How does geographic isolation contribute to evolution? Mandira P. Sep 1, Explanation: All members of a species form a single population. Related questions How do I determine the molecular shape of a molecule? What is the lewis structure for co2? What is the lewis structure for hcn? We can understand this with a decomposition described by [ 39 ]. When we condition on survival until t , we condition on survival of at least one trunk lineage the long red lineage in Fig 3.
Given this lineage, the other reproduction events are undistinguished—the genealogy of a family that survives until t can be decomposed into a trunk that lasts until t and that sprouts independent, unconditioned branches that may or may not survive until t. Since each subfamily Z s , k is a branching process with mean offspring number less than one, most of these die out quickly, and so Z t is composed of a cluster of recently split subfamilies.
If we take the expectation of Eq 7 , since , we get the limiting mean family size conditioned on nonextinction: 8. Both numerator and denominator in Eq 9 count only families that have arrived at the new patch, so we can consider a family conditioned on having successfully transited between the patches.
To obtain simply interpretable expressions, we assume that once any member of the family arrives at the patch, all other members are close by, and so have roughly equal opportunity to establish in the new patch. Suppose the family arrives with K individuals. On the other hand, in the absence of the patch of new habitat, each of the K newly arrived individuals would leave behind a lineage with mean total size , roughly half of which would stay in the new patch.
For more careful consideration of the geometry of the patch and the distinction between occupation density and number of individuals, see section Hitting and occupation , below. This suggests that 10 There are now two extremes to consider. If is small, then each family has a small chance of establishment, and the numerator is approximately. On the other hand, if p e is not too small, then the numerator would be close to 1, as any family that arrives will likely establish.
The remaining term is q S , the integral of Eq 3 over the patch. Since the patches are at distance R , if the new patch has width smaller than , then q S is approximately q R multiplied by the area of the new patch. If S is larger, then we should use the area of the closest strip of width to the original patch; see Fig 3 , and section The integral q S for more details.
Note that the approximation has the undesirable property that it is non-monotonic in s m , when we expect the rate of adaptation via migration to be a decreasing function of s m. Now that we have expressions for the mean rates of adaptation by new mutation, Eq 2 , and by migration from an already colonized patch, Eq 11 , it seems helpful to step back and review the assumptions underlying the asymptotic results we have used, or will use below.
Our results should hold exactly in the limit of large, circular patches sufficiently far apart, large population density, and small selection coefficients of equal magnitude. As for the migration rate, we assume that each patch is large enough to support a stable population of B alleles.
The geometry and size of the patch will also affect the approximation of Eq The last requirement is necessary because if the B allele fixes in large neighborhoods where it is deleterious, we cannot approximate its dynamics via a branching process. We also neglect the time for migration-selection equilibrium to be reached. As discussed above, we also assume that migration to a new patch takes place over a number of generations; if there are sufficient rare, long-distance migration events that would move between patches in a single hop, this would require a different analysis.
More details are given below in section Simulation methods , and the number of simulations used and parameter values are given in supplementary S1 and S2 Tables. Fig 4 summarizes how the results compare to theory, excluding parameter combinations that violate the assumptions discussed above, or where a majority of simulations did not adapt by 25, generations. For each, parameter combinations and numbers of simulations are in supplemental S1 and S2 Tables; shown are the median points and interquartile range lines of the time until the patch had at least adapted individuals, for each unique parameter combination.
The agreement is reasonable for both. This is to be expected for two reasons: First, we compute the time to reach B alleles, while theory predicts the time until the progenitor of those B alleles arose. S10 Fig shows simulations at more parameter values. We now turn to the question of whether the separated patches adapt by parallel genetic changes or by the spread of a migrant allele between the patches. As only a single mutation is required for individuals to adapt to the poor habitat patch, subsequent mutations that arise after an allele becomes established in the patch gain no selective benefit.
Similarly, an allele introduced into a patch by migration will not be favored by selection to spread, if the patch has already been colonized. Therefore, mutations selectively exclude each other from patches, over short time scales, and will only slowly mix together over longer time scales by drift and migration, an assumption we also made in [ 6 ].
In reality, different mutations will likely not be truly selectively equivalent, in which case this mixing occurs over a time-scale dictated by the difference in selection coefficients. We assume that once a B allele is introduced by migration or mutation it becomes established in the poor habitat patch rapidly if it escapes loss by demographic stochasticity.
As such, the question of how the second patch adapts simply comes down to whether a new mutation or a migrant allele is the first to become established in the second patch. To work out how adaptation is likely to proceed, we can compare Expressions 2 and 11 above for the effective rates of adaptation by new mutation and by migration.
We work in one dimension, as the square root term appearing for two dimensions is relatively unimportant. We first consider the order of magnitude that our parameters need to be on in order for adaptation via mutation or migration to dominate. Effective migration and mutation rates will be on the same order if , where R is the distance between the patches.
Equivalently, the rates are roughly equal if , which gives the critical gap size past which adaptation will be mostly parallel in terms of selection coefficients, patch width, and mutation rate. We can go beyond these rough calculations to find the probability of parallel adaptation if we are willing to take our approximations at face value.
Furthermore, the probability that the second adaptation is a new mutation, i. We tested this using the same simulations as Fig 4 , by using the empirical distributions of the respective times to adaptation to estimate the probability, for each parameter combination, that a new, successful mutation appears before a successful migrant arrives from another, already adapted patch. The results are compared to Eq 12 in Fig 5. Points correspond to parameter combinations of the simulations in Fig 4 , with the shade of grey and the red numeric label indicating the same probability estimated from these simulations see text for details.
Suppose that instead of patches there are many, each patch adapting through either migration or mutation. Suppose that patches 1, …, k are already adapted, and that these patches are at distances R 1 , …, R k away from this unadapted patch.
The total rate of adaptation through migration, from Eq 11 if patches do not interfere with each other, is 13 If the patch adapts through migration, the probability the adapted allele comes from patch i is We can also rescale time by the typical patch size, and introduce a parameter, say , making the properties other than the time scaling of the discrete model independent of the numerical sizes of the demes themselves.
We can say more if we suppose that all patches have the same area and are the same distance from each other i. Now we can compute most properties we might want about the process. For instance, the proportion of demes that shares the same origin as a randomly sampled deme is approximately Beta distributed—see [ 41 ]. The high connectedness of the discrete deme island model means that the expected number of distinct alleles grows with the log of the number of demes.
This strongly contrasts with the continuous spatial model where the local nature of dispersal means that doubling the species range will double the number of mutations expected. If a patch adapts through new mutation or a rare migrant lineage, the genomic region surrounding the selected site will hitchhike along with it [ 42 ], so that at least initially, all adapted individuals within a patch share a fairly long stretch of haplotype.
This association gets slowly whittled down by recombination during interbreeding at the edge of the patch, but there will always be longer LD nearby to the selected site [ 44 ]. When an already adapted patch colonizes another through migration, the newly colonized patch will inherit a long piece of haplotype around the selected site from the originating patch.
A large, linked haplotype may still arrive and fix in the new patch, in which case the haplotype has literally hitchhiked across geography!
Fig 6 shows a simulation of such an event, including the lineage that founds an adaptive population on a second patch, and the length of the haplotype shared between this lineage and one in the original patch. The time that the lineage is outside the region where the B allele is common dark red in Fig 6 , the haplotype that accompanies it is broken down rapidly.
After the lineage establishes on the patch, the rate of decay of the haplotype is slowed significantly, since most others with which it recombines have similar haplotypic backgrounds.
In the left panel, as in Fig 2 , color and contours show allele frequency across space and time; other simulation parameters are as in Figs 1B and 2. The right panel shows in red the extent of the shared haplotype between the two lineages around the selected locus at position 0 , which decreases as time progresses.
In so doing, we neglect recombination between B alleles since they are at low density in transit , and the possibility that more than one successful migrant family is in transit at once so that faster migrants would be more likely to have arrived first. Therefore, if L is the length of hitchhiking segment on, say, the right of the selected locus, then We can then use standard results on hitting times of d -dimensional Brownian motion that is killed at rate s m see [ 45 ] 2.
In particular, if the patch is circular with radius w and lies at distance R from the already adapted patch, then 16 where K 0 is a modified Bessel function of the second kind. We are interested in lineages that manage to reach the patch before being killed, i.
To keep the expressions simple, in the remainder of this section we only compute quantities for one dimension. A saddle point approximation provides an alternative route to these expressions. The form of Eq 17 implies that if Y is an exponential random variable with rate , then L has the same distribution as. Furthermore, the expected length of shared hitchhiking haplotype is 19 and. For two dimensions, asymptotics of Bessel functions show that L has the same tail behavior, but how other properties might change is unclear.
As a rule of thumb, Eq 18 means that families who successfully establish move at speed towards the new patch—if s m , the strength of the selection against them, is smaller, the need to move quickly is less imperative. Then, Eq 19 means that the haplotype length is roughly the length one would expect given the mean transit time, so more weakly deleterious transiting alleles arrive with them shorter haplotypes.
Coat color in the rock pocket mouse Chaetodipus intermedius is a classic example of local adaptation [ 14 , 15 ]. These mice live on rocky outcrops throughout the southwestern United States and northern Mexico, and generally have a light pelage similar to the predominant rock color. However, in some regions these mice live on darker substrates e. Some of the largest patches of dark rock range from 10km to km wide and lie 50—km from each other, and dark-colored populations of C. However, patches of all sizes occur across all scales in a heterogeneous manner.
This dark allele differs from the light allele by four amino acid changes, and has a dominant or partially dominant effect depending on the measure of coat color. The Pinacate allele is not present in a number of other populations with dark pelage, suggesting these populations have adapted in parallel [ 17 , 46 ].
However, [ 47 ] reasoned that, elsewhere in the range, multiple close dark outcrops may share a dark phenotype whose genetic basis has been spread by migration despite intervening light habitat. A key parameter above was the dispersal distance divided by the square root of strength of selection against the focal allele between patches,. The dark allele was entirely absent from Christmas pass, a site with light substrate 7km north of Tule, and 3km further from the lava flow.
Note that such apparent asymmetry is expected, since as noted above the migration—selection equilibrium can be highly stochastic. These numbers give us a sense of a plausible range of the parameters. We also need estimate of s m. While the Pinacate dark haplotype differs from the light haplotype at four amino acid residues, it is likely that not all of these changes are needed for a population to begin to adapt. Also, there a number of genes besides MC1R at which adaptive changes affecting pigmentation have been identified in closely related species and more broadly across vertebrates [ 51 ].
In Fig 7 we show the dependence of the probability of parallel mutation on the distance between lava flow patches using these parameters, showing that parallel mutation should become likely over a scale of tens to a few hundred kilometers between patches.
Top panel: The probability of parallel mutation as a function of the distance between environmental patches for two different cline widths and two different mutation rates using Eq The parameters were chosen to match the example of Chaetodipus intermedius ; see text.
Bottom panel: The initial genetic length of the haplotype shared between patches due to adaptation via migration as a function of the distance between environmental patches. As a rough rule of thumb 1cM is approximately 10 6 bases in a number of mammalian species. Given the large selection coefficient associated with the dark allele on the dark substrate, we expect the initial haplotype associated with either a new mutation or migrant allele to be large.
Fig 7 also shows how long the founding haplotype shared between populations is expected to be, from Eq The initial length can be quite long between geographically close patches tens of kilometers. However, for the wider cline width , adaptation by migration can still be likely for patches km apart, but the shared basis may be hard to detect, as the length of shared haplotype can be quite short.
This paper is an investigation into the basic question: What is the spatial resolution of convergent local adaptation? In other words, over what spatial scale of environmental patchiness will the process of adaptation develop independent solutions to evolutionary problems? The answer to this depends most strongly on , the dispersal distance divided by the square root of the strength of selection against the allele between patches.
It depends much more weakly on the selective benefit within the patches or perhaps surprisingly the population density, although these two factors will determine the time-scale over which adaptation will occur and note that population density could affect the selection coefficients. This is in contrast with models of panmictic populations [ 8 , 52 , 53 ] and geographically spread populations adapting to homogeneous selection pressures [ 6 ], where the probability of multiple, independently arising adaptive alleles increases with the population size.
However, in all of these models the dependence on the beneficial selection coefficient is absent or weak, due to the fact that selection both aids establishment of new alleles and the spread of existing alleles but see [ 53 ] for the complications of varying population sizes.
We have also shown that while weaker selection against alleles will make sharing of adaptations between patches easier, it also makes it harder to spot such sharing, since the lucky alleles that manage to colonize new patches move slower, and thus carry a shorter shared haplotype.
This issue is amplified by the fact that the length of haplotype shared within patches decays over time, potentially making the identification of shared adaptations to old selection pressures difficult. Perhaps the most useful rule-of-thumb quantities we found were the following. Equivalently, the critical gap size between patches past which adaptation is likely independent is. Finally, successfully transiting migrant lineages move at rate , and so shared haplotype lengths between patches will be of order.
In developing the set of approximations in the paper we have ignored a number of complicating factors. We now briefly discuss these. We have focused on relative rates of adaptation, since in applications where adaptation has occurred, the question is whether adaptations in distinct patches have appeared independently or not. However, any adaptation that does occur may have to make use of standing variation, if mutation rates are low. The case of a panmictic population was studied by [ 8 ], and we study the case of a continuous, spatial population in [ 54 ].
If parallelism in local adaptation of the sort we study here is due to standing variation rather than new mutation, then the dynamics of adaptation should not depend strongly on migration patterns but the initial spatial distribution of standing variation may. We have mostly ignored the issue of dominance by dealing with essentially haploid models, and appealing to the fact that the dynamics we study occur where the mutation is rare, and hence mostly present only in heterozygotes.
Our results should hold as a good approximation to dominant and partially dominant alleles with s m the selection against heterozygotes. If, however, the mutation is recessive, then it is essentially neutral where rare, and so would encounter much less resistance to spreading between patches.
The observation of morphological diversity not conforming to an inferred history of shared common ancestry is not unique to Hyalella Faria et al. It is likely that differentiation of lab stock occurred via drift or plasticity due to inevitable bottlenecks when establishing populations in captivity, as morphological divergence and local adaptation have been shown to occur rapidly in captivity Fragata et al.
However, differentiation was convergent toward the H. Only some of the populations were found to be interfertile and this did not strongly correlate with history of common ancestry or morphological similarity Figure 4.
The three interfertile populations were interfertile with each other in all possible combinations but never produced offspring with either of the reproductively isolated populations. The two reproductively isolated populations were shown to be completely reproductively isolated from all three of the interfertile populations as well as from each other.
At this time, the mechanism of reproductive isolation is unknown although amplexus was observed in all combinations, suggesting that the mode of reproductive isolation is gametic or postzygotic for the completely reproductively isolated populations, or at least not entirely behavioral. However, all of the heterospecific mating trials showed evidence of reproductive isolation, including interfertile combinations Figure 5.
This finding demonstrates viable hybridization between morphologically distinct populations and presents evidence of behavioral prezygotic reproductive isolation between populations of what were formerly considered populations of the H.
Therefore, divergence between populations likely occurred in the absence of gene flow; thus, sympatrically occurring populations likely represent secondary contact. It is unclear if divergence occurred directionally due to selection or drift during periods of geographic isolation.
However, molecular distance did not predict morphology or reproductive isolation, but geographic range size was found to be negatively correlated with interfertility Figure 6 , Table 7. It is remarkable that Hyalella was recovered as a monophyletic taxon as the depth of divergence between different Hyalella lineages is comparable to the depth of divergence observed among the other amphipod families included in our analysis depth of divergence between outgroups in Figure 1 is comparable to the divergence found within Hyalella.
We also only used a single mitochondrial locus because of the abundance of archived COI sequences for amphipods; however, the rate of divergence may be too rapid at the COI locus to properly estimate relationships with such deep divergence Figure 2. It is possible that divergence between lineages is approaching saturation which appears to have occurred around 0. However, separate analyses looking at each codon position revealed that first and second codon positions account for observed saturation of the COI locus while the third position conforms to the expectations of neutral evolution Figure S2.
A comparison of the amount of observed pairwise substitutions indicates that the third codon position is evolving approximately 2. It is important to point out that the present study recovered fewer haplotypes than previous authors despite sequencing the same locus and using the same sequences published by other authors on GenBank. This is likely due to the trimming of sequences to much fewer base pairs in order to have a complete alignment as different authors amplified different regions of the COI locus.
Therefore, it is likely that variable sites were eliminated that other authors used to identify haplotypes. Presumably, there is strong stabilizing selection within a population to maintain interfertility with other members of the same population. Therefore, it is more likely that genomic changes that lead to barriers to interfertility will be retained in smaller populations.
It is also less likely for larger populations to diverge from the reproductive type of ancestral populations if they experience stabilizing selection for interfertility due to lower susceptibility to drift in larger populations. Therefore, abundant and widespread taxa experiencing stabilizing selection may maintain interfertility with many different lineages, especially other widespread taxa, while local endemics experience drift or divergent selection.
The experimental observations presented herein are consistent with this hypothesis, but it requires further investigation. Identifying divergent loci associated with reproductive isolation could shed light on the factors that contribute to the evolution of reproductive isolation. Worsham involved in research design, field collections, laboratory experiments, molecular phylogeny, statistical analysis, manuscript preparation, and maps.
Julius involved in field collections and laboratory experiments. Nice involved in molecular phylogeny, statistical analysis, and manuscript preparation. Diaz involved in multivariate statistics and maps. Huffman involved in field collections, material support and manuscript preparation.
We are deeply grateful to Gary Wellborn for his invaluable contributions. Without the numerous volunteers that assisted with field collections, especially Helen Wukasch, Stephen Harding, Alex Zalmat, and Duy Le, this project would not have been possible.
We would like to thank Mackenzie Barnett for providing access to one of the sampling locations making this research possible. Fish and Wildlife Service. Geographic isolation facilitates the evolution of reproductive isolation and morphological divergence.
Ecol Evol. National Center for Biotechnology Information , U. Journal List Ecol Evol v. Published online Oct McLean L. Worsham , 1 , 2 Eric P. Julius , 1 Chris C.
Nice , 1 Peter H. Diaz , 3 and David G. Huffman 1. Eric P. Chris C. Peter H. Diaz 3 U. David G. Author information Article notes Copyright and License information Disclaimer. Worsham, Email: ude. Corresponding author. Email: ude. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.
This article has been cited by other articles in PMC. Associated Data Supplementary Materials. Abstract Geographic isolation is known to contribute to divergent evolution, resulting in unique phenotypes. Keywords: evolution, geographic isolation, molecular diversity, morphological diversity, reproductive isolation.
Table 1 Collection localities and count of dorsal mucronation for each Hyalella population. Open in a separate window. Establishment of stock cultures Stock cultures of Hyalella were established to provide a continuous source of live animals for experimentation and to control for the possibility of morphological differentiation due to phenotypic plasticity in situ. Molecular methods A molecular phylogeny based on the mitochondrial cytochrome C oxidase subunit I COI locus was constructed in order to analyze the relationship between morphological similarity, geographic factors, and a history of shared common ancestry.
Table 2 Replication of male—female pair combinations of Hyalella by population source and sex. Correlates of reproductive isolation To evaluate potential factors that might explain the occurrence of reproductive isolation, the results from the reproductive isolation experiment were arranged into a matrix. Table 3 Factors analyzed in ANOVAs to determine if geography can account for variation in the occurrence of reproductive isolation.
Phylogenetics Pairwise comparison of Hyalella sequences yielded 97 unique Hyalella haplotypes; three of the populations we sequenced had only one haplotype while the other two had two haplotypes each Table S5. Figure 1. Divergence events in Corylus were generally found to be older in Yang et al. Another source of error in their time-calibration may have come from their use of normally distributed priors on node age.
This approach that is typically reserved for secondary calibrations or geological events 37 and may lead to less accurate node ages if used for fossil calibrations as in Yang et al. Identifying and characterising the close relatives of the C. Corylus maxima is thought by some to belong to the species C.
However, estimated divergence times were older than one might expect within a single species: node age was 1. Confidence intervals suggest that divergence may have occurred as recently as 0. It may be that C. Even when genomic data is used the placement of C. The relationships in this group would be worth clarifying to shed more light on the closest relatives of the European hazelnut.
Previous studies 10 , 11 found C. It has been suggested that Corylus avellana could be crossed with C. Little is known about the thunbergii variety so further study could investigate whether the variety shares these traits with C. Corylus jacquemontii diverged from C. It would be valuable to investigate whether these two species can cross, as this could be used to introgress traits possessed by C.
The non-monophyly of species like C. The variety C. We suggest that results in this study and others 12 could be used as a platform for a taxonomic revision of the genus Corylus , particularly given the unclear boundary between variety and species and the wide range in number of species currently recognised by different authorities 6 , 7 , 8.
We found that the addition of the jump dispersal parameter improved the likelihood of both models we tested. Several other studies using BioGeoBEARS have found evidence for the widespread importance of jump dispersal in ancestral biogeographic models in animals 39 , 40 and plants 28 including temperate angiosperm lineages 41 , 42 , It is worth noting that the frequency of types of biogeographic events heavily depends on the size of the areas chosen e.
Indeed, our results using this model include improbable long distance dispersal events from China to North America and back again in rapid succession Fig. Consequently, we also examined results from models that did not include the jump dispersal parameter Figs 3a,c , 4 , which appear more realistic in most cases.
Much of the past diversification in hazel takes place in China and perhaps as a consequence of this, many dispersal events originate there Figs 3 , 4. We also see an important role for Eastern Russia Fig. All models highlight an important role for east Asia in the diversification of the genus, particularly within China.
The colonisations of North America by the ancestors of C. We estimated this repeated dispersal regardless of the model used with our nrDNA tree, indicating it is a reliable set of events and not model-dependent. The resulting disjunction has previously been inferred in plants 44 , 45 as well as animals As in this study, Whitcher and Wen 10 found that C.
Colonisation via the Bering land bridge is thought to contribute to the widespread pattern of disjunct distributions of temperate forest groups between North America and East Asia Our second question was: how important have different types of biogeographic events been in the history of Corylus?
To answer this we assessed the frequency of different types of biogeographic event. Range expansion, within-area speciation and founder-event speciation were generally the most common across our analyses Table 3 suggesting that these processes were important in the history of Corylus.
For example, range expansion and founder event speciation were more frequent in the nrDNA models and within-area speciation was more common in the cpDNA models Table 3.
Our cpDNA tree infers close relationships between geographically proximate species such as C. Again, the proportion of events inferred using models with jump dispersal must be treated with caution, though there appears to be scope for this process being important in Corylus. Our biogeographic models estimated several long distance dispersal events — thousands of kilometres over land. These lineages almost certainly passed through the regions between these areas but the history of dispersal across central Asia and the Middle East is unclear.
Few species currently inhabit these regions — C. The rarity of species records within these regions means that including the area in a model would likely not provide more resolution at the cost of increased model complexity and computation time. Further work could aim to identify the role of Central Asia and the Middle East in the biogeographic history of Corylus. Hazelnuts can be dispersed by both mammals and birds 52 , which could have helped species travel over these long distances.
Bird species have been known to disperse hazelnuts up to six kilometres 53 and mammals such as Sciurus carolinensis can store nuts for winter across more than a hectare of land around the plant 54 , which may have facilitated past dispersal. Dispersal across water bodies is also a possibility as hazelnuts can float and survive up to 90 days This characteristic may have facilitated dispersal across water bodies such as the Korea Strait, which lies between Japan and Korea and likely had to be crossed by species found in Japan today such as C.
Our ancestral range estimations consider only extant taxa and future work should include information from the fossil record to expand upon these models when fossils from the Betulaceae can be placed accurately. Further work incorporating these fossils and their morphology into the Corylus phylogenetic tree will develop our knowledge of the historical biogeography of the genus.
The two North American hazel species, C. Their ranges overlap extensively, with C. This proximity may have allowed interspecific hybridisation between the two species, leading to chloroplast capture 55 , where the chloroplast genome of one species is entirely replaced with that of another after introgression. Further work with high-throughput sequence data could shed light on history of gene flow between C. Experiments have shown that these species do not readily hybridise in experimental crosses 38 but this does not exclude the possibility that they were less reproductively isolated in the past.
Corylus ferox var. Due to the prospect of introgression and differences to the nrDNA tree, we suggest that the cpDNA tree represents the history of the chloroplast in Corylus rather than the species in the genus.
Observed non-monophyly of some species may also be the result of past hybridisation. Other potential sources of differentiation between cpDNA and nDNA may be different evolutionary rates among these two marker types 59 and incomplete lineage sorting The latter may be more prominent in recently diverged taxa or trees inferred with slowly evolving markers.
Further work is needed to assess the degree to which chloroplast capture has occurred, and then use this information to determine which taxa are true species and which are varieties. Our final question was: did tree form evolve once in Corylus or are different tree species the result of convergent evolution? Our models showed multiple instances of convergent evolution in growth form across the genus.
Even though many species of Carpinus and Ostrya are trees, our reconstructions both with and without outgroups estimate that shrub form is the most likely ancestral state of the Corylus genus. Taking this into account, tree form probably arose on four to five separate occasions across the genus. Therefore, convergent evolution as well as long-distance dispersal appears to have led to the geographic distribution of Corylus trees. There are many reasons why tree form may have evolved 61 and future work identifying shared ecological characteristics or constraints of the widespread tree species may shed light on the drivers behind the inferred convergence.
Our new phylogenetic trees resolve previously ambiguous relationships in the genus while adding additional taxa. Fossil calibrations allow the estimation of a detailed temporal scale of diversification in hazel, with most divergence events taking place in the Miocene. Our biogeographic models suggest that long distance dispersal was an important process in the history of the Corylus genus and instrumental in generating the diversity observed today, especially outside of east Asia.
We estimate that multiple cases of convergent evolution of tree form have occurred in hazel species, mostly in China.
Our study provides a platform for further investigations into potential introgression among geographically connected species, taxonomic revision of current varieties and the use of crop wild relatives to improve hazelnut production. Matzke, N. Ree, R. Maximum likelihood inference of geographic range evolution by dispersal, local extinction, and cladogenesis.
Ronquist, F. Dispersal-Vicariance Analysis: A new approach to the quantification of historical biogeography. Biol 46 , — Article Google Scholar. Paulay, G. Diversification in the tropical pacific: comparisons between marine and terrestrial systems and the importance of founder speciation.
Templeton, A. The reality and importance of founder speciation in evolution. BioEssays 30 , — Molnar, T. In: Wild crop relatives: Genomic and breeding resources forest trees ed. Kole, C. World Checklist of Selected Plant Families. Missouri Botanical Garden, St. Crop Evol. Whitcher, I. Google Scholar. Yang, Z. Resolving the speciation patterns and evolutionary history of the intercontinental disjunct genus Corylus Betulaceae using genome-wide SNPs.
Plant Sci. Pigg, K. Manchester, S. Palaeocarpinus dakotensis sp. Betulaceae: Coryloideae and associated staminate catkins, pollen, and leaves from the Paleocene of North Dakota. Crane, P. Early fossil history and evolution of the Betulaceae. Wehr, W. Early Tertiary flowers, fruits, and seeds of Washington State and adjacent areas.
0コメント