e-dokumen.com

Rodríguez-Clark et al.: Evaluating differentiation for ex situ conservationbeen crossed with canaries Serinus canaria domesticus as well as other finch species, to produce ‘colorbred’ canaries and cage-hardy varieties of siskins(Birkhead 2003, McCarthy 2006), although hybridization has declined in popularity and is generallyfrowned upon among aviculturists today (e.g. Aviculture Hub 2017).Recently, a third potential source of founders wasdiscovered: a population in Guyana, nearly 1000 kmdistant from the previously known Venezuelan range(Robbins et al. 2003; our Fig. 1). Surveys suggest thatthis population is healthy and stable, comprisinghundreds to thousands of individuals, and protections are in place to safeguard against illegal trade(SRCS 2017, society/). The Guyana populationthus represents a potentially attractive source offounders for captive breeding and reintroduction inVenezuela.However, the disjunct nature of this populationraises questions about its potential level of differentiation from other populations. Neotropical bird populations frequently have stronger phylogeographicstructure than their north temperate counterparts,presumably due to longer residence times (e.g. Smithet al. 2014). In addition, Guyanese red siskins may bephysiologically adapted to a different habitat; theGuyanese population inhabits a hot, low-elevationsavanna/forest ecotone, while Venezuelan siskinsoccur at higher elevations in more mesic habitats(Coats & Phelps 1985, Robbins et al. 2003). Thus, theGuyanese population may be sufficiently differentiated to make it undesirable as a source of foundersfor Venezuela. On the other hand, the entire SouthAmerican siskin radiation of 10 species appears tohave been recent and rapid (Beckman & Witt 2015),and a vicariant origin of the Guyana populationbased on savanna expansion in the region wouldimply isolation of just 8000 10 000 years (Van derHammen 1983).The Guyana population could also be the result ofrecent long-distance dispersal or an anthropogenicintroduction (Robbins et al. 2003). In either of thesescenarios, we would expect the gene pool of theGuyana population to be a subset of the Venezuelagene pool, with the likelihood of reduced diversity dueto founder effects. The red siskin is a seminomadic,flocking species which may therefore exhibit long-distance dispersal. Red siskin populations in Cuba andPuerto Rico may be derived from escaped cage birds(Raffaele 1983, Lever 1987, Collar 1992), and manyother feral populations of small finches exist in theGuianas and the Caribbean (Bond 1971). Yet, while185Guyana has been a source of bird trafficking for centuries (Hanks 2005), at the time of discovery of theGuyana population, traffic in red siskins was unknownin that country (Robbins et al. 2003).In the present paper, our aim was therefore toevaluate both nuclear and mitochondrial (mtDNA)genetic differentiation between the Guyanese andVenezuelan populations, in order to explore the possibility of using Guyanese individuals as founders fora captive breeding program to restore populations inVenezuela. Here we report the results of sequencecomparisons of 2 mtDNA genes (cytochrome B andcontrol region) and nuclear genetic variation at 312loci surveyed by the amplified fragment length polymorphism (AFLP) technique. A finding of little to nogenetic differentiation would support the use of individuals from Guyana in ex situ conservation effortsaimed at restoring red siskins in Venezuela, while afinding of significant differentiation would caution usin such an endeavor.MATERIALS AND METHODSSample collection and DNA extractionSamples of this species are extremely difficult to obtain both in the wild and from captive flocks. In thewild, birds are nomadic, high-flying, sparsely distributed, and Endangered, and permits for their captureand sampling are a major challenge, often requiringyears of effort. Many owners of captive birds do nothave proper paperwork and/or are reluctant to allowsampling of valuable, delicate birds. Thus, the samples we were able to obtain were not ideal. However,given the importance of the conservation question athand, we preferred to use available samples and carefully consider potential sources of bias when drawingwhat we believe are conservative inferences.The 5 samples available from Guyana (GU; Table 1)were from wild birds in adult plumage sampled at asingle location on 12 April 2000 following their unexpected discovery during an ornithological survey ofthe Rupununi Savanna, conducted with the permission of the Guyana Environmental Protection Agencyand Ministry of Amerinidian Affairs (Robbins et al.2003; our Fig. 1). Efforts by M.J.B. and M. Robbins,after this discovery, were instrumental in establishing legal protection for this species in Guyana. Samples were frozen in the field in liquid nitrogen andmaintained at 130 C or below.Most or all red siskins in captivity today worldwideare thought to derive from Venezuelan stock, al-

Endang Species Res 36: 183–196, 2018186Table 1. Spinus cucullatus specimens used in the present study. ‘VE’: ofpresumed Venezuelan origin; GU: of Guyanese origin; USNM: US NationalMuseum of Natural History, Washington, DC (USA)Mitochondrial DNA amplification,sequencing, and editingWe amplified 2 mitochondrial genes:cytochrome B, and a portion of the control region. For cytochrome B, we usedthe primers L14764 (5’-TGR TAC AAA‘VE’B200142000 2006 Florida1‘VE’B2001519982004Oregon1AAA ATA GGM CCM GAA GG-3’;‘VE’B2001620042004Oregon1 Full sib to SI-B-20017Sorenson et al. 1999) and H16060 (5’‘VE’B2001720042004Oregon1 Full sib to SI-B-20016TTT GGY TTA CAA GAC CAA TG-3’;‘VE’B2001820042006Oregon1Robbins et al. 2005) to amplify 2004Oregon2entire coding sequence and short flank‘VE’B2002120042006Florida2ing regions. For the control region, we‘VE’B2002220052010 California1designed species-specific primers tar‘VE’B2002320092010Oregon1 Full sib to SI-B-20024geting an initial segment of 670 bp:‘VE’B2002420102010Oregon1 Full sib to SI-B-20023RSCRL000 (5’-CTC TCT CCG 2010Oregon1TCT ATG GCC TGA A-3’) andGU USNMxxxxx2a2000WildRSCRH690 (5’-CAC TTG AAG GGCaGU USNMxxxxx42000WildTTA TTG AAG AGA C-3’). All ampliaGU USNMxxxxx52000Wildcons were sequenced on both strandsGU USNMxxxxx6a2000WildGU USNMxxxxx3a2000Wildwith additional internal primers, andareads were assembled with SequencherComplete ID numbers available upon request to safeguard the locationof the GU population ({Lindenmayer & Scheele 2017)5.0 (GeneCodes) to arrive at consensussequences for each individual. Full details of amplification and sequencingprotocols are given in the Supplement at www.int-res.though records of their origin are generally unavailcom/articles/suppl/n036p183 supp.pdf. No insertionsable. The 13 individuals used to represent Venezuelaor deletions were detected in either gene, and no stop(‘VE’) in the present study came from 5 captive flockscodons in the case of cytochrome B, suggesting thatin 3 US states (Table 1). They were donated for thisour sequences were of mitochondrial origin and notstudy following death from natural causes, and hadnuclear pseudogenes (Sorenson & Quinn 1998). Genesno morphological traits suggesting hybrid ancestry.were concatenated for all analyses.These ‘VE’ individuals consisted of 2 pairs of knownsiblings and 9 individuals with no known first-orderrelationships. However, given the small size of USAFLP scoringavicultural flocks and regular transfers among them,some level of relatedness and inbreeding is plausibleTo compare GU and ‘VE’ individuals across the nuamong all 13 (P. Hansen pers. comm.). Captive specclear genome, we developed a set of AFLP markersimens were stored at 20 C after death, shipped on(Vos et al. 1995, Bensch & Akesson 2005, Meudt &dry ice, and subsequently maintained at 80 C. GenClarke 2007). We used the protocol of Kingston &etic samples and voucher specimens will be deposiRosel (2004) with some modifications (see ‘Detailedted at the US National Museum of Natural Historymethods’ in the Supplement) to screen all 18 individ(USNM) of the Smithsonian Institution, in Washinguals for variation with 14 selective primer pair combiton, DC.nations (Table S1 in the Supplement). We used anGenomic DNA was extracted from all GU samplesABI Prism 3100 genetic analyzer to detect fragmentand ‘VE’ samples B20014 through B20021 using stansizes, multiplexing 2 selective PCR products labeleddard phenol chloroform extraction. (Sambrook et al.with different dyes in each run. Electropherograms1989). DNA from the remaining ‘VE’ samples waswere scored using GeneMapper 4.0, followingextracted on an automated Autogenprep 965 extracKingston & Rosel (2004). Polymorphic peaks weretor (Autogen) following the manufacturer’s instrucscored as dominant, biallelic markers (Vos et al.tions using a standard mouse tissue protocol. DNA1995). We used strict scoring criteria to minimize erconcentration and purity were assessed using a Nanrors, only scoring peaks larger than the second smalloDrop ND-1000 spectrophotometer (ThermoFisherest size standard (89 bp) and smaller than the secondScientific).OriginUSNMID numberHatch SamplingyearyearSourceaviaryKnownrelationships

Rodríguez-Clark et al.: Evaluating differentiation for ex situ conservationlargest (508 bp) or the last monomorphic peak,whichever was smallest. Peaks had to conform to thefollowing additional criteria, following Bonin et al.(2007): fluorescence intensity above 100; low baselinefluorescence; a clean negative control; a clear, singlebase width profile without ‘shoulders;’ no peak at thesame location in the co-loaded PCR product or sizestandard; no closer than 3 bp from another fragment;and strong sample amplification across the entire sizerange of fragments. Two coauthors (K.M.R.C. andB.D.) scored all fragments in all individuals separatelyand removed any loci with discrepancies. In total, wedeveloped 312 loci that could be reliably scored.Data analysesData from all individuals (5 GU, 13 ‘VE’) were usedfor the analyses described below.mtDNATo examine variation in mitochondrial DNA sequences, we constructed a median-joining network usingNetwork (Fluxus Technology Ltd. 2009). We also calculated total haplotype diversity within each population (Ha t), and nucleotide diversity (π) as well asdivergence with FST and ΦST as implemented in Arlequin (Excoffier et al. 2005).AFLPPopulation allele frequencies for all AFLP loci wereestimated from observed fragment frequencies in theGU and ‘VE’ samples using a Bayesian approach implemented in AFLP-SURV (Zhivotovsky 1999). Weapplied a non-uniform prior of allele frequenciescomputed by combining sample size and the numberof individuals without fragment presence to take intoaccount small sample sizes. We estimated allele frequencies assuming Hardy-Weinberg equilibrium(HWE; F IS 0), as well as assuming average F ISvalues ranging from 0.0625 (individuals related onaverage at the level of second cousins) to 0.75 (theequivalent of 4 generations of full-sib mating). In order to assess the effect of possible size homoplasy onour estimates of genetic divergence with these markers, we also calculated the average fragment size andthe Pearson correlation coefficient (r) between fragment size and frequency, along with its significance(Vekemans et al. 2002). Finally, to better separate and187understand possible sources of bias, we calculated relatedness (rab ) of each individual with respect to individuals from the country of origin of that individual,following Lynch & Milligan (1994), and comparedthem with levels of relatedness between known sibpairs, to test assumptions about probable levels of F IS.The percentage of polymorphic AFLP loci (at 5% orabove; PL), Nei’s gene diversity (H j , equivalent toexpected heterozygosity), and FST between the 2populations were computed with AFLP-SURV (Vekemans 2002), using the range of allele frequency estimates described above. FST values were tested forsignificance by comparing observed values with thedistribution of values in 10 000 random permutationsof individuals among groups, calculated on the basisof expected heterozygosity of dominant marker loci(Lynch & Milligan 1994, Vekemans 2002). We alsosearched for significant linkage disequilibrium (LD)among all locus pairs using an algorithm for dominant markers (Li et al. 2007). LD analysis can alsoreveal aspects of population structure not evident inindividual-locus analyses, since extensive non-random associations of allele frequencies across manyloci can indicate recent founder events and/or bottlenecks that would be expected if, for example, theGuyana population had a recent origin from few captive individuals.Population structure in the AFLP data was furtherexplored using a variety of methods. To visualize differentiation, we used NTSYSpc version 2.2 (Rolf2008) to first create a matrix of band-sharing between individuals as measured by the Jaccard similarity value, a metric appropriate for dominant locibecause it makes no assumption of homology amongband-absent genotypes (Ajmone-Marsan et al. 2002).We then represented the relationships revealed bythese values using an ordination technique, i.e. nonmetric multi-dimensional scaling (NMDS; Rolf 2008).Private alleles and diagnostic loci are expected toaccumulate in isolated populations over time, withincreasing numbers indicating increasing divergence (Schönswetter et al. 2004). Private alleles arethose for which fragment presence is observed in just1 population, and diagnostic loci are those which distinguish all individuals of a population or group fromall individuals of other populations or groups. Weestimated the number of private alleles (Np) anddiagnostic loci (Nd) from our AFLP data, and calculated the probability of the observed numbers occurring by chance using randomization (Manly 1997).We reconfigured our dataset 1000 times (consideringonly 1 randomly-selected sib per known sibset), reassigning individuals each time randomly to a group

188Endang Species Res 36: 183–196, 2018(GU or ‘VE’). For each reconfiguration, we tallied thenumber of private alleles and diagnostic loci betweenthose 2 groups. We then calculated a p-value as theproportion of null model iterations where Np and Ndexceeded or were equal to the observed values. Calculations were performed with Excel (Microsoft).In addition to classical analyses of predefinedgroups, we used 2 clustering methods to investigatepopulation structure in our samples. We first appliedmodel-based Bayesian clustering analyses using algorithms appropriate for dominant loci as implemented in STRUCTURE 2.3.4 (Pritchard et al. 2000,Bonin et al. 2007, Falush et al. 2007). These analysesassume HWE, an assumption likely to be violatedby the known and potential familial relationshipspresent in our samples. However, although family relationships and inbreeding can lead to an overestimation of the number of distinct population clusters K,there appears to be little effect on the correct assignment of individuals to populations for a fixed K(Falush et al. 2003, Pritchard et al. 2010). We chosenot to include prior information about sample originin STRUCTURE models, because the geographic origin of ‘VE’ samples is presumed rather than known.We used a standard admixture model to allow for thepossibility of mixed ancestry, assumed that allele frequencies in each population were correlated, andconducted unsupervised runs with K from 1 to 6groups. Although the correlated frequencies modelcan overestimate K in the presence of family relationships, it is more appropriate for populations that mayshare ancestry (Pritchard et al. 2010). We chose to runmore chains for shorter periods, in order to examinevariation among runs (Evanno et al. 2005), and usedequal burn-in and data collection periods of 10 000 iterations each, for 20 independent runs per model. Foreach run, we recorded the estimated posterior probability of the data given the assumed model, and usedSTRUCTURE Harvester (Earl & vonHoldt 2012) tocalculate ΔK, an ad hoc statistic based on the secondorder rate of change of the likelihood function withrespect to K (Evanno et al. 2005). We took the modelcorresponding to the modal value of the distributionof ΔK as indicating the likely number of populations,and then plotted each individual’s estimated membership coefficients in those populations (Q) from arepresentative run (as variation among runs was minimal). To consider the case of K 1 (i.e. lack of geneticstructuring), we examined variation in α, a model parameter indicating the extent of admixture: variationamong iterations beyond a range of 0.2 units or morein a single run indicates a lack of true structure. Wealso examined individual assignment and Q valuesfor models corresponding to peak ΔK, becauseroughly equal numbers of individuals assigned toeach putative population and a majority of admixedindividuals also indicate a lack of true structure(Pritchard et al. 2010). Finally, we also consideredwhich value of K had the highest ln Pr(X K ), which isrecommended as an additional indicator of the truevalue of K (Janes et al. 2017).We also used a model-free iterative reallocationmethod, FLOCK 3.1 (Duchesne & Turgeon 2012) toestimate the number of populations, K. This methodis robust to population inbreeding and non-zerorelatedness among sampled individuals because itcreates clusters based on maximizing multilocusgenetic similarity rather than minimizing deviationsfrom HWE and LD. In this method, samples are initially partitioned randomly into K clusters (K 2),allele frequencies are estimated for each of the Kclusters, and each individual is then reallocated tothe cluster that maximizes its likelihood score.Twenty repeated reallocations are performed withineach run, and 50 runs are carried out for each K.Strong consistency among runs, resulting in‘plateaus’ of identical mean log likelihood difference(LLOD) scores, is used to indicate the most likelynumber of clusters (Duchesne & Turgeon 2012).Although it is not run explicitly with K 1, FLOCKdoes test for K 1. In short, K 1 is the defaulthypothesis, and is retained if no plateau of length 6is found for any K 2.Once reference populations have been correctlyidentified, allocation programs take advantage of thisinformation and so are generally less prone to misallocations than are cluster programs. Thus, we alsoperformed reallocation procedures using the methodand software designed for AFLP data by Duchesne &Bernatchez (2002; AFLPOP) to reallocate individualsto populations (‘VE’, GU) based on the allele frequencies. We used the default settings (fixed correctionvalue for 0 frequencies 0.001, minimal LLOD to allocate specimens 0, number of artificial genotypesto compute p-values 500). AFLPOP calculates theLLOD score for each genotype (the difference between the log likelihood of the most likely referencefor the genotype and that of its second most likely reference) and the mean LLOD (MLLOD) over all genotypes. Higher differentiation between references willtend to produce higher MLLOD scores. Marker lociwith minor allele frequency of 5% were consideredmonomorphic and uninformative for the purpose ofour FLOCK and AFLPOP analyses. We therefore discarded loci with either 17 or 18 ‘band present’ phenotypes over all 18 genotypes. Because FLOCK and

Rodríguez-Clark et al.: Evaluating differentiation for ex situ conservationAFLPOP do not accept loci with missing scores, thosewere also removed. Thus 78 loci were retainedamong the 312 loci originally developed and scored.Although the 5 GU samples were from wild adults,given the fact that they all came from the same location, it is possible they were related. Similarly, our 13‘VE’ samples could have additional family relationships unknown to their breeders. With small samplesizes, close relatedness among individuals couldproduce distorted estimates of allele frequencies,mimicking population structure. Therefore, we considered the following hypotheses to explain the presence of GU and ‘VE’ genetic clusters: (1) GU and ‘VE’belonged to 2 distinct populations, (2) GU and ‘VE’ belonged to the same population, but the GU genotypeswere strongly inbred (related at the level of full sibs) sothat enough differentiation was generated to bepicked up by various algorithms, (3) GU and ‘VE’ belonged to the same population, but the ‘VE’ genotypeswere the inbred ones (related at the level of half sibs).We designed a specific procedure in order to testhypotheses that moderate or very high levels of relatedness among the ‘VE’ or GU genotypes mighthave been sufficient to explain the high MLLOD scoreobtained from running the reallocation procedure ofthe AFLPOP program. Essentially, we kept the empirical (real) set of either ‘VE’ or GU genotypes and generated 100 simulated sets of the other genotypes,using allelic frequencies based on the assumption thatall 18 actual specimens (‘VE’ GU) originated fromthe same population. To test hypothesis 2, we simulated sets of 5 genetic full sibs. To test hypothesis 3,we simulated sets of 13 half sibs. Each simulated setstood in place of the empirical GU or ‘VE’ genotypes,respectively, depending on the simulation. We thusran the reallocation procedure of AFLPOP with eachset of the 5 simulated full sibs and 13 ‘VE’ genotypes,or the 13 simulated half sibs and the 5 GU genotypes.For each of the 100 simulations, the MLLOD score ofthe reallocation result was calculated (see Fig. S1a,bin the Supplement). To obtain p-values, we locatedthe MLLOD score from the reallocation of empiricalGU and ‘VE’ genotypes within each distribution of the100 MLLOD scores from both simulation procedures.RESULTSmtDNA variationAll 18 individuals examined had cytochrome B(MT-CYB) sequences consistent with a previouslypublished sequence for this species. No previously189published sequence for the control region (CR) wasfound. Out of a total of 1813 bp sequenced across the2 genes (1143 bp in MT-CYB and 670 bp in C

comparisons of 2 mtDNA genes (cytochrome B and control region) and nuclear genetic variation at 312 loci surveyed by the amplified fragment length poly-morphism (AFLP) technique. A finding of little to no genetic differentiation would support the use o