Evolution of <Emphasis Type="Italic">cd59</Emphasis> gene in mammals

The CD59-coding sequences were obtained from 5 mammals by PCR and BLAST, and combined with the available sequences in GenBank, the nucleotide substitution rates of mammalian cd59 were calculated. Results of synonymous and nonsynonymous substitution rates revealed that cd59 experienced negative selection in mammals overall. Four sites experiencing positive selection were found by using “site-specific” model in PAML software. These sites were distributed on the molecular surface, of which 2 sites located in the key functional domain. Furthermore, “branch-site-specific” model detected 1 positive site in cd59a and cd59b lineages which underwent accelerated evolution caused by positive selection after gene duplication in mouse.     

Extensive concerted evolution of rice paralogs and the road to regaining independence

Many genes duplicated by whole-genome duplications (WGDs) are more similar to one another than expected. We investigated whether concerted evolution through conversion and crossing over, well-known to affect tandem gene clusters, also affects dispersed paralogs. Genome sequences for two Oryza subspecies reveal appreciable gene conversion in the approximately 0.4 MY since their divergence, with a gradual progression toward independent evolution of older paralogs. Since divergence from subspecies indica, approximately 8% of japonica paralogs produced 5-7 MYA on chromosomes 11 and 12 have been affected by gene conversion and several reciprocal exchanges of chromosomal segments, while approximately 70-MY-old "paleologs" resulting from a genome duplication (GD) show much less conversion. Sequence similarity analysis in proximal gene clusters also suggests more conversion between younger paralogs. About 8% of paleologs may have been converted since rice-sorghum divergence approximately 41 MYA. Domain-encoding sequences are more frequently converted than nondomain sequences, suggesting a sort of circularity--that sequences conserved by selection may be further conserved by relatively frequent conversion. The higher level of concerted evolution in the 5-7 MY-old segmental duplication may reflect the behavior of many genomes within the first few million years after duplication or polyploidization.     

Rapid divergence of gene duplicates on the Drosophila melanogaster X chromosome

The recent sequencing of several eukaryotic genomes has generated considerable interest in the study of gene duplication events. The classical model of duplicate gene evolution is that recurrent mutation ultimately results in one copy becoming a pseudogene, and only rarely will a beneficial new function evolve. Here, we study divergence between coding sequence duplications in Drosophila melanogaster as a function of the linkage relationship between paralogs. The mean K(a)/K(s) between all duplicates in the D. melanogaster genome is 0.2803, indicating that purifying selection is maintaining the structure of duplicate coding sequences. However, the mean K(a)/K(s) between duplicates that are both on the X chromosome is 0.4701, significantly higher than the genome average. Further, the distribution of K(a)/K(s) for these X-linked duplicates is significantly shifted toward higher values when compared with the distributions for paralogs in other linkage relationships. Two models of molecular evolution provide qualitative explanations of these observations-relaxation of selective pressure on the duplicate copies and, more likely, positive selection on recessive adaptations. We also show that there is an excess of X-linked duplicates with low K(s), suggesting a larger proportion of relatively young duplicates on the D. melanogaster X chromosome relative to autosomes.     

Excess of amino acid substitutions relative to polymorphism between X-linked duplications in Drosophila melanogaster

We have obtained sequence polymorphism data from 13 genes belonging to 5 gene families in Drosophila melanogaster where the K(a)/K(s) between copies is greater than 1. Twelve of these 13 loci are X-linked. In general, there is evidence of purifying selection in all families, as inferred both from levels of silent and replacement variation and insertion/deletion variation, suggesting that the loci are likely functional. Shared polymorphisms indicative of gene conversion between paralogs are rare among the X-linked families, in contrast to available data from autosomal duplicates. McDonald-Kreitman tests between duplicates reveal an excess of amino-acid fixations between copies in the X-linked families, suggesting that the divergence between these loci was driven by positive selection. In contrast, available data from autosomal duplicates show a deficit of fixations, consistent with gene conversion being a strong homogenizing force.     

Early evolutionary history and genomic features of gene duplicates in the human genome

Background

Human gene duplicates have been the focus of intense research since the development of array-based and targeted next-generation sequencing approaches in the last decade. These studies have primarily concentrated on determining the extant copy-number variation from a population-genomic perspective but lack a robust evolutionary framework to elucidate the early structural and genomic characteristics of gene duplicates at emergence and their subsequent evolution with increasing age.

Results

We analyzed 184 gene duplicate pairs comprising small gene families in the draft human genome with 10 % or less synonymous sequence divergence. Human gene duplicates primarily originate from DNA-mediated events, taking up genomic residence as intrachromosomal copies in direct or inverse orientation. The distribution of paralogs on autosomes follows random expectations in contrast to their significant enrichment on the sex chromosomes. Furthermore, human gene duplicates exhibit a skewed gradient of distribution along the chromosomal length with significant clustering in pericentromeric regions. Surprisingly, despite the large average length of human genes, the majority of extant duplicates (83 %) are complete duplicates, wherein the entire ORF of the ancestral copy was duplicated. The preponderance of complete duplicates is in accord with an extremely large median duplication span of 36 kb, which enhances the probability of capturing ancestral ORFs in their entirety. With increasing evolutionary age, human paralogs exhibit declines in (i) the frequency of intrachromosomal paralogs, and (ii) the proportion of complete duplicates. These changes may reflect lower survival rates of certain classes of duplicates and/or the role of purifying selection. Duplications arising from RNA-mediated events comprise a small fraction (11.4 %) of all human paralogs and are more numerous in older evolutionary cohorts of duplicates.

Conclusions

The degree of structural resemblance, genomic location and duplication span appear to influence the long-term maintenance of paralogs in the human genome. The median duplication span in the human genome far exceeds that in C. elegans and yeast and likely contributes to the high prevalence of complete duplicates relative to structurally heterogeneous duplicates (partial and chimeric). The relative roles of regulatory sequence versus exon-intron structure changes in the acquisition of novel function by human paralogs remains to be determined.

Electronic supplementary material

The online version of this article (doi:10.1186/s12864-015-1827-3) contains supplementary material, which is available to authorized users.     

Duplicated genes evolve slower than singletons despite the initial rate increase

Background

Gene duplication is an important mechanism that can lead to the emergence of new functions during evolution. The impact of duplication on the mode of gene evolution has been the subject of several theoretical and empirical comparative-genomic studies. It has been shown that, shortly after the duplication, genes seem to experience a considerable relaxation of purifying selection.

Results

Here we demonstrate two opposite effects of gene duplication on evolutionary rates. Sequence comparisons between paralogs show that, in accord with previous observations, a substantial acceleration in the evolution of paralogs occurs after duplication, presumably due to relaxation of purifying selection. The effect of gene duplication on evolutionary rate was also assessed by sequence comparison between orthologs that have paralogs (duplicates) and those that do not (singletons). It is shown that, in eukaryotes, duplicates, on average, evolve significantly slower than singletons. Eukaryotic ortholog evolutionary rates for duplicates are also negatively correlated with the number of paralogs per gene and the strength of selection between paralogs. A tally of annotated gene functions shows that duplicates tend to be enriched for proteins with known functions, particularly those involved in signaling and related cellular processes; by contrast, singletons include an over-abundance of poorly characterized proteins.

Conclusions

These results suggest that whether or not a gene duplicate is retained by selection depends critically on the pre-existing functional utility of the protein encoded by the ancestral singleton. Duplicates of genes of a higher biological import, which are subject to strong functional constraints on the sequence, are retained relatively more often. Thus, the evolutionary trajectory of duplicated genes appears to be determined by two opposing trends, namely, the post-duplication rate acceleration and the generally slow evolutionary rate owing to the high level of functional constraints.

     

<Emphasis Type="Italic">Hox</Emphasis> cluster duplication in the basal teleost <Emphasis Type="Italic">Hiodon alosoides</Emphasis> (Osteoglossomorpha)

Large-scale—even genome-wide—duplications have repeatedly been invoked as an explanation for major radiations. Teleosts, the most species-rich vertebrate clade, underwent a “fish-specific genome duplication” (FSGD) that is shared by most ray-finned fish lineages. We investigate here the Hox complement of the goldeye (Hiodon alosoides), a representative of Osteoglossomorpha, the most basal teleostean clade. An extensive PCR survey reveals that goldeye has at least eight Hox clusters, indicating a duplicated genome compared to basal actinopterygians. The possession of duplicated Hox clusters is uncoupled to species richness. The Hox system of the goldeye is substantially different from that of other teleost lineages, having retained several duplicates of Hox genes for which crown teleosts have lost at least one copy. A detailed analysis of the PCR fragments as well as full length sequences of two HoxA13 paralogs, and HoxA10 and HoxC4 genes places the duplication event close in time to the divergence of Osteoglossomorpha and crown teleosts. The data are consistent with—but do not conclusively prove—that Osteoglossomorpha shares the FSGD. Electronic supplementary material  The online version of this article (doi:) contains supplementary material, which is available to authorized users.

Regulatory Evolution of a Duplicated Heterodimer Across Species and Tissues of Allopolyploid Clawed Frogs (<Emphasis Type="Italic">Xenopus</Emphasis>)

Changes in gene expression contribute to reproductive isolation of species, adaptation, and development and may impact the genetic fate of duplicated genes. African clawed frogs (genus Xenopus) offer a useful model for examining regulatory evolution, particularly after gene duplication, because species in this genus are polyploid. Additionally, these species can produce viable hybrids, and expression divergence between coexpressed species-specific alleles in hybrids can be attributed exclusively to cis-acting mechanisms. Here we have explored expression divergence of a duplicated heterodimer composed of the recombination activating genes 1 and 2 (RAG1 and RAG2). Previous work identified a phylogenetically biased pattern of pseudogenization of RAG1 wherein one duplicate—RAG1β—was more likely to become a pseudogene than the other one—RAG1α. In this study we show that ancestral expression divergence between these duplicates could account for this. Using comparative data we demonstrate that regulatory divergence between species and between duplicated genes varies significantly across tissue types. These results have implications for understanding of variables that influence pseudogenization of duplicated genes generated by polyploidization, and for interpretation of the relative contributions of cis versus trans mechanisms to expression divergence at the cellular level. Electronic supplementary material  The online version of this article (doi:) contains supplementary material, which is available to authorized users.     

A Revised Evolutionary History of the CYP1A Subfamily: Gene Duplication, Gene Conversion, and Positive Selection

Members of cytochrome P450 subfamily 1A (CYP1As) are involved in detoxification and bioactivation of common environmental pollutants. Understanding the functional evolution of these genes is essential to predicting and interpreting species differences in sensitivity to toxicity caused by such chemicals. The CYP1A gene subfamily comprises a single ancestral representative in most fish species and two paralogs in higher vertebrates, including birds and mammals. Phylogenetic analysis of complete coding sequences suggests that mammalian and bird paralog pairs (CYP1A1/2 and CYP1A4/5, respectively) are the result of independent gene duplication events. However, comparison of vertebrate genome sequences revealed that CYP1A genes lie within an extended region of conserved fine-scale synteny, suggesting that avian and mammalian CYP1A paralogs share a common genomic history. Algorithms designed to detect recombination between nucleotide sequences indicate that gene conversion has homogenized most of the length of the chicken CYP1A genes, as well as the 5′ end of mammalian CYP1As. Together, these data indicate that avian and mammalian CYP1A paralog pairs resulted from a single gene duplication event and that extensive gene conversion is responsible for the exceptionally high degree of sequence similarity between CYP1A4 and CYP1A5. Elevated nonsynonymous/synonymous substitution ratios within a putatively unconverted stretch of ∼250 bp suggests that positive selection may have reduced the effective rate of gene conversion in this region, which contains two substrate recognition sites. This work significantly alters our understanding of functional evolution in the CYP1A subfamily, suggesting that gene conversion and positive selection have been the dominant processes of sequence evolution. Electronic Supplementary Material Electronic Supplementary material is available for this article at and accessible for authorised users. [Reviewing Editor: Dr. Yves Van de Peer]     

Relaxed Selection Among Duplicate Floral Regulatory Genes in Lamiales

Polyploidization is a prevalent mode of genome diversification within plants. Most gene duplicates arising from polyploidization (paralogs) are typically lost, although a subset may be maintained under selection due to dosage, partitioning of gene function, or acquisition of novel functions. Because they experience selection in the presence of other duplicate loci across the genome, interactions among genes may also play a significant role in the maintenance of paralogs resulting from polyploidization. Previously, we identified duplicates of the genes LFY/FLO and AP3/DEF that directly interact in a floral regulatory pathway and are thought to be the result of ancient polyploidization in the Lamiales (> 50 mya). Although duplicates of MADS box genes including AP3/DEF are common throughout the angiosperm lineage, LFY/FLO duplicates in Lamiales are the first reported outside of tetraploid taxa. In order to explore hypotheses for the joint preservation of these interacting floral regulatory genes including novel LFY/FLO paralogs, here we clone FLO and DEF duplicates from additional Lamiales taxa and apply codon substitution models to test how selection acts on both genes following duplication. We find acceleration in the ratio of nonsynonymous-to-synonymous nucleotide substitutions for one (FLO) or both (DEF) paralogs that appears to be due to relaxed purifying selection as opposed to positive selection and shows a different pattern among functional domains of these genes. Several mechanisms are discussed that might be responsible for preservation of co-orthologs of FLO and DEF in Lamiales, including interactions among the genes of this regulatory pathway. Electronic Supplementary Material Electronic Supplementary material is available for this article at and accessible for authorised users. [Reviewing Editor: Dr. Yves Van de Peer]     

Heterogeneous Selective Pressure Acting on Influenza B Victoria- and Yamagata-Like Hemagglutinins

As a consequence of immune pressure, influenza virus hemagglutinin presents some of its amino acids under positive selection. Several authors have reported the existence of influenza A hemagglutinin codons under positive selective pressure (PSP). In this framework, the present work objectives were to demonstrate the presence of PSP and evaluate its effects on Victoria- and Yamagata-like influenza B viruses. Methodology adopted consisted in estimating the acceptance rate of nonsynonymous substitutions (ω = d_N/d_S) that describe the strength of selective pressure and identifying codons that may be positively selected, applying a set of continuous-time Markov chain codon-substitution models. Two groups of HA1 sequences (140 from Yamagata and 60 from Victoria lineage) were used. All the model maximum-likelihood estimates were obtained using codeml software application (PAML 3.15). The hypothesis of no existence of sites under PSP was rejected for both lineages (p < 0.001), using likelihood ratio tests. These results demonstrate the presence of positive selection acting on hemagglutinin of both Yamagata- and Victoria-like influenza B viruses. Several different sites were identified to be under PSP on Yamagata and Victoria hemagglutinins. Sites found with a posterior probability > 0.95 were codons 197 and 199 in both lineages, codon 75 in the Yamagata lineage, and codon 129 in the Victoria lineage. The detected amino acids are located at or near antigenic sites in influenza A virus H3 hemagglutinin. Electronic supplementary material  The online version of this article (doi:) contains supplementary material, which is available to authorized users.     

Retention of protein complex membership by ancient duplicated gene products in budding yeast

To investigate functional divergence of gene duplicates, we examined the protein-protein interactions and coexistence in complexes of paralogs resulting from an ancient whole-genome duplication in yeast. Strikingly, half the surveyed paralog pairs were found to be co-clustered in protein complexes, and were more conserved and highly expressed than non-co-clustered paralogs; however, their discordant expression patterns and conservation rates indicate differential regulation of subfunctionalized paralogs. These results highlight the value of protein complex membership in studying functional divergence among gene duplicates.     

Comparative genomic analysis of teleost fish <Emphasis Type="Italic">bmal</Emphasis> genes

Bmal1 (Brain and muscle ARNT like 1) gene is a key circadian clock gene. Tetrapods also have the second Bmal gene, Bmal2. Fruit fly has only one bmal1/cycle gene. Interrogation of the five teleost fish genome sequences coupled with phylogenetic and splice site analyses found that zebrafish have two bmal1 genes, bmal1a and bmal1b, and bmal2a; Japanese pufferfish (fugu), green spotted pufferfish (tetraodon) and Japanese medaka fish each have two bmal2 genes, bmal2a and bmal2b, and bmal1a; and three-spine stickleback have bmal1a and bmal2b. Syntenic analysis further indicated that zebrafish bmal1a/bmal1b, and fugu, tetraodon and medaka bmal2a/bmal2b are ancient duplicates. Although the dN/dS ratios of these four fish bmal duplicates are all <1, implicating they have been under purifying selection, the Tajima relative rate test showed that fugu, tetraodon and medaka bmal2a/bmal2b have asymmetric evolutionary rates, suggesting that one of these duplicates have been subject to positive selection or relaxed functional constraint. These results support the notion that teleost fish bmal genes were derived from the fish-specific genome duplication (FSGD), divergent resolution following the duplication led to retaining different ancient bmal duplicates in different fishes, which could have shaped the evolution of the complex teleost fish timekeeping mechanisms. Electronic supplementary material  The online version of this article (doi:) contains supplementary material, which is available to authorized users.     

Evolution of cd59 gene in mammals

The CD59-coding sequences were obtained from 5 mammals by PCR and BLAST, and combined with the available sequences in GenBank, the nucleotide substitution rates of mammalian cd59 were calcu- lated. Results of synonymous and nonsynonymous substitution rates revealed that cd59 experienced negative selection in mammals overall. Four sites experiencing positive selection were found by using "site-specific" model in PAML software. These sites were distributed on the molecular surface, of which 2 sites located in the key functional domain. Furthermore, "branch-site-specific" model detected 1 positive site in cd59a and cd59b lineages which underwent accelerated evolution caused by positive selection after gene duplication in mouse.     

The Roles of Gene Duplication,Gene Conversion and Positive Selection in Rodent Esp and Mup Pheromone Gene Families with Comparison to the Abp Family

Three proteinaceous pheromone families, the androgen-binding proteins (ABPs), the exocrine-gland secreting peptides (ESPs) and the major urinary proteins (MUPs) are encoded by large gene families in the genomes of Mus musculus and Rattus norvegicus. We studied the evolutionary histories of the Mup and Esp genes and compared them with what is known about the Abp genes. Apparently gene conversion has played little if any role in the expansion of the mouse Class A and Class B Mup genes and pseudogenes, and the rat Mups. By contrast, we found evidence of extensive gene conversion in many Esp genes although not in all of them. Our studies of selection identified at least two amino acid sites in β-sheets as having evolved under positive selection in the mouse Class A and Class B MUPs and in rat MUPs. We show that selection may have acted on the ESPs by determining K_a/K_s for Exon 3 sequences with and without the converted sequence segment. While it appears that purifying selection acted on the ESP signal peptides, the secreted portions of the ESPs probably have undergone much more rapid evolution. When the inner gene converted fragment sequences were removed, eleven Esp paralogs were present in two or more pairs with K_a/K_s >1.0 and thus we propose that positive selection is detectable by this means in at least some mouse Esp paralogs. We compare and contrast the evolutionary histories of all three mouse pheromone gene families in light of their proposed functions in mouse communication.     

A Population-Genetic Lens into the Process of Gene Loss Following Whole-Genome Duplication

Whole-genome duplications (WGDs) have occurred in many eukaryotic lineages. However, the underlying evolutionary forces and molecular mechanisms responsible for the long-term retention of gene duplicates created by WGDs are not well understood. We employ a population-genomic approach to understand the selective forces acting on paralogs and investigate ongoing duplicate-gene loss in multiple species of Paramecium that share an ancient WGD. We show that mutations that abolish protein function are more likely to be segregating in retained WGD paralogs than in single-copy genes, most likely because of ongoing nonfunctionalization post-WGD. This relaxation of purifying selection occurs in only one WGD paralog, accompanied by the gradual fixation of nonsynonymous mutations and reduction in levels of expression, and occurs over a long period of evolutionary time, “marking” one locus for future loss. Concordantly, the fitness effects of new nonsynonymous mutations and frameshift-causing indels are significantly more deleterious in the highly expressed copy compared with their paralogs with lower expression. Our results provide a novel mechanistic model of gene duplicate loss following WGDs, wherein selection acts on the sum of functional activity of both duplicate genes, allowing the two to wander in expression and functional space, until one duplicate locus eventually degenerates enough in functional efficiency or expression that its contribution to total activity is too insignificant to be retained by purifying selection. Retention of duplicates by such mechanisms predicts long times to duplicate-gene loss, which should not be falsely attributed to retention due to gain/change in function.     

Phylogeny reconstruction in the Caesalpinieae grade (Leguminosae) based on duplicated copies of the sucrose synthase gene and plastid markers

The Caesalpinieae grade (Leguminosae) forms a morphologically and ecologically diverse group of mostly tropical tree species with a complex evolutionary history. This grade comprises several distinct lineages, but the exact delimitation of the group relative to subfamily Mimosoideae and other members of subfamily Caesalpinioideae, as well as phylogenetic relationships among the lineages are uncertain. With the aim of better resolving phylogenetic relationships within the Caesalpinieae grade, we investigated the utility of several nuclear markers developed from genomic studies in the Papilionoideae. We cloned and sequenced the low copy nuclear gene sucrose synthase (SUSY) and combined the data with plastid trnL and matK sequences. SUSY has two paralogs in the Caesalpinieae grade and in the Mimosoideae, but occurs as a single copy in all other legumes tested. Bayesian and maximum likelihood phylogenetic analyses suggest the two nuclear markers are congruent with plastid DNA data. The Caesalpinieae grade is divided into four well-supported clades (Cassia, Caesalpinia, Tachigali and Peltophorum clades), a poorly supported clade of Dimorphandra Group genera, and two paraphyletic groups, one with other Dimorphandra Group genera and the other comprising genera previously recognized as the Umtiza clade. A selection analysis of the paralogs, using selection models from PAML, suggests that SUSY genes are subjected to a purifying selection. One of the SUSY paralogs, under slightly stronger positive selection, may be undergoing subfunctionalization. The low copy SUSY gene is useful for phylogeny reconstruction in the Caesalpinieae despite the presence of duplicate copies. This study confirms that the Caesalpinieae grade is an artificial group, and highlights the need for further analyses of lineages at the base of the Mimosoideae.     

Detecting amino acid sites under positive selection and purifying selection

An excess of nonsynonymous over synonymous substitution at individual amino acid sites is an important indicator that positive selection has affected the evolution of a protein between the extant sequences under study and their most recent common ancestor. Several methods exist to detect the presence, and sometimes location, of positively selected sites in alignments of protein-coding sequences. This article describes the "sitewise likelihood-ratio" (SLR) method for detecting nonneutral evolution, a statistical test that can identify sites that are unusually conserved as well as those that are unusually variable. We show that the SLR method can be more powerful than currently published methods for detecting the location of positive selection, especially in difficult cases where the strength of selection is low. The increase in power is achieved while relaxing assumptions about how the strength of selection varies over sites and without elevated rates of false-positive results that have been reported with some other methods. We also show that the SLR method performs well even under circumstances where the results from some previous methods can be misleading.     

An ancient frame-shifting event in the highly conserved KPNA gene family has undergone extensive compensation by natural selection in vertebrates

One of the prevailing arguments in evolutionary theory is that the duplicates of genes can acquire novel functionality. This is because only one of the paralogs need maintain the ancestral function, leaving room for natural experimentation due to a respite in purifying selection. Although many duplicates can subsequently become disabled by nullifying mutations, a few may also go on to diverge along a novel evolutionary trajectory. Here, evidence is provided that demonstrates how this scenario may not always be true. Rather, in the case of the highly conserved KPNA importin family, an initial relaxation in selection induced a frameshift that was later suppressed and heavily compensated for as part of a reparative and optimizing process. Despite a resulting divergence, there remains a distinct preservation of both sequence and functionality among the paralogs. This would indicate that duplicates can be retained by selection for reasons related to their redundant functionality. It also shows that, even when positive selection is inferred in duplicate genes, this may be of a compensatory nature rather than one representing any biochemical innovation. Generally, this development would perhaps be a more common outcome for gene duplication than is currently maintained.     

Positive Darwinian Selection at Single Amino Acid Sites Conferring Plant Virus Resistance

Explicit evaluation of the accuracy and power of maximum likelihood and Bayesian methods for detecting site-specific positive Darwinian selection presents a challenge because selective consequences of single amino acid changes are generally unknown. We exploited extensive molecular and functional characterization of amino acid substitutions in the plant gene eIF4E to evaluate the performance of these methods in detecting site-specific positive selection. We documented for the first time a molecular signature of positive selection within a recessive resistance gene in plants. We then used two statistical platforms, Phylogenetic Analysis Using Maximum Likelihood and Hypothesis Testing Using Phylogenies (HyPhy), to look for site-specific positive selection. Their relative power and accuracy are assessed by comparing the sites they identify as being positively selected with those of resistance-determining amino acids. Our results indicate that although both methods are surprisingly accurate in their identification of resistance sites, HyPhy appears to more accurately identify biologically significant amino acids using our data set. Electronic supplementary material  The online version of this article (doi:) contains supplementary material, which is available to authorized users. J. R. Cavatorta and A. E. Savage have contributed equally to this work.