hung et al 2012

Molecular Phylogenetics and Evolution xxx (2012) xxx–xxx

Contents lists available at SciVerse ScienceDirect

Molecular Phylogenetics and Evolution

journal homepage: www.elsevier .com/ locate /ympev

Recent allopatric divergence and niche evolution in a widespread Palearctic bird,the common rosefinch (Carpodacus erythrinus)

Chih-Ming Hung a,b,c,⇑, Sergei V. Drovetski d, Robert M. Zink a,b

a Bell Museum, University of Minnesota, St. Paul, MN 55108, USAb Department of Ecology, Evolution and Behavior, University of Minnesota, St. Paul, MN 55108, USAc Department of Life Science, National Taiwan Normal University, Taipei 116, Taiwand Tromsø University Museum, NO-9037 Tromsø, Norway

a r t i c l e i n f o a b s t r a c t

Article history:Received 3 December 2011Revised 8 September 2012Accepted 14 September 2012Available online xxxx

Keywords:CoalescenceSimulationSpecies distribution modelNiche divergencePhylogeography

1055-7903/$ - see front matter � 2012 Elsevier Inc. Ahttp://dx.doi.org/10.1016/j.ympev.2012.09.012

⇑ Corresponding author at: Department of Life ScieUniversity, Taipei 116, Taiwan. Fax: +1 612 624 6777

E-mail address: [email protected] (C.-M. Hung)

Please cite this article in press as: Hung, C.-M.,finch (Carpodacus erythrinus). Mol. Phylogenet.

A previously published phylogeographic analysis of mtDNA sequences from the widespread Palearcticcommon rosefinch (Carpodacus erythrinus) suggested the existence of three recently diverged groups, cor-responding to the Caucasus, central-western Eurasia, and northeastern Eurasia. We re-evaluated themtDNA data using coalescence methods and added sequence data from a sex-linked gene. The mtDNAgene tree and SAMOVA supported the distinctiveness of the Caucasian group but not the other twogroups. However, UPGMA clustering of mtDNA UST-values among populations recovered the threegroups. The sex-linked gene tree recovered no phylogeographic signal, which was attributed to recentdivergence and insufficient time for sorting of alleles. Overall, coalescence methods indicated a lack ofgene flow among the three groups, and population expansion in the central-western and northeasternEurasia groups. These three groups corresponded to named subspecies, further supporting their validity.A species distribution model revealed potential refugia at the Last Glacial Maximum. These three groups,which we hypothesized are in the early stages of speciation, provided an opportunity for testing tenets ofecological speciation. We showed that the early stages of speciation were not accompanied by ecologicalniche divergence, consistent with other avian studies.

� 2012 Elsevier Inc. All rights reserved.

1. Introduction

Quaternary climatic fluctuations repeatedly shifted the distri-butions of species, presumably resulting in multiple cycles ofretraction into refugia during glacial advance and subsequentexpansion during interglacial periods (Hewitt, 2000, 2004). Thesefluctuations in range could have had at least three consequences:(1) initiation of divergence among populations, (2) continuedmaintenance of already distinct populations, or (3) merging ofonce-differentiated populations. Phylogeographic studies can pro-vide a description of the current geographic deployment of geneticvariation and yield inferences about which of these potential histo-ries was experienced by a particular species (Avise, 2000). Forexample, the discovery of three phylogroups in an extant speciesimplies the existence of at least three glacial refugia at the Last Gla-cial Maximum (LGM; 21,000 years ago), assuming that the groupsevolved prior to the last glacial period. A species distribution model(SDM), often referred to as an ecological niche model (Petersonet al., 1999), can be used to determine the sizes and locations of

ll rights reserved.

nce, National Taiwan Normal..

et al. Recent allopatric divergenEvol. (2012), http://dx.doi.org/1

LGM refugia and then relate them to the current population struc-ture (Carnaval et al., 2009; Knowles and Alvarado-Serrano, 2010).One could also discover whether these intraspecific groups havediverged ecologically, providing a test of the tenets of ecologicalspeciation (Schluter, 2009; McCormack et al., 2010). Petersonet al. (1999) suggested that ecological divergence only occurredafter speciation; however, there are relatively few tests of nichedivergence early in the speciation process.

The common rosefinch (Carpodacus erythrinus) inhabits varioustypes of woodlands and grasslands, spanning much of the Palearc-tic. During the past century, its range has expanded westward intoScandinavia, France, Spain and England (Cramp and Perrins, 1994;Payevsky, 2008), although persistent breeding is limited to Scandi-navia. Pavlova et al. (2005) surveyed mtDNA variation and postu-lated the existence of three taxa corresponding to the Caucasus(CA), northeastern Eurasia (NEE), and central-western Eurasia(CWE, Fig. 1). However, the geographic limits of the groups wereunclear and maximum likelihood estimates of migration suggesteda low level of gene flow among groups. In this study, we re-examined the conclusion of Pavlova et al. (2005) by (1) usingrobust coalescence methods and adding a Z-linked gene, (2) usinga SDM to reconstruct the current and LGM distributions of thecommon rosefinch, to search for sites of potential refugia, and to

ce and niche evolution in a widespread Palearctic bird, the common rose-0.1016/j.ympev.2012.09.012

http://dx.doi.org/10.1016/j.ympev.2012.09.012

mailto:[email protected]


http://www.sciencedirect.com/science/journal/10557903

http://www.elsevier.com/locate/ympev


DOR TÖV

TYV

GOR IRK KRY YEK

MEZ MED MOS

KUR

KRD SAK KHA

MAG ANA

KAM ALM

C. e. grebnitskii

C. e. erythrinus

C. e. kubanensis C. e. ferghanensis

Fig. 1. Breeding range of the common rosefinch (Carpodacus erythrinus) with four subspecies indicated and location of sample sites (re-drawn from Fig. 1 in Pavlova et al.(2005). Gray area indicates breeding range. Large solid circles indicate sample sizes of 10 or more individuals. Dotted lines partition the population into three groups asindicated by the pairwise mtDNA UST analysis (Pavlova et al., 2005) and the UPGMA tree based on a matrix of mtDNA UST values (Fig. 4 in this study). ALM is Almaty, ANA isAnadyr’, DOR is Dornod, GOR is Gorno-Altay, IRK is Irkutsk, KAM is Kamchatka, KHA is Khabarovsk, KRD is Krasnodar, KRY is Krasnoyarsk, KUR is Kursk, MAG is Magadan, MEDis Medvedevo, MEZ Mezen’, MOS is Moscow, SAK is Sakhalin, TÖV is Töv, TYV is Tyva, and YEK is Yekaterinburg.

2 C.-M. Hung et al. / Molecular Phylogenetics and Evolution xxx (2012) xxx–xxx

assess the relative sizes of populations at each time, and (3) testingfor ecological divergence between these groups to determine ifniche differentiation has accompanied the early stages of lineagedivergence (Warren et al., 2008).

2. Methods and materials

2.1. Genetic analyses

We used the ND2 sequence data for 186 common rosefinchesfrom Pavlova et al.’s (2005) study (Fig. 1 and Table S1, NCBI acces-sion numbers AY703261-AY703446). In addition, one Z-linked in-tron, ADAMTS6 (Backström et al., 2006), was sequenced for 28individuals from 15 localities (Table S1; 12, 7, and 9 individualsfrom the three mtDNA-defined groups, CWE, NEE and CA, respec-tively). Phases of sequences containing indels were sorted manuallyby subtracting chromatogram peaks upstream of the indel in the re-verse primer sequences from the double peaks downstream of theindel in the forward primer sequences. This procedure was re-peated in the alternative direction and allowed the two alleles ofa heterozygous individual to be determined (Sousa-Santos et al.,2005; Dolman and Moritz, 2006). Alleles present in individuals withmultiple heterozygous sites but no indel(s) were resolved usingPHASE 2.1.1 (Stephens et al., 2001; Stephens and Scheet, 2005).

2.2. Detection of population division

NETWORK 4.5.1.6 (fluxus-engineering.com) was used to gener-ate minimum spanning networks (Polzin and Daneschmand, 2003)for mtDNA and ADAMTS6 data to detect geographic partitioning ofhaplotypes and alleles.

Pairwise UST-values of mtDNA (but not ADAMTS6 due to smallsample sizes) among populations were estimated using DnaSP 5(Librado and Rozas, 2009). To determine the overall geographicpattern of population structure, we pooled individuals from eachlocality and then clustered the matrix of pairwise UST values usingthe unweighted pair-group method using arithmetic averages (UP-GMA) implemented in MEGA 4 (Tamura et al., 2007). We combinedthe populations with sample sizes smaller than five with samplesin the nearest and sufficiently large population to avoid bias inUST estimates. If the combination was not geographically reason-able, the samples were excluded from analyses. For example, theSakhalin population was not combined with the Kamchatka orMagadan population because they are separated by a wide areaof sea or a long geographic distance.

Please cite this article in press as: Hung, C.-M., et al. Recent allopatric divergenfinch (Carpodacus erythrinus). Mol. Phylogenet. Evol. (2012), http://dx.doi.org/1

We also applied spatial analysis of molecular variance imple-mented in SAMOVA 1.0 (Dupanloup et al., 2002) to the mtDNA datato explore population division. Based on an F-statistic, SAMOVAuses a simulated annealing procedure to determine groups of pop-ulations that are geographically and genetically homogeneous andmaximally differentiated from each other (i.e., the configurationhas the largest UCT). We tested population partitions (K) rangingfrom 2 to 8 with 100 to 300 initial conditions each based on 105

annealing permutations. Given that populations with small samplesizes might be erroneously identified as isolated groups in SAM-OVA (Dupanloup, pers. comm.), we combined the localities withsmall sample sizes according to the same rules we used for calcu-lating the pairwise UST-values.

2.3. Estimates of demographic parameters

A coalescence-based program, IMa (Hey and Nielsen, 2007), wasused to estimate the effective population sizes of two current pop-ulations and their common ancestral populations (h1, h2, and ha;h = 4 Nel), migration rates (m1 and m2; m = m/l), and divergencetimes (t = tl) between groups of populations determined fromthe procedures outlined above. The UPGMA phenogram based onpairwise UST values and SAMOVA suggested two possible groupstructures, a three-group model and a two-group model (see Sec-tion 3.1). The three-group model included CA, CWE, and NEE. Thetwo-group model consisted of a combined CWE and NEE (termedME) and CA. We estimated the demographic parameters associatedwith the various groupings in the two alternative models. We didnot use IMa2 (Hey, 2010) because we cannot specify a tree topologyfor the three groups. A minimum of two independent analyses of>2 � 107 steps after a burn-in of 106 steps were performed for eachpairwise comparison. Plots of trend lines and the effective samplesize values (ESS > 250) were examined to assess convergence inparameter estimates. The IMa estimates were based on all availablemtDNA and ADAMTS6 sequence data combined. To convert thescaled demographic parameters to absolute values, we calculatedthe geometric mean of the substitution rates of ND2 and ADAMTS6by multiplying the sequence lengths by 4 � 10�8 substitutions/site/year for ND2 (Arbogast et al., 2006) and 1.62 � 10�9 substitutions/site/year for ADAMTS6 (Ellegren, 2007; given the mean divergenceof Z-linked introns between chicken and turkey is 1.2 times higherthan that of autosomal introns, which have an average rate of1.35 � 10�9 substitutions/site/year) and assumed a generation timeof 2 years (Stjernberg, 1979). The substitution rate used in thisstudy was higher than widely used rates for avian mtDNA (Lovette,



C.-M. Hung et al. / Molecular Phylogenetics and Evolution xxx (2012) xxx–xxx 3

2004; Weir and Schluter, 2008) but more appropriate for recentdivergences (Ho et al., 2005; Arbogast et al., 2006; Herman andSearle, 2011; but see Emerson 2006).

2.4. Group delimitation using hypothesis testing

We tested alternative population models using a coalescentsimulation program, SIMCOAL 2.1.2 (Laval and Excoffier, 2004) tosimulate ND2 sequences. The simulations were performed usingthe demographic parameters estimated from IMa based on mtDNAand ADAMTS6, i.e., current effective population sizes (the effectivepopulation size of mtDNA, Nef, was assumed to be half of the Ne ofpopulations), migration rates, divergence times, exponentialgrowth rates (calculated from the current and ancestral populationsizes), a substitution rate of 4 � 10�8 substitutions/site/year, andthe generation time of two years.

One two-group model and two three-group models were simu-lated (Fig. 2). In the latter case, two models were considered due tothe uncertainty in gene flow from NEE to CWE based on the IMaanalyses (see Section 3.1). One three-group model postulated nogene flow (Fig. 2a) and the other assumed a low level of gene flow(2 � 10�6 per generation per individual; see Section 3.1) from NEEto CWE (Fig. 2b). Given that the dispersal distance of rosefinches isunlikely to span the entire geographic range, the widely distrib-uted ME (in the two-group model; Fig. 2c) was not assumed tobe panmictic. We simulated ME by connecting two subgroups witha substantial level of migration, 100 individuals per generation atpresent but decreasing exponentially backwards in time as thepopulation size decreased (Fig. 2c). This level of migration is largeenough to prevent population differentiation (Hudson and Coyne,2002). The ratio of Ne for the two subgroups in ME was 12:1(Fig. 2c) similar to that for CWE and NEE.

We simulated 5000 datasets of 186 ND2 sequences (1041 bp)for each hypothesized model. The simulated datasets were ana-lyzed using Arlequin 3.1 (Schneider et al., 2000) to calculate UST

values (Excoffier et al., 1992) and the distribution and the upperand lower bounds of 95% ranges of these values were recorded.The UST values for simulated data were compared with the empir-ical UST to evaluate which model was most consistent with the ob-served data.

2.5. Species distribution modeling

Online databases ORNIS and GBFI together with our own mu-seum data resulted in 196 breeding localities (see Fig. S1), whichwe input into MAXENT v 3.2.2. (Phillips et al., 2006) to infer aSDM. Because the species has only colonized western Europe dur-ing the past few decades (Cramp and Perrins, 1994), we excluded

NCA = 83,000

NCWE = 3,516,000

NNEE = 308,000

Na = 63,000 t= 39,200

(a) (b)

m= 2 × 1

NCA = 83,000

NCWE =3,516,0

t= 39,200

Fig. 2. Outline of two three-group models (a and b) and one two-group model (c) develoNCWE, NNEE and NME indicate effective population sizes of ancestral, CA, CWE, NEE and Mpresent. Arrows indicate gene flow. The immigration rate (m) in the (b) model is shownwith substantial levels of gene flow.


breeding records from Scandinavia, France, Spain and England.We obtained climatic data (19 layers) for present and LGM condi-tions from the Worldclim bioclimatic database (Hijmans et al.,2005). We ran MAXENT five times (using 30% of values for training)and examined the jackknifed contributions of each bioclimaticlayer, picking seven layers (bioclim layers 1, 5, 9, 10, 11, 15 and18) that contributed 5% or more to the total variation for the finalanalysis (Brown and Knowles, 2012). For the final model, we con-structed a SDM from the average of 10 MAXENT runs using alllocality points, which we displayed using DIVA-GIS ver. 7.1.7.2(Hijmans et al., 2004).

2.6. Niche divergence tests

We used the program ENMTools v.1.3 (Warren et al., 2010) toperform niche identity tests and background tests for whether taxathat were defined according to the genetic data (i.e., the CA, CWEand NEE groups) exhibit niche divergence or conservatism (or nei-ther). The niche identity test assesses whether the similarity be-tween ecological niches of two taxa is significantly different froma null distribution generated from random draws of pooled empir-ical occurrence points. The identity test often suggests that two taxahave significantly different niches (Godsoe, 2010; Warren et al.,2008). However, this inference is only true when the two groupstolerate the exact same set of environmental conditions and havethe same suite of environmental conditions available to them(Warren et al., 2010). Because this is unlikely for allopatric popula-tions that are distributed across environmental gradients, the back-ground test determines whether two taxa are more or less similarthan expected based on the differences in the environmental back-ground where they occur. The background test determines whetheran empirical niche similarity index between two taxa is signifi-cantly smaller (niche divergence) or greater (niche conservatism)than a null distribution of the niche similarity indices betweenone taxon and random points from the environment within theother taxon’s range; two null distributions result because the com-parisons are done in both directions (Warren et al., 2010). For thebackground tests, we divided the large central area of rosefinch dis-tribution into eastern (CWEE) and western (CWEW) groups to ac-count further for the large geographic distance involved.

3. Results

3.1. Population structure

The mtDNA network revealed that 9 of the 13 haplotypes fromthe Caucasus (CA) formed a clade, which was separated by a fewsteps from the remaining haplotypes (Fig. 3a). The haplotypes in

NCA = 72,000

NME = 3,990,000 + 333,000

Na = 72,000 t =33,200

(c)

0-6

00

NNEE = 308,000

Na = 63,000

ped to test hypothesized demographic histories for the common rosefinch. Na, NCA,E groups, respectively. t indicates time of divergence in units of generations from

in the figure. The dashed lines indicate that the groups are connected to each other



Fig. 3. Network of (a) mtDNA and (b) a Z-linked intron (ADAMTS6) haplotypes of the common rosefinch by the minimum spaning criterion. Colors on the network indicatethe locations at which each haplotype was detected. Black circles indicate central-western Eurasian (CWE) haplotypes, white circles indicate northeastern Eurasian (NEE)ones, and gray circles indicate Caucasian (CA) ones. Open triangles indicate unsampled or extinct haplotypes. Sizes of circles are proportional to haplotype frequencies. Shortlines indicate one mutation step.

Fig. 4. UPGMA phenogram showing relationships among rosefinch populationsamples. The tree is reconstructed based on the matrix of UST values. The threeclades correspond to the central-western Eurasian (CWE), northeastern Eurasian(NEE), and Caucasian (CA) groups. Populations with sample sizes smaller than fivewere combined with samples in the nearest and sufficiently large population.Therefore, DOR + TÖV is the combination of Dornod and Töv’s samples, IRK + KRY isthat of Irkutsk and Krasnoyarsk, MED + MEZ is that of Medvedevo and Mezen, andMOS + KUR is that of Moscow and Kursk.


northeastern Eurasia (NEE) were not separated from those in cen-tral-western Eurasia (CWE). Thus, the mtDNA network revealedlimited geographic structure. The allele network for ADAMTS6showed no geographic structure (Fig. 3b). The UPGMA phenogrambased on the matrix of mtDNA pairwise UST values (see Table S2for the UST values) clustered rosefinch populations into three dis-tinct groups corresponding to CA, NEE, and CWE groups (Fig. 4).The SAMOVA supported a two-group structure, CA and all remain-ing populations (i.e., CWE plus NEE). The UCT value was highestwhen K = 2 (UCT = 0.349) and decreased as K increased.

IMa analyses based on mtDNA, and mtDNA plus ADAMTS6, sug-gested recent isolation with little or no gene flow followed by pop-ulation expansion (except for CA), for both the three-group andtwo-group models. In addition, the IMa analyses revealed that thevalues of demographic parameters for the two datasets differed inmagnitude, with the combined data (i.e., mtDNA plus ADAMTS6)suggesting longer divergence times, greater effective populationsizes and lower migration rates (Table S3). We used results basedon the combined dataset in our simulations. For the three-groupmodel, IMa revealed that the current population sizes of CWE(7,032,000; 95% credibility interval [CI] = 4,130,000 to 17,303,000)and NEE (616,000; 95% CI = 334,000–1,839,000) are significantlylarger than that of the ancestral population (136,000; 95%CI = 58,000–240,000), whereas the current population size of CA


(167,000; 95% CI = 80,000–438,000) was not significantly largerthan that of the ancestral population (Fig. 5a and b). There was littleif any migration between any pair of groups (Fig. 5c–e). There could



Fig. 5. The marginal posterior probabilty distribution of the IMa parameters under the three-group model. The three groups are Caucasus (CA), central-western Eurasia (CWE)and northeastern Eurasia (NEE). The IMa analyses were based on mtNDA and a Z-linked gene (ADMATS6) combined. Effective population sizes, h, for CA, CWE, NEE and theirancestral populations are shown in (a and b). Migration rates, m, in two directions between CA and CWE, between CWE and NEE, and between CA and NEE in the unite of pergeneration per individual are shown in (c–e). Divergence times, t, for the three pairwise comparisons in the unite of year are shown in (f).


be some migration from NEE to CWE (2 � 10�6 per generation perindividual; 95% CI = �0 to 1.6 � 10�5 [we could not reject the pos-sibility of zero gene flow]; Fig. 5d). The divergence times betweenCA and CWE (78,300 years ago; 95% CI = 52,200–117,000 yearsago), between CWE and NEE (78,800 years ago; 95% CI = 52,000–117,200 years ago), and between CA and NEE (76,200 years ago;95% CI = 42,200–141,700 years ago) were similar and recent(Fig. 5f), suggesting contemporaneous origins of the three groups(see Hung et al. 2012). The IMa results for the two-group model re-vealed that the current population size of ME (8,645,000; 95%CI = 5,396,000–18,196,000) was significantly larger than the ances-tral population (147,000; 95% CI = 88,000–282,000), but that of CA(144,000; 95% CI = 70,000–394,000) was not. The IMa resultsshowed no migration between the two groups. The estimateddivergence time (66,400 years ago; 95% CI = 45,500–96,600 yearsago) was similar to that of the three-group model.

Comparison of the simulated and empirical data supported thethree-group models over the two-group model (Fig. 6). The empir-ical UST value was within the 95% distribution range of UST valuesof the three-group models regardless of the level of gene flow


(Fig. 6a and b) but out of the 95% range of UST values for thetwo-group (Fig. 6c). Therefore, we could reject the two-group butnot either of the three-group models. The similarity between thedistributions of UST values for the two three-group models, onewith and one without gene flow, suggested that the level of geneflow from NEE to CWE was negligible.

3.2. Current and LGM distributions predicted by SDM

The AUC of the MAXENT distribution model was 0.88, indicatinggood quality in the predictions (AUC > 0.7 = informative; Baldwin,2009; Swets, 1988). The SDM under current climatic conditionspredicted the breeding range of common rosefinches includingnewly colonized areas in Scandinavia (Fig. 7a). Although the modelpredicts occurrence in France, Portugal and Spain the species hasnot yet become established in these areas, but there are sporadicrecords (Cramp and Perrins, 1994). Therefore, the predicted cur-rent distribution of common rosefinches in Europe was somewhatwider than the observed distribution but consistent with the trend



Fig. 6. Histograms of UST values for 5000 simulated sequence datasets. (a) Results from simulations of the three-group model without gene flow. (b) Results from simulationof the three-group model with a low level of gene flow from NEE to CWE. (c) Results from simulations of the two-group model. The dashed lines represent the 2.5 and 97.5percentiles in the distribution. The black triangles represent the UST value of our empirical data.


of recent expansion towards the western and northwestern Pale-arctic (Cramp and Perrins, 1994; Payevsky, 2008).

The predicted distribution at the LGM suggested that commonrosefinches mainly occurred in four areas (i.e., refugia): (1) Kam-chatka and Magadan, (2) southern Siberia, Mongolia, China and Ja-pan, (3) Caucasus and (4) southern Europe (Fig. 7b). The first threepredicted refugia sustainably or partially overlapped with the pres-ent distributions of the three haplotype groups. Comparing the ex-tent of the present and LGM distributions suggested that theCaucasian population (CA) and Kamchatka and Magadan popula-tions (corresponding to NEE) had not undergone expansionwhereas the other populations (corresponding to CWE) expandedwestwards since the LGM (Fig. 7).

3.3. Niche divergence

The pairwise niche identity tests (not shown) suggested that theniches of different groups of rosefinches were significantly differ-ent based on the niche similarity index, Schoener’s D (P < 0.001).However, we found no instances of either niche divergence or


niche conservatism in the eight pairwise background tests, whichtook into account the environment available to each group (Fig. 8).

4. Discussion

Our re-analyses of the mtDNA data, and analyses of the mtDNAdata plus the sex-linked sequences, support the existence of threegroups of populations in the common rosefinch, namely the Cauca-sus, northeastern Eurasia, and central-western Eurasia. These threegroups are not reciprocally monophyletic for either mtDNA or theZ-linked gene tree, suggesting that the rosefinches are at an earlystage of diversification. Our estimates of gene flow were essentiallyzero, indicating that the groups are evolving independently. Hencewe infer that speciation has commenced. In addition, the SDM im-plies that these three groups might have occupied distinct LGMrefugia. Further supporting our hypothesis of the distinctness ofthese groups, we note that the three groups, CA, NEE and CWE, cor-respond to named subspecies, C. e. kubanensis, C. e. grebnitskii andC. e. erythrinus, respectively (Cramp and Perrins, 1994; Fig. 1). Thesubspecies C. e. kubanenesis is larger and more rosy-red colored



Fig. 7. Predicted distributions of the common rosefinch (Carpodacus erythrinus) for (a) present and (b) LGM. Darker colors reflect areas with higher probabilities of occurrence.


than the nominate subspecies C. e. erythrinus, and C. e. grebnitskiihas more red or carmine on its upperparts than C. e. erythrinus(Clement et al., 1993; Cramp and Perrins, 1994). A fourth subspe-cies, C. e. ferghanensis, represented only by our Almaty samplewas not genetically distinct, and fell within C. e. erythrinus.Although we do not propose elevating these taxa to species, anyanalysis of recent trends in biodiversity should recognize their his-torical existence.

4.1. Recent history of common rosefinches revealed by phylogeographyand SDM

The application of SDMs to evolutionary studies is promising(Carnaval et al., 2009; Kozak et al., 2008) although their perfor-mance has been under debate (e.g., Araújo et al., 2009; Bealeet al., 2008). Our SDM under current climatic condition predictedthe occurrence of common rosefinches in western Europe eventhough our model did not incorporate breeding sites from this re-gion. The fact that the species has recently become common inScandinavia and has bred locally in other European areas (Crampand Perrins, 1994; Payevsky, 2008), and that the considerable partof the predicted range conforms to the known distribution, sug-gests that our SDM is relatively robust. However, our SDM under-estimated the distribution of this species in northeastern Eurasiaand overestimated that in China and Japan when compared withthe known distribution. The inconsistency may reflect limitations


of SDMs and/or the fact that our knowledge about the distributionof this species in these regions is incomplete.

In general, it is difficult to determine the specific geographiclocations of LGM refugia, irrespective of whether one uses a SDMor makes inferences from the extant phylogeographic pattern.We consider our SDM and genetic results to be reciprocally illumi-nating. Our SDM (Fig. 7) is a snapshot at the maximum of the lastglacial period, which began ca. 100,000 years before (Lisiecki andRaymo, 2007), and hence it is unclear how many refugia existedduring this period. Our genetic results suggested three recently(e.g., ca. 75,000 ybp) isolated groups (Figs. 4 and 6), which we thinkhelps guide interpretation of the SDM. Assuming that the threegroups originated prior to the LGM, we suggest that two refugia ex-isted in the eastern Palearctic at the LGM, even though our SDMshows a relatively narrow gap (i.e., the coast near the northwesterntip of the Sea of Okhotsk) between them (Fig. 7b). However, manyextant phylogroups are parapatrically distributed or separated bynarrow gaps (e.g., Hung et al., 2012; Zink et al., 2008). The lackof sorting of mtDNA haplotypes from the Palearctic (excludingCA; Fig. 3a) coupled with the inference of negligible gene flow be-tween NEE and CWE, is consistent with our estimate of divergencetime (ca. 75,000 ybp). Haploid, non-recombining mtDNA genetrees require isolation lasting approximately Ne generations toreach reciprocal monophyly (Hudson and Coyne 2002). Given thelarge estimates of Ne for the two main Palearctic groups(7,032,000 and 616,000), considerable time would have toelapse before the mtDNA gene tree would become reciprocally



Fig. 8. Background tests for niche divergence. Arrows indicate observed niche similarity index (Schoener’s D) between two groups and are compared with a null distributionfor the niche similarity indices between one group and random points from the range of the other group and vice versa. (a) Gray histogram indicates a null distribution of NEEversus the background of CWEE, and black indicates the opposite comparison. (b) Gray indicates a null distribution of CA versus the background of NEE, and black indicatesthe opposite comparison. (c) Gray indicates a null distribution of CA versus the background of CWEW, and black indicates the opposite comparison. (d) Gray indicates a nulldistribution of CWEW versus the background of CWEE, and black indicates the opposite comparison. Niche similarity values smaller than a null distribution indicate nichedivergence, larger values indicate niche conservatism, and values within a null distribution indicate neither niche divergence nor niche conservatism.


monophyletic. Therefore, there has been insufficient time for themtDNA gene tree to evolve to reciprocal monophyly. Given thatthe average Ne of Z-linked genes is three times greater than mtDNA,we expected that the ADSMTS6 gene tree would show less geo-graphic structure than the mtDNA gene tree, which we observed.

Combining coalescence and SDM analyses allowed additionalinsight into rosefinch population history. Our coalescence analysesrevealed that CWE and NEE experienced population expansion,whereas CA did not (Fig. 5). Our SDM analyses suggested a similarhistory except that NEE had similar distributions between thepresent and LGM predictions. However, the degree of expansionfor CWE was at least 10 times greater than that for NEE accordingto the coalescence analyses (Fig. 5), and from the SDM it would ap-pear logical given the large area into which CWE expanded, and therelatively stable distribution of NEE.

The predicted LGM occurrence of rosefinches in areas south ofwhere they presently occur in southern Europe is relevant to pasthypotheses about the distribution of genetic variation followingglacial retreat. Hewitt (2004) suggested that genetic variability de-creases in the direction of colonization from a refugium owing tosampling effects, termed ‘‘leading edge expansion.’’ However, ifthe refugium no longer has suitable habitat, as is the case for com-mon rosefinches in southern Europe, this prediction is not testable.In this case, one might look to the southern extent of the currentrange as being the closest to refugial populations, and hence cur-rently harboring the greatest genetic variability, but we lackeddata for a strong test of this hypothesis. Thus, a SDM can modifypredictions about post-glacial patterns of genetic variation.

The SDM is consistent with a history of isolation for rosefinch-es in the Caucasus (Fig. 7). Compared to CWE and NEE, CA has arelatively smaller effective population size and has not experi-enced population expansion. These attributes can explain why


approximately 70% of haplotypes in CA form a clade in themtDNA tree (Pavlova et al. 2005) and haplotype network(Fig. 3), whereas haplotypes in the other two groups do not. Thatis, lineage sorting is a function of Ne, and therefore one expectsfaster sorting in the CA group. We predict that if the isolationand current population size continues, the Caucasian populationwill be the first group to become a new species. This exampleillustrates that relatively small and isolated populations on is-lands, continental habitat islands, or mountaintops are potentgrounds for lineage diversification.

4.2. Niche evolution between the recently diverged common rosefinchgroups

As the three rosefinch groups are in the early stages of specia-tion, our study can help to show whether ecological divergencedrives speciation or accrues after speciation. We found no evidencefor niche divergence, a result consistent with that for other groupsof birds that are considerably more differentiated (Peterson et al.,1999; McCormack et al., 2010). The significant tests for niche iden-tity coupled with the insignificant background tests reinforce theimportance of considering environment availability when testingniche divergence (McCormack et al., 2010). In other words, the re-sults imply that rosefinches in different areas do not have the sameniche space available to them.

The lack of support for niche divergence or conservatism mayindicate that rosefinches can survive in varied habitats, as sug-gested by the significant niche identity tests. This may explainthe recent population expansion of rosefinches shown by geneticdata and current expansion to new ranges in Europe. On the otherhand, given the general lack of niche divergence, one might expectthat the groups would invade each other’s ranges. That is, during




the ensuing 21,000 years since the LGM, and a dispersal distancetypical for passerine birds of P1 km per generation (Barrowclough1980), there would have been ample time for range expansion andhomogenization of the three groups. However, our estimates ofgene flow suggest this is not occurring. Hence it is possible thatthe forms exclude each other behaviorally (Waters, 2011) or wehave not captured the most important niche dimensions.

Acknowledgments

We thank S. Rohwer for his roles in gathering specimens and A.Pavlova for doing the original analyses, and S. Birk for loans.K. Kozak and D. Shepard helped in conducting ecological nichemodeling. T. Rodrigues assisted with map making. B. McKay, H.Vazquez and two anonymous reviewers helped to improve themanuscript. We are grateful to the University of Minnesota Super-computing Institute for assistance with the computations. Supportcame from NSF (DEB 9707496, and DEB 0212832), FCT (PTDC/BIA-BEC/103435/2008) and the Dayton-Wilkie fund of the Bell Museum.

Supplementary material

Supplementary data associated with this article can be found, inthe online version, at http://dx.doi.org/10.1016/j.ympev.2012.09.012.

References

Araújo, M.B., Thuiller, W., Yoccoz, N., 2009. Reopening the climate envelope revealsmacroscale associations with climate in European birds. Proc. Natl. Acad. Sci.USA 106, E45–E46.

Arbogast, B.S., Drovetski, S.V., Curry, R.L., Boag, P.T., Seutin, G., Grant, P.R., Grant,B.R., Anderson, D.J., 2006. The origin and diversification of Galapagosmockingbirds. Evolution 60, 370–382.

Avise, J.C., 2000. Phylogeography: The History and Formation of Species. HarvardUniversity Press, Cambridge.

Backström, N., Brandström, M., Gustafsson, L., Qvarnström, A., Cheng, H., Ellegren,H., 2006. Genetic mapping in a natural population of collared flycatchers(Ficedula albicollis): conserved synteny but gene order rearrangements on theavian z chromosome. Genetics 17, 377–386.

Baldwin, R.A., 2009. Use of maximum entropy modeling in wildlife research.Entropy 11, 854–866.

Barrowclough, G.F., 1980. Genetic and phenotypic differentiation in a wood warbler(genus Dendroica) hybrid zone. Auk 97, 655–668.

Beale, C.M., Lennon, J.J., Gimona, A., 2008. Opening the climate envelope reveals nomacroscale associations with climate in European birds. Proc. Natl. Acad. Sci.USA 105, 14908–14912.

Brown, J.L., Knowles, L.L., 2012. Spatially explicit models of dynamic histories:examination of the genetic consequences of Pleistocene glaciation and recentclimate change on the American pika. Mol. Ecol. 21, 3757–3775.

Carnaval, A.C., Hickerson, M.J., Haddad, C.F.B., Rodrigues, M.T., Moritz, C., 2009.Stability predicts genetic diversity in the Brazilian Atlantic forest hotspot.Science 323, 785–789.

Clement, P., Harris, A., Davis, J., 1993. Finches and Sparrows: An IdentificationGuide. Princeton University Press, Princeton.

Cramp, S., Perrins, C.M., 1994. Handbook of the Birds of Europe, the Middle East andNorth Africa, vol. 5. Oxford University Press, Oxford.

Dolman, G., Moritz, C., 2006. A multilocus perspective on refugial isolation anddivergence in rainforest skinks (Carlia). Evolution 60, 573–582.

Dupanloup, I., Schneider, S., Excoffier, L., 2002. A simulated annealing approach todefine the genetic structure of populations. Mol. Ecol. 11, 2571–2581.

Ellegren, H., 2007. Molecular evolutionary genomics of birds. Cytogenet. GenomeRes. 117, 120–130.

Emerson, B.C., 2006. Alarm Bells for the molecular clock? No support for Ho et al’.smodel of time-dependent molecular rate estimates. Syst. Biol. 56, 337–345.

Excoffier, L., Smouse, P., Quattro, J., 1992. Analysis of molecular variance inferredfrom metric distances among DNA haplotypes: application to humanmitochondrial DNA restriction data. Genetics 131, 479–491.

Godsoe, W., 2010. Regional variation exaggerates ecological divergence in nichemodels. Syst. Biol. 59, 298–306.

Herman, J.S., Searle, J.B., 2011. Post-glacial partitioning of mitochondrial geneticvariation in the field vole. Proc. Roy. Soc. B 278, 3601–3607.

Hewitt, G.M., 2000. The genetic legacy of the quaternary. Nature 405, 907–913.Hewitt, G.M., 2004. Genetic consequences of climatic oscillations in the quaternary.

Philos. Trans. Roy. Soc. B 359, 183–195.


Hey, J., 2010. Isolation with migration models for more than two populations. Mol.Biol. Evol. 27, 905–920.

Hey, J., Nielsen, R., 2007. Integration within the Felsenstein equation for improvedMarkov chain Monte Carlo methods in population genetics. Proc. Natl. Acad. Sci.USA 104, 2785–2790.

Hijmans, R.J., Guarino, L., Bussink, C., Mathur, P., Cruz, M., Barrentes, I., Rojas, E.,2004. DIVA-GIS. Vsn. 7.1.7. A geographic information system for the analysis ofspecies distribution data. <http://www.diva-gis.org>.

Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G., Jarvis, A., 2005. Very highresolution interpolated climate surfaces for global land areas. Int. J. Climatol. 25,1965–1978.

Ho, S.Y.W., Phillips, M.J., Cooper, A., Drummond, A., 2005. Time dependency ofmolecular rate estimates and systematic overestimation of recent divergencetimes. Mol. Biol. Evol. 22, 1561–1568.

Hudson, R.R., Coyne, J.A., 2002. Mathematical consequences of the genealogicalspecies concept. Evolution 56, 1557–1565.

Hung, C.-M., Drovetski, S.V., Zink, R.M., 2012. Multilocus coalescence analysessupport a mtDNA-based phylogeographic history for a widespread Palearcticpasserine bird, Sitta europaea. Evolution 66, 2850–2864.

Knowles, L.L., Alvarado-Serrano, D.F., 2010. Exploring the population geneticconsequences of the colonization process with spatio-temporally explicitmodels: insights from coupled ecological, demographic and genetic models inmontane grasshoppers. Mol. Ecol. 19, 3727–3745.

Kozak, K.H., Graham, C.H., Wiens, J.J., 2008. Integrating GIS-based environmentaldata into evolutionary biology. Trends Ecol. Evol. 23, 141–148.

Laval, G., Excoffier, L., 2004. SIMCOAL 2.0: a program to simulate genomic diversityover large recombining regions in a subdivided population with a complexhistory. Bioinformatics 20, 2485–2487.

Librado, P., Rozas, J., 2009. DnaSP v5: a software for comprehensive analysis of DNApolymorphism data. Bioinformatics 25, 1451–1452.

Lisiecki, L.E., Raymo, M.E., 2007. Plio-Pleistocene climate evolution: trends andtransitions in glacial cycle dynamics. Quat. Sci. Rev. 26, 56–69.

Lovette, I.J., 2004. Mitochondrial dating and mixed support for the ‘‘2% rule’’ inbirds. Auk 121, 1–6.

McCormack, J.E., Zellmer, A.L., Knowles, L.L., 2010. Does niche divergenceaccompany allopatric divergence in Aphelocoma jays as predicted underecological speciation?: insights from tests with niche models. Evolution 64,1231–1244.

Pavlova, A., Zink, R.M., Rohwer, S., 2005. Evolutionary history, population genetics,and gene flow in the common rosefinch (Carpodacus erythrinus). Mol.Phylogenet. Evol. 36, 669–681.

Payevsky, V.A., 2008. Breeding and demographic parameters and range expansion ofthe common rosefinch (Carpodacus erythrinus). Ring 30, 63–69.

Peterson, A.T., Sobersón, J., Sánchez-Cordero, V., 1999. Conservatism of ecologicalniches in evolutionary time. Science 285, 1265–1267.

Phillips, S.J., Anderson, R.P., Schapire, R.E., 2006. Maximum entropy modeling ofspecies geographic distributions. Ecol. Model. 19, 231–259.

Polzin, T., Daneschmand, S.V., 2003. On Steiner trees and minimum spanning treesin hypergraphs. Oper. Res. Lett. 31, 12–20.

Schluter, D., 2009. Evidence for ecological speciation and its alternative. Science323, 737–741.

Schneider, S., Roessli, D., Excoffier, L., 2000. Arlequin: A Software for PopulationGenetics Data Analysis. Version 2. Genetics and Biometry Lab, Dept ofAnthropology, Univ of Geneva, Switzerland. <http://anthro.unige.ch/software/arlequin/>.

Sousa-Santos, C., Robalo, J.I., Collarrs-Pereira, M.-J., Almada, V.C., 2005.Heterozygous indels as useful tools in the reconstruction of DNA sequencesand in the assessment of ploidy level and genomic constitution of hybridorganisms. DNA Sequence 16, 462–467.

Stephens, M., Scheet, P., 2005. Accounting for decay of linkage disequilibrium inhaplotype inference and missing-data imputation. Am. J. Hum. Genet. 76, 449–462.

Stephens, M., Smith, N., Donnelly, P., 2001. A new statistical method for haplotypereconstruction from population data. Am. J. Hum. Genet. 68, 978–989.

Stjernberg, T., 1979. Breeding biology and population dynamics of the scarletrosefinch Carpodacus erythrinus. Acta Zool. Fennica 157, 1–88.

Swets, J.A., 1988. Measuring the accuracy of diagnostic systems. Science 240, 1285–1293.

Tamura, K., Dudley, J., Nei, M., Kumar, S., 2007. MEGA4: molecular Evolutionarygenetics analysis (MEGA) software version 4.0. Mol. Biol. Evol. 24, 1596–1599.

Warren, D.L., Glor, R.E., Turelli, M., 2008. Environmental niche equivalency versusconservatism: quantitative approaches to niche evolution. Evolution 62, 2868–2883.

Warren, D.L., Glor, R.E., Turelli, M., 2010. ENMTools: a toolbox for comparativestudies of environmental niche models. Ecography 33, 607–611.

Waters, J.M., 2011. Competitive exclusion: phylogeography’s ‘elephant in theroom’? Mol. Ecol. 20, 4388–4394.

Weir, J.T., Schluter, D., 2008. Calibrating the avian molecular clock. Mol. Ecol. 17,2321–2328.

Zink, R.M., Pavlova, A., Drovetski, S., Rohwer, S., 2008. Mitochondrialphylogeographics of five widespread Eurasian bird species. J. Ornithol. 149,399–413.




http://www.diva-gis.org

http://anthro.unige.ch/software/arlequin/

http://anthro.unige.ch/software/arlequin/


hung et al 2012

Documents