. (38). This approach yielded nearly 300 transcription factors with We previously showed that in white-throated sparrows, ESR1 predicted differential affinity for binding sites within the ZAL2 vs. is expressed not only in TnA, but also in other regions of the ZAL2m alleles of ESR1 (P < 0.05, Benjamini–Hochberg correc- brain thought to regulate social behavior (28). Although ESR1 tion for multiple comparisons). Using our published RNA-seq expression depends on morph in most of these regions, the di- data (34, 35), we found that 120 of those transcription factors rection of that effect varies. In other words, in some regions, are expressed in TnA in both morphs (Fig. 4A and SI Appendix, expression is higher in WS birds, whereas in other regions, it is Fig. S4), suggesting a clear mechanism by which the level of ESR1 higher in TS birds. For example, in contrast to its expression in expression could be influenced by genotype. These transcription TnA, ESR1 expression in POM is higher in TS males than in WS factors were neither overrepresented on ZAL2m nor enriched for males in both nestlings and adults (28, 30). In parental adult differential expression by morph (SI Appendix,Fig.S4). males, this expression in POM predicts parental provisioning Differential gene expression and allelic imbalance do not al- more accurately than does morph, suggesting a potentially causal ways involve genetic differentiation. These phenomena can also role for ESR1 expression in parenting (28). This relationship be caused by epigenetic factors, such as DNA methylation. Of 49 between ESR1 and parental behavior was detected only in POM CpG sites in the regulatory sequences of ESR1 (Fig. 4A), 22% and not in any other region in which ESR1 expression was are “unshared,” meaning that because of sequence divergence, measured. Clearly, the potential influence of ESR1 on any par- the site is present on only one of the two alleles. The number of ticular behavior depends not only on the level at which the gene unshared CpGs is higher on ZAL2 (Fig. 4C and SI Appendix, is expressed, but also on where it is expressed; in other words, Table S8), suggesting that genetic divergence could contribute to expression must be influenced by local, region-specific conditions.

4of8 | www.pnas.org/cgi/doi/10.1073/pnas.2011347117 Merritt et al. Downloaded by guest on September 29, 2021 AB

DE NEUROSCIENCE

Fig. 4. Mechanisms underlying allelic imbalance in ESR1.(A) ESR1 is alternatively spliced. Dark blue or red regions are CREs; transcribed regions are in light colors. Black lines within CREs represent 42 fixed differences distinguishing ZAL2 from ZAL2m. Lollipops represent CpG sites. Stacked circles represent transcription factors that are expressed in TnA and for which a binding site is disrupted by a fixed difference. (B) Cis-regulatory variation in ESR1 contributes to variation in activity of the CREs in avian DT40 cells in vitro. Activity, in relative light units (RLU), of the ZAL2m-luciferase (LUC; red) and ZAL2-LUC (blue) constructs are shown normalized to activity of the ZAL2-LUC construct (± SEM). n = 6, *P < 0.05 (SI Appendix, Table S8). (C) The Venn diagram shows the number of shared and unshared CpG sites within the CREs. (D) In TnA, methylation of these CREs depended on the allele. In WS birds, the ZAL2m sequence was less methylated than ZAL2, but this difference was not detectable when considering only shared CpG sites (SI Appendix, Table S9). Boxplots indicate median, interquartile range, and the 10th to 90th percentiles (TS, n = 14; WS, n = 18). (E) Correlation matrices, using data from WS birds, demonstrate covariation in methylation of CpG sites along the sequences. Similarly methylated clusters that significantly predicted allele-specific expression are outlined in yellow; those that did not are outlined in black (SI Appendix, Fig. S5). Unshared sites are marked by boxes to the left of each matrix. *P < 0.05.

In the present study, we have shown that these local factors dictate elements of many other genes. Thus, the ZAL2m allele of ESR1 allelic expression. In both adults and nestlings, the ZAL2m allele is likely coevolving with other genes inside the supergene. One of ESR1 is expressed at higher levels in TnA, whereas expression potential coevolving gene is GRM1, encoding the metabotropic of the ZAL2 allele is greater in HYP and POM (Fig. 3). These glutamate receptor mGluR, which in rats interacts with ERα findings suggest that cis-regulatory variation interacts with region- situated in cell membranes (49). This possibility is particularly dependent trans-regulatory and epigenetic factors to create com- intriguing because the time course of the behavioral effects of E2 plex patterns of expression that lead to rich phenotypic variation. administration seen in this study strongly suggests a nongenomic Because of its well-characterized ZAL2/2m system, the white- rather than genomic mechanism of action (see also refs. 24, 50). throated sparrow is an excellent model for exploring this inter- Also inside the supergene are the genes NCOA7, encoding nu- esting interplay between genetic and epigenetic regulation of clear receptor coactivator 7, and TBP, encoding the TATA box- social behavior. binding protein, both of which interact directly with ERα as The adaptive value of chromosomal inversions has been the transcriptional coactivators (51). Our gene network analysis subject of much interest and debate for nearly a century (44–47). of expression in TnA showed that ESR1 is situated within an Because they suppress recombination, inversions can link co- interconnected module of 157 genes that are differentially adapted alleles together, ensuring coinheritance (45, 48). In expressed between the morphs and that predict territorial singing other words, an inversion can lock together alleles that interact (34). Of these 157 genes, 115 are inside the ZAL2m supergene. well with each other, benefiting the individual and the alleles Thus, we believe that there is great potential in this system to alike. Because it is a transcription factor, ERα interacts with identify coadapted alleles and to show how they have coevolved a large number of other proteins as well as with regulatory inside the supergene.

Merritt et al. PNAS Latest Articles | 5of8 Downloaded by guest on September 29, 2021 Linking coadapted alleles is just one of several hypothesized we made a series of four infusions of antisense or scrambled oligos, two functions of inversions. For example, inversions may capture per day. locally adapted alleles (52) or contribute to speciation (53). The Behavioral testing. Behavioral testing was conducted, as described previously ZAL2/2m arrangement in white-throated sparrows is a special (24), on the day after the fourth infusion. In brief, before a behavioral trial, we placed the focal bird in its cage in an empty sound-attenuating booth to case, however, because of the strong disassortative mating. acclimate it to that environment. After 1 h, we placed the opponent in its Nearly every breeding pair consists of one WS and one TS bird cage immediately adjacent to the focal bird’s cage with an opaque barrier (15, 16), meaning that the supergene is maintained by social between the cages that visually isolated the birds from each other. At the behavior itself (43) independently of geography or local condi- same time, a wax moth larva that had just been injected with 300 μgof m tions. Clearly, ZAL2 is not driving a speciation event, nor has it cyclodextrin-encapsulated 17β-estradiol (E2) or cyclodextrin alone as a con- spread because it is favored in particular environments. Instead, trol (CON) was placed on the floor of the focal bird’s cage. This oral dosage these heteromorphic chromosomes seem more akin to early- of E2 increases plasma E2 to a similar extent in WS and TS birds and produces evolving sex chromosomes (18, 35). Inversions are common on levels typical of breeding WS females within 15 min, as shown by Merritt sex chromosomes, perhaps because they capture sex-determining et al. (24). Ten minutes after the focal bird consumed the larva, the opaque genes together with alleles that benefit a particular sex; in other barrier was removed, and the birds were allowed to interact for 10 min. words, they can bind together a male-determining gene with al- Videos of the trials were later scored for attacks directed toward the op- leles that enhance male fitness (54). Thus, inversions are par- ponent and time spent in proximity to the opponent. After a washout pe- riod of 48 h, each focal bird participated in a second trial with the other ticularly adaptive in the case of antagonistic selection, under hormone treatment (CON or E2). All trials were counterbalanced with re- which certain alleles are beneficial to only one of two alternative spect to the order of treatment. On the day after the second behavioral trial, life-history strategies (e.g., male or female). We hypothesize that brains were harvested, and each cannula was injected with bromophenol m in white-throated sparrows, the ZAL2 arrangement harbors a blue to confirm its placement. Levels of ESR1 and ESR2 expression were collection of alleles beneficial to the WS strategy, characterized measured in samples of TnA using qPCR. The final sample size for each by increased territorial aggression and low parenting, whereas group, each of which included multiple males and females, was n = 6 for WS the ZAL2 harbors alleles that favor the TS strategy. ESR1-KD, n = 7 for WS scrambled, n = 6forTSESR1-KD, and n = 7forTS We recently reported that expression of the gene for vasoac- scrambled. We used separate generalized linear models to test whether the tive intestinal peptide (VIP), located only 345 kb away from effect of hormone treatment depended on morph separately in birds re- ESR1, differs between the morphs and predicts behavior (55). ceiving scrambled or ESR1-KD oligonucleotides, as well as whether the effect WS birds of both sexes have higher levels of VIP expression in of hormone treatment depended on antisense in TS or WS birds (SI Ap- the anterior hypothalamus, a region in which VIP expression is pendix, Table S2). We then followed up significant two-way interactions by testing for the effect of treatment within each experimental group: WS causal for aggression in zebra finches (Taeniopygia guttata) and ESR1-KD, WS scrambled, TS ESR1-KD, and TS scrambled (SI Appendix, violet-eared waxbills (Uraeginthus granatina) (56). In parental Table S3). white-throated sparrows, TS birds have higher levels of VIP expression in the infundibular region (55), which contains the VIP Quantification of ESR1 Allelic Imbalance. cell population that controls prolactin release (57). We fully Animals. Adults and nestlings of both sexes were collected during the expect that additional genes will be shown to contribute to the breeding season at Penobscot Experimental Forest near Argyle, ME. Adults alternative life history strategies of the WS and TS morphs, and were collected during the peak of territorial behavior, after pair formation believe that this species represents an important model with and territory establishment but before or early in the incubation stage (18, which to understand the role of cis-regulatory variation in the 19, 34). Before collecting the adults, we characterized their behavioral re- evolution of behavior. sponses to territorial intrusion by conducting STIs (62). Birds remained on their territories for at least 24 h before we returned to collect tissue, in order Materials and Methods to minimize the effects of the STI itself on gene expression. Later, during the parental phase of the breeding season, we collected nestlings on posthatch day All procedures involving animals were approved by the Emory University 7, 2 d before natural fledging (19, 63). All brains were rapidly dissected from Institutional Animal Care and Use Committee, were in keeping with all the skulls, frozen on dry ice, and stored at −80 °C. RNA and DNA were federal, state, and local laws, and adhered to guidelines set forth by the NIH extracted from samples of TnA, HYP, and POM, and RNA was reverse- Guide for the Care and Use of Laboratory Animals (58). All data were ana- transcribed into cDNA for the allelic imbalance assay (30, 34). lyzed in RStudio (v1.2.1335; R v3.6.0). The α level was set at P ≤ 0.05 unless Allelic imbalance assay. Allelic imbalance was assessed using a multiplexed otherwise specified. More detailed information is provided in SI Appendix, qPCR assay with probes specific for each allele of ESR1. Primers and probes Materials and Methods. were designed by Integrated DNA Technologies to target an A/G SNP within ESR1 exon 1A. Only WS birds were used in these assays, because TS birds do Effects of ESR1 Knockdown on Aggression. not have the ZAL2m allele. We calculated the relative expression of each Animals. We collected 33 white-throated sparrows on the campus of Emory allele within each reaction and then averaged across three replicate reac- University in Atlanta, GA during the fall migration, when gonads are tions. The ratio for each cDNA sample was then normalized to the average regressed and plasma levels of sex steroids are low. We maintained the birds of the ratios calculated from WS genomic DNA (gDNA) samples (mean, 0.99; in a nonbreeding state by housing them under short day lengths (10-h minimum, 0.95; maximum, 1.02) to correct for the relative affinity of each light:14-h dark) (59, 60). Our rationale for testing birds in nonbreeding condition was that the morph difference in ESR1 expression in TnA is pro- probe to its target sequence. We tested for allelic imbalance within each nounced in nonbreeding birds (61) and ERα would be relatively unoccupied region and age using one-sample t tests, comparing the allelic ratios with a and thus manipulable with exogenous E2 (24). Each bird was fitted with null ratio of 1. To determine whether the degree of imbalance varied by age bilateral 26-gauge, stainless steel guide cannulae aimed at TnA (AP, 0.0 mm; or region, we analyzed the data in a two-way mixed ANOVA with region ’ ML, 1.9 mm; DV, −3.6 mm). Birds were allowed to recover from surgery for and age as factors (SI Appendix, Table S5), followed by Tukey s honest sig- at least 3 d before dominance testing. After recovery, dominance relation- nificant difference test (SI Appendix, Table S6). Associations between allelic ’ ships between pairs of birds were established following Merritt et al. (24). imbalance and territorial responses were tested using Pearson s correlation. Dyads were comprised of a dominant bird, which served as the focal animal, and a subordinate opponent. Analysis of Cis-Regulatory Variation in ESR1. Antisense. ESR1 expression was manipulated with custom-designed locked Analysis of transcription factor binding sites. We examined the CREs in ESR1 to m nucleic acid 15-mer antisense oligonucleotides, avoiding ZAL2/2m SNPs identify transcription factor binding sites disrupted by ZAL2/2 .Topredict (ESR1-KD; TTCAAAGGTGGCACT). A scrambled control oligonucleotide was differential transcription factor binding between the two alleles, we used designed to consist of the same nucleotides but in a random order (TAG- sTRAP (38, 64), a tool that predicts the binding affinities of transcription CATGTCAGATCG). Both the ESR1-KD and scrambled sequences were entered factors to each allele and ranks them according to the extent to which the into BLAST (NCBI), and Zonotrichia albicollis refseq_rna was searched to variation affects their binding affinity. We then cross-referenced the list of confirm no significant alignments to other transcripts, particularly ESR2 factors predicted to bind differentially to the two alleles with the list of (XM_014270428.2). Starting on the day after the establishment of a dyad, factors that are expressed in TnA using our RNA-seq data (34).

6of8 | www.pnas.org/cgi/doi/10.1073/pnas.2011347117 Merritt et al. Downloaded by guest on September 29, 2021 Luciferase assays. We cloned each allele of the 2-kb CREs A, B, and C into the bisulfite-converted, N-masked reference genome, and then assigned to an pGL3 control vector upstream of the luciferase gene, LUC (SI Appendix, Fig. allele (65). Allele-specific methylation was determined by Bismark (66). To S3). Constructs were cotransfected with Renilla luciferase into cultured cells. arrive at an overall level of methylation for each of the two alleles for each At 24 h after transfection, luciferase activity was quantified using the bird, we averaged the percent methylation across all the CpGs in that allele. Promega Dual-Glo assay system on a Biotek Synergy plate reader. The value When including unshared CpG sites, we treated that site on the other allele for the ZAL2m allele was normalized to the value for the ZAL2 allele, as 0% methylation. Effects of allele and CRE were tested using linear mixed meaning that the ZAL2m data were expressed relative to a value of 1 for the models, followed by contrasts of estimated marginal means for shared sites ZAL2 allele (Fig. 4B). Effects of CRE, allele, and interactions between CRE and only, as well as for all possible CpG sites. allele were assessed using two-way ANOVA, followed by Student’s pairwise t tests. Data Availability. Sequencing data reported in this manuscript are deposited Analysis of DNA methylation. We bisulfite-converted gDNA extracted from in the National Center for Biotechnology Information under accession samples of TnA from the behaviorally characterized, free-living adults de- PRJNA647725. scribed above. Amplicons containing shared and unshared CpG sites in the three CREs and exon 1A of ESR1 were amplified using PCR with primers not ACKNOWLEDGMENTS. We thank Carlos Rodriguez-Saltos, T. J. Libecap, falling on SNPs or CpGs (SI Appendix, Table S10). Amplicons were pooled per Suzanne Mays, William Hudson, Emily Kim, and Aubrey Kelly for technical sample, indexed, and then pooled for a single sequencing run (PE300) on an assistance. This work was supported by NIH Grant 1R01MH082833 (to D.L.M.), Illumina MiSeq system. Reads were filtered, trimmed, and aligned to a NSF Grant IOS-1627789 (to D.L.M.), and NIH Grant 1F31MH114509 (to J.R.M.).

1. E. O. Wilson, Sociobiology: The New Synthesis, (Harvard University Press, 2000). 27. L. Stetzik et al., Inhibiting ERα expression in the medial amygdala increases prosocial 2. N. Niepoth, A. Bendesky, How natural genetic variation shapes behavior. Annu. Rev. behavior in male meadow voles (Microtus pennsylvanicus). Behav. Brain Res. 351, Genomics Hum. Genet., 10.1146/annurev-genom-111219-080427 (2020). 42–48 (2018). 3. A. Bendesky, C. I. Bargmann, Genetic contributions to behavioural diversity at the 28. B. M. Horton et al., Estrogen receptor α polymorphism in a species with alternative gene-environment interface. Nat. Rev. Genet. 12, 809–820 (2011). behavioral phenotypes. Proc. Natl. Acad. Sci. U.S.A. 111, 1443–1448 (2014). 4. G. E. Robinson, C. M. Grozinger, C. W. Whitfield, Sociogenomics: Social life in mo- 29. K. A. Rosvall et al., Neural sensitivity to sex steroids predicts individual differences in lecular terms. Nat. Rev. Genet. 6, 257–270 (2005). aggression: Implications for behavioural evolution. Proc. Biol. Sci. 279, 3547–3555 5. N. Snyder-Mackler, J. Tung, Vasopressin and the neurogenetics of parental care. (2012). Neuron 95,9–11 (2017). 30. K. E. Grogan, B. M. Horton, Y. Hu, D. L. Maney, A chromosomal inversion predicts the 6. J. E. Madrid, K. J. Parker, A. G. Ophir, Variation, plasticity, and alternative mating expression of sex steroid-related genes in a species with alternative behavioral phe- tactics: revisiting what we know about the socially monogamous prairie vole. Adv. notypes. Mol. Cell. Endocrinol. 495, 110517 (2019). Study Behav. 52, 203–242 (2020). 31. C. V. Mello, T. Kaser, A. A. Buckner, M. Wirthlin, P. V. Lovell, Molecular architecture of 7. A. Bendesky et al., The genetic basis of parental care evolution in monogamous mice. the zebra finch arcopallium. J. Comp. Neurol. 527, 2512–2556 (2019). Nature 544, 434–439 (2017). 32. M. Cheng, M. Chaiken, M. Zuo, H. Miller, Nucleus taenia of the amygdala of birds: NEUROSCIENCE 8. J. Wang et al., A Y-like social chromosome causes alternative colony organization in Anatomical and functional studies in ring doves (Streptopelia risoria) and European fire ants. Nature 493, 664–668 (2013). starlings (Sturnus vulgaris). Brain Behav. Evol. 53, 243–270 (1999). 9. C. Küpper et al., A supergene determines highly divergent male reproductive morphs 33. A. Reiner et al.; Avian Brain Nomenclature Forum, Revised nomenclature for avian in the ruff. Nat. Genet. 48,79–83 (2016). telencephalon and some related brainstem nuclei. J. Comp. Neurol. 473, 377–414 10. S. Lamichhaney et al., Structural genomic changes underlie alternative reproductive (2004). strategies in the ruff (Philomachus pugnax). Nat. Genet. 48,84–88 (2016). 34. W. M. Zinzow-Kramer et al., Genes located in a chromosomal inversion are correlated 11. H. B. Thorneycroft, Chromosomal polymorphism in the white-throated sparrow, Zo- with territorial song in white-throated sparrows. Genes Brain Behav. 14, 641–654 notrichia albicollis (Gmelin). Science 154, 1571–1572 (1966). (2015). 12. H. B. Thorneycroft, A cytogenetic study of the white-throated sparrow, Zonotrichia 35. D. Sun, I. Huh, W. M. Zinzow-Kramer, D. L. Maney, S. V. Yi, Rapid regulatory evolution albicollis (Gmelin). Evolution 29, 611–621 (1975). of a nonrecombining autosome linked to divergent behavioral phenotypes. Proc. 13. K. Ladjali-Mohammedi, J. J. Bitgood, M. Tixier-Boichard, F. A. Ponce De Leon, Inter- Natl. Acad. Sci. U.S.A. 115, 2794–2799 (2018). national system for standardized avian karyotypes (ISSAK): Standardized banded 36. J. M. Buerstedde et al., Light chain gene conversion continues at high rate in an ALV- karyotypes of the domestic fowl (Gallus domesticus). Cytogenet. Cell Genet. 86, induced cell line. EMBO J. 9, 921–927 (1990). 271–276 (1999). 37. P. Winding, M. W. Berchtold, The chicken B cell line DT40: A novel tool for gene 14. J. W. Thomas et al., The chromosomal polymorphism linked to variation in social disruption experiments. J. Immunol. Methods 249,1–16 (2001). behavior in the white-throated sparrow (Zonotrichia albicollis) is a complex re- 38. T. Manke, M. Heinig, M. Vingron, Quantifying the effect of sequence variation on arrangement and suppressor of recombination. Genetics 179, 1455–1468 (2008). regulatory interactions. Hum. Mutat. 31, 477–483 (2010). 15. E. M. Tuttle et al., Divergence and functional degradation of a sex chromosome-like 39. D. R. Rubenstein et al., Sex-specific fitness effects of unpredictable early life condi- supergene. Curr. Biol. 26, 344–350 (2016). tions are associated with DNA methylation in the avian glucocorticoid receptor. Mol. 16. B. M. Horton et al., Behavioral characterization of a white-throated sparrow homo- Ecol. 25, 1714–1728 (2016). zygous for the ZAL2m chromosomal rearrangement. Behav. Genet. 43,60–70 (2013). 40. S. J. Siller, D. R. Rubenstein, A tissue comparison of DNA methylation of the gluco- 17. J. K. Davis et al.; NISC Comparative Sequencing Program, Haplotype-based genomic corticoid receptor gene (NR3C1) in European starlings. Integr. Comp. Biol. 59, 264–272 sequencing of a chromosomal polymorphism in the white-throated sparrow (2019). (Zonotrichia albicollis). J. Hered. 102,380–390 (2011). 41. M. Okhovat, A. Berrio, G. Wallace, A. G. Ophir, S. M. Phelps, Sexual fidelity trade-offs 18. J. B. Falls, J. G. Kopachena, “White-throated sparrow (Zonotrichia albicollis), version promote regulatory variation in the prairie vole brain. Science 350, 1371–1374 (2015). 2.0” in The Birds of North America, A. F. Poole, Ed. (Cornell Lab of Ornithology, 2010). 42. J. Purcell, A. Brelsford, Y. Wurm, N. Perrin, M. Chapuisat, Convergent genetic archi- 19. B. M. Horton, I. T. Moore, D. L. Maney, New insights into the hormonal and behav- tecture underlies social organization in ants. Curr. Biol. 24, 2728–2732 (2014). ioural correlates of polymorphism in white-throated sparrows, Zonotrichia albicollis. 43. D. R. Rubenstein et al., Coevolution of genome architecture and social behavior. Anim. Behav. 93, 207–219 (2014). Trends Ecol. Evol. 34, 844–855 (2019). 20. M. N. Romanov, J. B. Dodgson, R. A. Gonser, E. M. Tuttle, Comparative BAC-based 44. M. Wellenreuther, L. Bernatchez, Eco-evolutionary genomics of chromosomal inver- mapping in the white-throated sparrow, a novel behavioral genomics model, using sions. Trends Ecol. Evol. 33, 427–440 (2018). interspecies overgo hybridization. BMC Res. Notes 4, 211 (2011). 45. T. Dobzhansky, Genetics of the Evolutionary Process, (Columbia University Press, 21. E. D. Ketterson, V. Nolan Jr., Hormones and life histories: An integrative approach. 1970), Vol. vol. 139. Am. Nat. 140, S33–S62 (1992). 46. M. Kirkpatrick, How and why chromosome inversions evolve. PLoS Biol. 8, e1000501 22. J. C. Wingfield, R. E. Hegner, A. M. Dufty Jr., G. F. Ball, The “challenge hypothesis”: (2010). Theoretical implications for patterns of testosterone secretion, mating systems, and 47. A. H. Sturtevant, A case of rearrangement of genes in Drosophila. Proc. Natl. Acad. breeding strategies. Am. Nat. 136, 829–846 (1990). Sci. U.S.A. 7, 235–237 (1921). 23. D. L. Maney, H. S. Lange, M. Q. Raees, A. E. Reid, S. E. Sanford, Behavioral phenotypes 48. T. Dobzhansky, A. H. Sturtevant, Inversions in the chromosomes of Drosophila persist after gonadal steroid manipulation in white-throated sparrows. Horm. Behav. pseudoobscura. Genetics 23,28–64 (1938). 55, 113–120 (2009). 49. P. Dewing et al., Membrane estrogen receptor-α interactions with metabotropic 24. J. R. Merritt et al., Rapid effects of estradiol on aggression depend on genotype in a glutamate receptor 1a modulate female sexual receptivity in rats. J. Neurosci. 27, species with an estrogen receptor polymorphism. Horm. Behav. 98, 210–218 (2018). 9294–9300 (2007). 25. S. Ogawa, D. B. Lubahn, K. S. Korach, D. W. Pfaff, Behavioral effects of estrogen 50. S. A. Heimovics et al., Rapid effects of 17β-estradiol on aggressive behavior in song- receptor gene disruption in male mice. Proc. Natl. Acad. Sci. U.S.A. 94, 1476–1481 birds: Environmental and genetic influences. Horm. Behav. 104,41–51 (2018). (1997). 51. G. Lazennec, T. R. Ediger, L. N. Petz, A. M. Nardulli, B. S. Katzenellenbogen, Mecha- 26. B. C. Trainor, H. H. Kyomen, C. A. Marler, Estrogenic encounters: How interactions nistic aspects of estrogen receptor activation probed with constitutively active es- between aromatase and the environment modulate aggression. Front. Neuroendocrinol. trogen receptors: Correlations with DNA and coregulator interactions and receptor 27,170–179 (2006). conformational changes. Mol. Endocrinol. 11, 1375–1386 (1997).

Merritt et al. PNAS Latest Articles | 7of8 Downloaded by guest on September 29, 2021 52. M. Kirkpatrick, N. Barton, Chromosome inversions, local adaptation and speciation. 60. L. L. Matragrano, M. M. LeBlanc, A. Chitrapu, Z. E. Blanton, D. L. Maney, Testosterone Genetics 173, 419–434 (2006). alters genomic responses to song and monoaminergic innervation of auditory areas 53. A. Navarro, N. H. Barton, Accumulating postzygotic isolation genes in parapatry: A in a seasonally breeding songbird. Dev. Neurobiol. 73, 455–468 (2013). new twist on chromosomal speciation. Evolution 57, 447–459 (2003). 61. D. L. Maney, B. M. Horton, W. M. Zinzow-Kramer, Estrogen receptor alpha as a me- 54. B. Charlesworth, The evolution of sex chromosomes. Science 251, 1030–1033 (1991). diator of life-history trade-offs. Integr. Comp. Biol. 55, 323–331 (2015). 55. B. M. Horton, C. M. Michael, M. R. Prichard, D. L. Maney, Vasoactive intestinal peptide 62. B. M. Horton, M. E. Hauber, D. L. Maney, Morph matters: Aggression bias in a poly- as a mediator of the effects of a supergene on social behaviour. Proc. Biol. Sci. 287, morphic sparrow. PLoS One 7, e48705 (2012). 20200196 (2020). 63. L. Y. Huynh, D. L. Maney, J. W. Thomas, Chromosome-wide linkage disequilibrium 56. J. L. Goodson, A. M. Kelly, M. A. Kingsbury, R. R. Thompson, An aggression-specific caused by an inversion polymorphism in the white-throated sparrow (Zonotrichia cell type in the anterior hypothalamus of finches. Proc. Natl. Acad. Sci. U.S.A. 109, albicollis). Heredity 106, 537–546 (2011). 13847–13852 (2012). 64. M. Okhovat, S. M. Maguire, S. M. Phelps, Methylation of avpr1a in the cortex of wild 57. D. L. Maney, S. J. Schoech, P. J. Sharp, J. C. Wingfield, Effects of vasoactive intestinal prairie voles: Effects of CpG position and polymorphism. R. Soc. Open Sci. 4, 160646 peptide on plasma prolactin in passerines. Gen. Comp. Endocrinol. 113, 323–330 (2017). (1999). 65. D. Sun et al, Genome-wide variation in DNA methylation linked to developmental 58. National Research Council, Guide for the Care and Use of Laboratory Animals, (Na- stage and chromosomal suppression of recombination in white-throated sparrows. tional Academies Press, Washington, DC, 8th Ed., 2011). bioRxiv:10.1101/2020.07.24.220582 (25 July 2020). 59. D. L. Maney, Endocrine and genomic architecture of life history trade-offs in an avian 66. F. Krueger, S. R. Andrews, Bismark: A flexible aligner and methylation caller for model of social behavior. Gen. Comp. Endocrinol. 157, 275–282 (2008). Bisulfite-Seq applications. Bioinformatics 27, 1571– 1572 (2011).

8of8 | www.pnas.org/cgi/doi/10.1073/pnas.2011347117 Merritt et al. Downloaded by guest on September 29, 2021