Turkish Honeybees: Genetic Variation and Evidence for a Fourth ...

42

Table 1. Presence or absence of diagnostic restriction sites in four regions of honeybee mtDNA, andstructure of the noncoding intergenic region

Gene EnzymeWestEuropean

Type 1, easternMediterranean

Type 2, easternMediterranean African

MiddleEastern

Cytochrome b BglII 1 1 1 2 1COI HincII 1 2 2 2 2lsrRNA EcoRI 2 1 2 2 2COI XbaI 2 1 1 2 2COI XbaI 2 2 1 2 2Noncoding sequence PQ Q Q P0Q P1Q

PQQ P0QQPQQQ P0QQQ

Western European, eastern Mediterranean, and African are three lineages of honeybee mtDNA characterized byrestriction site and length differences. Type 1 and type 2 are two variants of the eastern Mediterranean lineage.‘‘Gene’’ indicates the approximate location of restriction sites, ‘‘Enzyme’’ the restriction enzyme. The ‘‘1’’ signindicates presence of a restriction site, ‘‘2’’ indicates its absence. A novel pattern of restriction sites was foundin bees from Hatay, Turkey; here it is called ‘‘Middle Eastern.’’ Primers for cytochrome b reported in Crozier et al.(1991); others in Hall and Smith (1990). Terminology for structure of the noncoding region follows that of Cornuetet al. (1991).

Turkish Honeybees: GeneticVariation and Evidence for aFourth Lineage of Apismellifera mtDNA

M. R. Palmer, D. R. Smith, andO. Kaftanoglu

The mtDNA of bees from 84 colonies ofTurkish honeybees (Apis mellifera) wassurveyed for variation at four diagnosticrestriction sites and the sequence of anoncoding intergenic region. These colo-nies came from 16 locations, ranging fromEuropean Turkey and the western Medi-terranean coast to the Caucasus Moun-tains along the Georgian border, the east-ern Lake Van region, and the extremesouth. Combined restriction site and se-quence data revealed four haplotypes.Three haplotypes belonged to the easternMediterranean mtDNA lineage. The fourthhaplotype, which had a novel restrictionsite pattern and noncoding sequence,was found in samples from the extremesouth, near the Syrian border. We foundtwo different noncoding sequencesamong the eastern Mediterranean haplo-types. The ‘‘Caucasian’’ sequence match-es that described from A. m. caucasica,and the ‘‘Anatolian’’ sequence matchesthat of A. m. carnica. The frequency of the‘‘Caucasian’’ sequence was highest (98–100%) in sites near the Georgian borderand decreased steeply to the south andwest. Elsewhere the Anatolian sequencewas found. In European Turkey (Thrace) arestriction site polymorphism previouslyreported from A. m. carnica in Austria andthe Balkans was present at high frequen-cy. A novel mtDNA haplotype with aunique restriction site pattern and noncod-ing sequence was found among beesfrom Hatay, in the extreme south near theSyrian border. This haplotype differed fromthe three previously known lineages ofhoneybee mtDNA—African, western Eu-ropean, and eastern Mediterranean—andmay represent a fourth mitochondrial lin-eage.

Honeybees (Apis mellifera) are geographi-cally diverse, with as many as 25 subspe-cies currently recognized (Ruttner 1988;Sheppard et al. 1997). The honeybee mi-tochondrial genome has provided abun-dant data for studies of honeybee phylog-eny and biogeography. Studies of mitochon-drial DNA (mtDNA) restriction site poly-morphisms (e.g., Crozier et al. 1991; Gar-nery et al. 1993; Hall and Smith 1991; Meix-

ner et al. 1993; Moritz et al. 1994; Sheppardet al. 1996; Smith 1991a,b; Smith andBrown 1990; Smith et al. 1991) and se-quence polymorphisms (Arias and Shep-pard 1996; Garnery et al. 1992; Lee andHall 1996) have revealed three main lin-eages of honeybee mtDNA: western Euro-pean, eastern Mediterranean, and African.

The mitochondrial genome of honey-bees also contains a noncoding region lo-cated between a leucine tRNA gene andthe cytochrome oxidase II gene (COII; Cor-nuet et al. 1991). This noncoding sequencehas a complex structure, which gives riseto both sequence and length variation(Hall and Smith 1991). Two basic ele-ments, called ‘‘P’’ (or ‘‘P0’’) and ‘‘Q’’ arefound in the noncoding sequence; lengthvariation stems from presence or absenceof the P element and from tandem repeatsof the Q element. The Q element itself con-sists of three subunits—Q1, Q2, and Q3;Q3 is usually identical to the P elementfrom the same genome (Cornuet et al.1991). Sequence variation includes bothbase substitutions and insertions/dele-tions. The three mitochondrial lineagesdiffer in length and sequence of the non-coding region: in the eastern Mediterra-nean lineage one finds a single Q element;in the western European and African lin-eages one finds a P element followed byone, two, or three repeats of the Q ele-ment. The P element of the western Eu-ropean lineage has a 15-base deletion rel-ative to the African ‘‘P0’’ element (Cornuetet al. 1991). These lineage-specific differ-ences are summarized in Table 1 and Fig-ure 1.

The geographic distribution of these mi-tochondrial lineages corresponds roughlywith the distributions of honeybee sub-species: western European mtDNA isfound primarily in A. m. mellifera and

some Spanish honeybees, eastern Medi-terranean mtDNA in A. m. ligustica, A. m.carnica, and A. m. caucasica, and AfricanmtDNA in A. m. scutellata and other beesfrom Africa.

The bees of Turkey are particularly rel-evant for studies of honeybee biogeogra-phy. Turkey is located at the geographiccrossroads of Europe, Asia, and the Mid-dle East and contains a wide range of cli-mates and habitats within its borders. Notsurprisingly, the honeybees of Turkey arealso quite diverse; based on morphomet-ric, behavioral, and ecological data, Rutt-ner (1988) suggested that four subspeciesoccur in Turkey: A. m. anatoliaca, A. m.caucasica, A. m. meda, and A. m. syriaca.According to Ruttner, A. m. caucasica oc-curs in the extreme northeast of Anatolia(Asian Turkey), with bees resembling A.m. caucasica occurring along the easternBlack Sea coast as far as Samsun. A. m.meda is found in the southeast, and A. m.syriaca in the extreme south, near the bor-der with Syria. A. m. anatoliaca occursthroughout the rest of Turkey, includingEuropean Turkey.

Genetic studies of the honeybees of Tur-key include allozyme studies (Asal et al.1995; Kandemir and Kence 1995) and astudy of mtDNA restriction site polymor-phisms (Smith et al. 1997). The restrictionsite study showed that our Turkish sam-ples possessed the eastern Mediterraneanlineage of honeybee mtDNA. Even thoughtwo subspecies—A. m. anatoliaca and A.m. caucasica—were believed to be repre-sented in our samples, the only variationwe found was among the bees of Thrace(European Turkey), where we found anXbaI site previously known only from A. m.carnica (Meixner et al. 1993; Smith andBrown 1990).

Here we examine mtDNA restriction site

Brief Communications 43

Figure 1. Structure and sequence of the noncoding region located between leucine tRNA and cytochrome oxidaseII genes in the A. mellifera mitochondrial genome. The ‘‘P’’ and ‘‘Q’’ terminology follows the usage of Cornuet etal. (1991). (A) Structure of the noncoding region in three major lineages of honeybee mtDNA (W 5 westernEuropean, E 5 eastern Mediterranean, A 5 African) and the novel mitochondrial haplotype found in Hatay (M 5Middle Eastern). In African and western European mtDNAs, the Q element may be repeated one to three times.(B) Sequence of the noncoding region. Sequence of each element and subunit (P, Q1, Q2, Q3) shown separately. A5 A. m. scutellata from African lineage; W 5 A. m. mellifera from western European lineage; M 5 Syrian sequencefrom the proposed Middle Eastern lineage; E-c 5 A. m. caucasica from the eastern Mediterranean lineage; E-l 5 A.m. ligustica from the east Mediterranean lineage. A. m. ligustica and A. m. scutellata sequences from Garnery et al.(1992).

Figure 2. Approximate ranges of A. m. anatoliaca, A. m. caucasica, A. m. meda, and A. m. syriaca in Turkey assuggested by the morphometric studies of Ruttner (1988; dotted lines), and distribution of four mtDNA haplotypesfound in this study (pie charts). Samples were collected from starred locations. Key to pie charts: Gray shadingindicates the frequency of haplotypes of the eastern Mediterranean mitochondrial lineage with type 1 restrictionsite pattern and Anatolian noncoding intergenic sequence; striped shading, eastern Mediterranean mitochondriallineage, type 2 restriction site pattern, Anatolian noncoding sequence; black shading, eastern Mediterranean mi-tochondrial lineage, type 1 restriction site pattern, Caucasian noncoding sequence; white, Syrian sequence withMiddle Eastern restriction site pattern. Restriction site patterns are described in Table 1, sequences are shown inFigure 1.

and sequence variation in bees collectedfrom 16 localities in Turkey: the 12 sitesreported earlier (Smith et al. 1997) and 4additional sites in southern and easternTurkey. We survey the four restrictionsites, which distinguish three major mt-DNA lineages within A. mellifera, and se-quence the noncoding intergenic region ofthe mitochondrial genome (Cornuet et al.1991). Our restriction site survey and se-quence data reveal a novel mtDNA haplo-type, which may constitute a fourth mito-chondrial lineage in A. mellifera.

Methods

CollectionsAdult worker honeybees were collectedfrom comb (or in one case from a swarm)and frozen in liquid nitrogen or preservedin 70% ethanol. Samples were collectedfrom colonies in the following locations(see Figure 2). In 1994, samples were col-lected from Bursa (one colony) and vil-lages near Giresun (four colonies). In June1995, samples were collected from Thrace(7 colonies from villages near Tekirdag);Gokceada (10 colonies); the Black Seacoast near Bolu and Yedigoller (4 colo-nies); Menemen, Aegean Agricultural Re-search Institute (9 colonies); Beypazari (7colonies); Erzurum (7 colonies); Ardahan(5 colonies); villages near Ardanuc (6 col-onies); villages near Artvin (5 colonies);and the villages of Posof, Savsat, and Sun-gullu near the Georgian border (16 colo-nies). The bees from Bolu, Savsat, andBeypazari came from breeding coloniesmaintained by the Beekeeping Project ofthe Turkish Development Foundation(TKV); the rest were collected in situ. In1998, samples were collected from Bitlis (9colonies), Mus (2 colonies), Van (4 colo-nies), and Hatay (14 colonies).

These collection sites fall within theranges of A. m. anatoliaca, A. m. caucasica,A. m. meda, and A. m. syriaca as describedby Ruttner (1988; see Figure 2). However,migratory bee-keeping is widely practicedin Turkey, and Caucasian bees (A. m. cau-casica) are highly prized by beekeepers.Both factors (especially the latter) canlead to transplantation and mixing of pop-ulations. Although none of the samplesused in this study were subjected to mor-phometric analysis, we are confident thatthey are, for the most part, representativeof regional populations. First, Guler (1996)and Guler et al. (1999) carried out a mor-phometric analysis of other samples fromsome of our 1995 collection sites. Theirsamples from Posof, Sungullu, and Arda-

44 The Journal of Heredity 2000:91(1)

Table 2. Frequency of three restriction site patterns in the mtDNA of Turkish honeybees from 17localities: eastern Mediterranean (type 1 and 2) and ‘‘Middle Eastern’’

Eastern Mediterranean

Locality

Type 1

Number Percent

Type 2

Number Percent

Middle Eastern

Number Percent

Tekirdag 1 14 6 86 0 0Gokceada 10 100 0 0 0 0Bursa 1 100 0 0 0 0Bolu 2 100 0 0 0 0Menemen 9 100 0 0 0 0Fethiye 9 100 0 0 0 0Beypazari 6 100 0 0 0 0Giresun 4 100 0 0 0 0Ardahan 6 100 0 0 0 0Posof and Savsat 9 100 0 0 0 0Artvin 5 100 0 0 0 0Ardanuc 6 100 0 0 0 0Erzurum 6 100 0 0 0 0Bitlis 9 100 0 0 0 0Van 4 100 0 0 0 0Mus 2 100 0 0 0 0Hatay 6 43 0 0 8 57

The restriction site patterns are described in Table 1.

han corresponded to published character-istics of A. m. caucasica. They also exam-ined samples from Gokceade, Beypazari,Fethiye, and Thrace; although they foundregional variation among these popula-tions, all correspond to A. m. anatoliaca.Second, some samples were taken from re-search apiaries that maintain honeybeesubspecies stocks (e.g., Aegean Agricul-tural Research Station, Menemen; Bee-keeping Project of the Turkish Develop-ment Foundation (TKV), Beypazari; honeybeebreeding station, Ardahan). Others arefrom regions in which migratory bee keep-ing is rare or prohibited (e.g., Gokceada,Ardahan). Finally, since A. m. caucasica isso highly prized, ‘‘foreign’’ populations ofhoneybees are not likely to be importedinto its range in Turkey, though A. m. cau-casica is likely to be exported and estab-lished in new locations. In the final analy-sis, however, subspecies designations arenot critical to our study. We are docu-menting variation present in honeybeemtDNA and its geographic distribution.The distribution of mtDNA haplotypesmay not always match subspecies desig-nations based on morphometrics.

Laboratory MethodsTotal DNA was prepared by proteinase Kdigestion of single thoraces, followed byphenol extraction and ethanol precipita-tion as described in Smith et al. (1997) orwith Qiagen Tissue Prep kits (Qiagen, Va-lencia, CA). The four regions of the mito-chondrial genome used to diagnose lin-eages of honeybee mtDNA were amplifiedby means of the polymerase chain reac-tion (PCR; Saiki et al. 1985) and digestedwith the appropriate restriction enzymes,as described in Smith et al. (1997). Theamplified regions, restriction enzymes,and pattern of restriction sites character-istic of each honeybee mtDNA lineage areshown in Table 1.

One of these amplified regions extendsfrom the 39 end of cytochrome oxidase I(COI) to the 59 end of COII; this fragmentincludes the noncoding intergenic regiondiscussed above. The noncoding regionwas sequenced using the internal primer59-GGCAGAATAAGTGCATTG-39 (Cornuet etal. 1991). Sequencing reactions were car-ried out using the fMOL (Promega) cyclesequencing protocol with 32P-labeled se-quencing primer.

Results

Three restriction site patterns were foundin the mitochondrial genomes of these

Turkish honeybee samples. Most colonieshad the restriction site pattern character-istic of the eastern Mediterranean mtDNAlineage (Tables 1 and 2). As reported ear-lier (Smith et al. 1997), a variant of thispattern in which there are two (ratherthan one) XbaI sites in the amplified COIfragment was found in six of seven colo-nies from Thrace. A novel restriction sitepattern was found in some colonies fromHatay (Tables 1 and 2); this haplotypemay represent a fourth mitochondrial lin-eage. Here we refer to both the haplotypeand the mitochondrial lineage as ‘‘MiddleEastern.’’

We also found three sequence variantsin the noncoding intergenic region of Turk-ish honeybee mtDNA (Figure 1). All of ournoncoding sequences differed in one re-spect from those reported by Garnery etal. (1992). In their study the noncoding se-quence began with ATTTCCCC in A. m. li-gustica and A. m. carnica, and ATTTCCC-(single-base deletion) only in A. m.caucasica; all of our Turkish sequences, re-gardless of origin, began ATTTCCC-.

One of our three mtDNA sequencesmatched (except for the missing ‘‘C’’ men-tioned above) the sequence reported forA. m. carnica, and a second matched thesequence reported for A. m. caucasica(Garnery et al. 1992). Here we refer tothese sequences as ‘‘Anatolian’’ and ‘‘Cau-casian,’’ respectively.

We call the third sequence ‘‘Syrian’’ (notreported in the literature), as it was foundin bees collected in Hatay, near the Syrianborder and within the reported range(Ruttner 1988) of A. m. syriaca. The novelSyrian sequence contained a P element

( like the African and west European lin-eages) and one Q element. The Syrian P(we call it ‘‘P1’’) is identical to the AfricanP0 element except for one base substitu-tion and a 3-base insertion (Figure 1). TheSyrian Q element differs from those of theother lineages by a series of base substi-tutions, additions, and deletions.

The Syrian sequence was always foundwith Middle Eastern restriction site pat-tern. Although we did not sequence alleight examples of mtDNA with the MiddleEastern restriction site pattern, we inferthat they all had the Syrian intergenic se-quence, based on size differences in theamplified mtDNA fragment: the Anatolianintergenic sequence (Q) is shorter thanthe Syrian (P1Q).

Table 3 shows the frequency and geo-graphic distribution of the three intergenicsequences. Caucasian sequence is highestnear the Georgian border and decreases tothe west and south. Fourteen of 16 colo-nies (87.5%) from sites near the Georgianborder (Posof, Savsat) and 10 of 15 colo-nies (66%) from localities 20–40 km fromthe Georgian border (Ardahan, Ardanuc,and Artvin) had the Caucasian sequence.Two of 7 colonies (29%) from Erzurum (ap-proximately 240 km south of the Georgianborder) and 3 of 12 colonies (25%) in theLake Van region (Van, Mus, Bitlis) had theCaucasian sequence. The novel Syrian se-quence was found only in our samplesfrom Hatay. All other colonies sampledfrom other localities had only the Anato-lian sequence. Of these, six of seven col-onies from Thrace had the eastern Medi-terranean type 2 restriction site pattern,while the rest had the type 1 pattern.


Table 3. Frequency of three noncoding mitochondrial sequences—‘‘Anatolian,’’ ‘‘Caucasian,’’ and‘‘Syrian’’—in Turkish honeybees

Locality

Anatolian

Number Percent

Caucasian

Number Percent

Syrian

Number Percent

Tekirdag 7 100 0 0 0 0Gokceada 1 100 0 0 0 0Bursa 1 100 0 0 0 0Bolu 4 100 0 0 0 0Menemen 2 100 0 0 0 0Beypazari 7 100 0 0 0 0Giresun 4 100 0 0 0 0Ardahan 0 0 5 100 0 0Posof and Savsat 2 12.5 14 87.5 0 0Artvin 2 40 3 60 0 0Ardanuc 3 60 2 40 0 0Erzurum 5 71 2 29 0 0Bitlis 3 50 3 50 0 0Van 4 100 0 0 0 0Mus 2 100 0 0 0 0Hataya 4 50 0 0 4 50

Anatolian and Caucasian match published A. m. carnica and A. m. caucasica sequences, respectively (Garnery etal. 1992); Syrian is found in bees with the Middle Eastern restriction site pattern (Tables 1 and 2).

a Four of each restriction site pattern (Table 2) were selected for sequencing.

Discussion

Our restriction site and sequence datacombined show four mitochondrial vari-ants in Turkey. Three of the four haploty-pes belong to the eastern Mediterraneanmitochondrial lineage; the fourth, foundonly in samples from Hatay, does not cor-respond to any of three previously report-ed mitochondrial lineages.

The most common and widespreadmtDNA haplotype in our samples has theAnatolian intergenic sequence and the re-striction site pattern ‘‘eastern Mediterra-nean type 10 (Table 1). In Thrace (Euro-pean Turkey) a haplotype with theAnatolian intergenic sequence and the re-striction site pattern ‘‘eastern Mediterra-nean type 2’’ occurs at high frequency.Since this restriction site pattern is alsofound among A. m. carnica from Austria,Slovenia, and Croatia (Meixner et al. 1993;Smith and Brown 1990), it suggests mater-nal gene flow among the bees of Thrace,the Balkans, and southern Austria. Thefact that this restriction site was not foundin any bees from Anatolia suggests theremay be a barrier to maternal gene flow be-tween these two regions—though moresampling, especially in northwest Anato-lia, is needed to test this.

A third haplotype has the eastern Med-iterranean (type 1) restriction site patternand the Caucasian intergenic sequence.This matches the mitochondrial haplotypepreviously described from A. m. caucasica(Garnery et al. 1992; Smith 1988). Wefound this sequence in high frequencynear the Georgian border and in lower fre-quency in Erzurum and the region aroundLake Van.

The ‘‘homeland’’ of A. m. caucasica is inthe Caucasus mountains, southern valleysof the Caucasus, and the higher reaches ofthe Little Caucasus mountains (Ruttner1988), primarily in Georgia and neighbor-ing republics. The full extent of the rangeof this bee is not clear. Ruttner states thatbees resembling A. m. caucasica occuralong the Black Sea coast of Anatolia asfar as Samsun (Ruttner 1988; pp. 178, 192–198), but so far we have no evidence of theCaucasian sequence along the southernBlack Sea coast, though our samples fromthis region are admittedly small.

The Caucasian sequence is found in Er-zurum, significantly south of the proposedrange of A. m. caucasica. If the intergenicsequence described by Garnery et al.(1992) for A. m. caucasica is indeed char-acteristic of the entire subspecies, thenour results indicate a wide zone of inter-action between A. m. caucasica and A. m.anatoliaca, at least from Lake Van to theGeorgian border (Figure 2). This could bedue either to natural gene flow and dis-persal or to transportation of A. m. cau-casica by humans. More extensive collec-tions from the northwestern area ofTurkey and the heart of the A. m. caucas-ica range in Georgia would enable us todetermine if the intergenic sequence wecall ‘‘Caucasian’’ is actually characteristicof all or most A. m. caucasica. Nothing inour mtDNA data suggested that the beesfrom the Lake Van region, supposedly inthe range of A. m. meda, were in any waydistinct from other Anatolian populations.

The fourth haplotype, which is charac-terized by a novel restriction site patternand intergenic sequence, does not match

any of the three previously documentedhoneybee mitochondrial lineages and mayconstitute a fourth lineage. This was foundin high frequency (roughly 50%) in beesfrom Hatay. This southern city lies at thenorthern edge of the proposed range of A.m. syriaca: along the eastern coast of theMediterranean north of the Negev desert,crossing parts of Israel, Jordan, Syria, andLebanon. Little genetic work has been car-ried out on the native honeybees of theMiddle East, principally because importedA. m. ligustica, A. m. carnica, and otherraces have largely replaced the nativehoneybee where modern apiculture ispracticed (Lensky Y, personal communi-cation to D.R.S.). Our data suggests that afourth mitochondrial lineage occursamong Middle Eastern honeybees. If thiswere so, it could change our ideas aboutthe relationships among mitochondrial lin-eages and the biogeography of A. melli-fera.

Earlier morphometric studies (summa-rized in Ruttner 1988) indicated four sub-species groups: M, C, A, and O or Oriental.Ruttner (1988) pointed out that the mor-phometrically based groups do not nec-essarily reflect phylogenetic relationships,and we have found that the match be-tween morphometric subspecies groupsand mtDNA lineages is not exact (e.g., seeSmith 1991a). For example, the morpho-metric C branch includes the subspeciesA. m. carnica and A. m. ligustica, a group-ing also supported by mtDNA data, asboth these subspecies typically carrymtDNA belonging to the eastern Mediter-ranean lineage. However, the morphomet-ric O branch includes A. m. caucasica, A.m. anatoliaca, and A. m. syriaca, a group-ing not supported by mtDNA data. ThemtDNA typically carried by A. m. caucasi-ca and A. m. anatoliaca also belongs to theeastern Mediterranean lineage. We do nothave any verifiable samples of A. m. syr-iaca, but our data show that some bees ofthe Middle East have a distinctive mtDNAhaplotype.

An advantage of mitochondrial dataover morphometric data is that DNA se-quence data can be analyzed easily in aphylogenetic context. This is desirable forinferring the relationships among popula-tions and for inferring the history of pop-ulation movements. A drawback of mtDNAis that it is inherited uniparentally. Whenformerly isolated populations of honey-bees come into contact, whether throughrange expansion or human transportation,mating between members of different pop-ulations can lead to introgression of mt-


DNA haplotypes into new populations.This is probably one source of discor-dance between morphometric and mito-chondrial datasets. However, mtDNA pre-serves information on the relatedness ofqueens and queen lines, and it is an ex-cellent source of data for inferring the his-tory and phylogeography of A. mellifera.

On a practical note, honeybee subspe-cies and ecotypes differ in many physio-logical, ecological, and behavioral charac-ters that make them particularly well-suited to their local environments. The va-riety of subspecies and ecotypes provideample genetic variation for the selectivebreeder. Brother Adam, a respected (evenrevered) honeybee breeder and bee biol-ogist, visited Turkey in 1954 and 1962 inthe course of his quest for honeybees en-dowed with qualities desired by profes-sional beekeepers. In his publications de-scribing his travels he praised theexcellent qualities of the Anatolian bees(Adam 1954, 1964, 1977). Genetic studiesof Turkish honeybee populations will helpin understanding their geographic varia-tion and may aid in maintaining and utiliz-ing their genetic diversity.

From the Department of Entomology, Haworth Hall,University of Kansas, Lawrence, KS 66045 (Palmer andSmith) and Cukurova Universitesi, Ziraat Fakultesi,Adana, Turkey (Kaftanoglu). Work in Turkey was sup-ported by NATO Science for Stability program TU-POL-LINATION Project to O. Kaftanoglu. Work at the Univer-sity of Kansas was supported in part by a grant to O.R. Taylor, Jr., and D. Smith from the U.S. Department ofAgriculture competitive grants program. We thank Dr.Ferat Genc, Dr. Ahmet Guler, and Hakan Kaftanoglu fortheir help in collecting samples, and Ozgen Aksu, Ne-cati Dikilitas, Ahmet Inci, and Dr. Ali Ihsan Ozturk fortheir help in locating suitable colonies. We thank Dr.Harrington Wells (Tulsa University) and Dr. IbrahimCakmak (Uludag University, Bursa) for samples fromBursa and Giresun. We also thank Cengiz Erdem ( Yu-zuncu Yil University, Van) for providing samples fromVan, Bitlis, and Mus, and Dr. Nuray K. Sahinler for pro-viding bees from Hatay. J. Therrien and two anony-mous reviewers provided valuable comments on themanuscript. Address correspondence to Deborah R.Smith at the address above or e-mail: [email protected].

q 2000 The American Genetic Association

References

Adam, 1954. In search of the best strains of bees: sec-ond journey. Bee World 35:193–203, 233–244.

Adam, 1964. In search of the best strains of bee: con-cluding journeys. Bee World 45:70–83, 104–118.

Adam, 1977. In search of the best strains of bee: sup-plementary journey to Asia Minor, 1973. Bee World 58:57–66.

Arias MC and Sheppard WS, 1996. Molecular phyloge-netics of honey bee subspecies (Apis mellifera L.) in-ferred from mitochondrial DNA sequences. Mol PhylogEvol 5:557–566.

Asal S, Kocabas S, Elmaci C, and Yildiz MA, 1995. En-

zyme polymorphism in honey bee (Apis mellifera L.)from Anatolia. Turk J Zool 19:153–156.

Cornuet J-M, Garnery L, and Solignac M, 1991. Putativeorigin and function of the intergenic region betweenCOI and COII of Apis mellifera L. mitochondrial DNA.Genetics 128:393–403.

Crozier YC, Koulianos S, and Crozier RH, 1991. An im-proved test for Africanized honeybee mitochondrialDNA. Experientia 47:968–969.

Garnery L, Cornuet J-M, and Solignac M, 1992. Evolu-tionary history of the honey bee Apis mellifera inferredfrom mitochondrial DNA analysis. Mol Ecol 1:145–154.

Garnery L, Solignac M, Celebrano G, and Cornuet J-M,1993. A simple test using restricted PCR-amplified mi-tochondrial DNA to study the genetic structure of Apismellifera L. Experientia 49:1016–1021.

Guler A, 1996. Turkiyedeki onemli balraisi (Apis melli-fera L.) irk ve ekotiplerinin morfolojik ozelliklerinin be-lirlenmesi ve performanslarinin saptanmasi [Morpho-metric characteristics and performances of honeybee(Apis mellifera L.) races and ecotypes in Turkey] (PhDdissertation). Cukurova Universitesi Fen Bilimleri En-stitusu, Adana, Turkey.

Guler A, Kaftanoakgglu O, Bek Y, and Yeninar H, 1999.Turkiyedeki cesitli balarisi (Apis mellifera) irk ve eko-tiplerinin morfolojik karakterler acisindan iliskilerinindiskriminant analiz yontemi ile saptanmasi. TURBITAKDoga 23:337–344.

Hall HG and Smith DR, 1991. Distinguishing African andEuropean honey bee matrilines using amplified mito-chondrial DNA. Proc Natl Acad Sci USA 88:4548–4552.

Kandemir I and Kence A, 1995. Allozyme variability ina central Anatolian honeybee (Apis mellifera L.) popu-lation. Apidologie 26:503–510.

Lee ML and Hall HG, 1996. Identification of mitochon-drial DNA of Apis mellifera (Hymenoptera: Apidae) sub-species groups by multiplex allele-specific amplifica-tion with competing fluorescent-labeled primers. AnnEntomol Soc Am 89:20–27.

Meixner MD, Sheppard WS, and Poklukar J, 1993. Asym-metrical distribution of a mitochondrial DNA polymor-phism between 2 introgressing honey bee subspecies.Apidologie 24:147–153.

Moritz RFA, Cornuet J-M, Kryger P, Garnery L, and Hep-burn HR, 1994. Mitochondrial DNA variability in SouthAfrican honeybees (Apis mellifera L.). Apidologie 25:169–178.

Ruttner F, 1988. Biogeography and taxonomy of honeybees. Berlin: Springer-Verlag.

Saiki RK, Scharf S, Faloona F, Mullis KB, Horn GT, ErlichHA, and Arnheim N, 1985. Enzymatic amplification ofb-globin genomic sequences and restriction site anal-ysis for diagnosis of sickle cell anemia. Science 230:1350–1354.

Sheppard WS, Arias MC, Grech A, and Meixner MD,1997. Apis mellifera ruttneri, a new honey bee subspe-cies from Malta. Apidologie 28:287–293.

Sheppard WS, Rinderer TE, Meixner MD, Yoo HR, Stel-zer JA, Schiff NM, Kamel SM, and Krell R, 1996. HinfIvariation in mitochondrial DNA of Old World honey beesubspecies. J Hered 87:35–40.

Smith DR, 1988. Mitochondrial DNA polymorphisms infive Old World subspecies of honey bees and in NewWorld hybrids. In: Africanized honey bees and beemites (Needham GR, Page RE, Delfinado-Baker M, andBowman CE, eds). Chichester: Ellis Horwood; 303–312.

Smith DR, 1991a. African bees in the Americas: insightsfrom biogeography and genetics. Trends Ecol Evol 6:17–21.

Smith DR, 1991b. Mitochondrial DNA and honey beebiogeography. In: Diversity in the genus Apis (SmithDR, ed). Boulder, CO: Westview Press; 131–176.

Smith DR and Brown WM, 1990. Restriction endonucle-ase cleavage site and length polymorphisms in mito-chondrial DNA of Apis mellifera mellifera and A. m. car-

nica (Hymenoptera: Apidae). Ann Entomol Soc Am 83:81–88.

Smith DR, Palopoli MF, Talyor BR, Garnery L, CornuetJ-M, Solignac M, and Brown WM, 1991. Geographicoverlap of two classes of mitochondrial DNA in Spanishhoney bees (Apis mellifera iberica). J Hered 82:96–100.

Smith DR, Slaymaker A, Palmer M, and Kaftanolgu O,1997. Turkish honey bees belong to the east Mediter-ranean mitochondrial lineage. Apidologie 28:269–274.

Received May 26, 1998Accepted August 25, 1999

Corresponding Editor: Robert Wayne

Classifying GenealogicalOrigins in HybridPopulations Using DominantMarkers

L. M. Miller

In hybrid studies, potential for error is highwhen classifying genealogical origins ofindividuals (e.g., parental, F1, F2) based ontheir genotypic arrays. For codominantmarkers, previous researchers have con-sidered the probability of misclassificationby genotypic inspection and proposed al-ternative maximum-likelihood approachesto estimating genealogical class frequen-cies. Recently developed dominant mark-er systems may significantly increase thenumber of diagnostic loci available for hy-brid studies. I examine probabilities ofclassification error based on the numberof dominant loci. As in earlier studies, I as-sume that only parental and first- and sec-ond-generation hybrid crosses betweentwo taxa potentially exist. Thirteen loci withdominant expression from each parentaltaxon (i.e., 26 total loci) are needed to re-duce classification error below 5% for F2

individuals, compared to 13 codominantloci for the same error rate. Use of loci insimilar numbers from both taxa most effi-ciently increases power to characterize allgenealogical classes. In contrast, classi-fication of backcrosses to one parentaltaxon is wholly dependent on loci from theother taxon. Use of dominant diagnosticmarkers may increase the power and ex-pand the use of maximum-likelihood meth-ods for evaluating hybrid mixtures.

Nason and Ellstrand (1993) and Epifanio andPhilipp (1997) addressed theoretical aspectsof using diagnostic codominant markers toestimate frequencies of genealogical classesin hybridized populations. Their work wasinspired by recognition of the great poten-tial for classification error when genealogi-cal origins of individuals are assigned based


on their genotypic arrays (genotypic inspec-tion) (Avise and van den Avyle 1984; Camp-ton 1990). When hybrids survive past thefirst generation (i.e., F1 hybrids successfullyreproduce), error arises because genotypicarrays of second- and later-generation cross-es overlap with those of parental and first-generation crosses. Assignment based ongenotype inspection can thus lead to incor-rect conclusions about the proportions ofvarious crosses (genealogical classes) in ahybrid mixture. Nason and Ellstrand (1993)considered this error and suggested the useof maximum-likelihood methods, which pro-duce unbiased estimates of hybrid class fre-quencies. Epifanio and Philipp (1997) ex-panded this work to quantify the extent ofclassification error based on the number ofdiagnostic codominant loci. With the adventof several new dominant marker systemshaving great potential for identifying diag-nostic loci, it is important to consider clas-sification error for the case of dominantmarkers.

Recently researchers have developedseveral new types of polymerase chain re-action (PCR)-based DNA marker systemsthat are typically scored assuming domi-nant inheritance [e.g., random amplifiedpolymorphic DNA (RAPD) (Welsh andMcClelland 1990; Williams et al. 1990), AFLP(Vos et al. 1995), inter-simple-sequence re-peats ( ISSR) (Gupta et al. 1994; Zietkiewiczet al. 1994), SINE-PCR (Greene and Seeb1997)]. Dominant markers are character-ized by the presence or absence of bandson a gel, with each band considered a lo-cus. The copy number of alleles producingthe band (one or two) cannot be deter-mined so that band-present homozygotescannot be distinguished from heterozy-gotes. This results in less genetic informa-tion per locus than codominant markerswhen applied to questions of populationgenetic structure, paternity, and hybridiza-tion (Fritsch and Rieseberg 1996).

The loss of information due to domi-nance is countered by positive attributesthat make dominant marker systems use-ful for hybrid analyses. Each techniqueproduces numerous bands, so that singlereactions can screen multiple loci for di-agnostic bands. By varying PCR primerswithin marker systems, and by combiningsystems, essentially unlimited numbers ofloci can be generated. Therefore dominantmarkers may provide for faster and moreefficient discovery of diagnostic loci com-pared to other techniques (e.g., allozymes,microsatellites, and introns). Diagnosticloci at varying taxonomic levels have beendocumented for RAPD (Crawford et al.

1993), AFLP (Biesmann et al. 1997), andISSR (Wolfe et al. 1998). Diagnostic lociwill be more prevalent for interspecies hy-brids, but these techniques should also in-crease the likelihood of finding sufficientloci for intraspecific hybrid studies (i.e.,subspecies, populations). For example,Williams et al. (1998) screened 17 RAPDprimers and found three of them that pro-duced 15 markers (bands) that distin-guish two subspecies of largemouth bass(Micropterus salmoides). In contrast, only2 of 28 allozyme loci were diagnostic forthese subspecies (Philipp et al. 1983).

Boecklen and Howard (1997) assessed theuse of both codominant and dominantmarkers in hybrid studies, but restrictedtheir analysis to repeated backcrossing toone parental taxon. I now consider the useof dominant markers under the model of Na-son and Ellstrand (1993), which includes allfirst- and second-generation hybrids. I focusprimarily on the issue of genealogical mis-classification addressed by Epifanio and Phi-lipp (1997). An understanding of the ap-proaches taken by these authors forcodominant markers is required first.

Codominant Markers

The model of Nason and Ellstrand (1993)assumes that a population consists of onlysix genealogical classes. Two parentaltaxa, A and B, and first- and second-gen-eration products of mating between them,produce the six classes: P1 and P2 (crosseswithin parental taxa), F1 (P1 3 P2), BP1 andBP2 (backcrosses F1 3 P1 and F1 3 P2, re-spectively), and F2 (F1 3 F1). No advanced-generation crosses exist [e.g., F3 (F2 3 F2)or BC-2 (BP1 3 P1)]. When taxa-specific di-agnostic alleles at codominant loci areidentified for two parental taxa, the mul-tilocus genotype of each individual in ahybrid population can be assigned to oneof seven mutually exclusive and exhaus-tive genotypic categories. These authorsconsidered the general case in which, atthe population level, loci have some diag-nostic alleles and some alleles shared byboth parental taxa. One genotypic cate-gory, called ‘‘ambiguous,’’ consists of in-dividuals that share alleles at all loci; thiscondition is not possible when at least onelocus is diagnostic (i.e., all alleles are taxaspecific). When diagnostic loci are pres-ent, only six categories are possible.

For codominant, diagnostic loci, Epifanioand Philipp (1997) described six genotypiccategories. Genotypic category A consists ofhomozygotes for taxon A alleles at all loci(typical of the P1 class); category B consists

of homozygotes for taxon B alleles at all loci(typical of the P2 class); category H consistsof heterozygotes for A and B alleles at allloci (typical of the F1 class); categories AIand BI consist of homozygotes for one, butnot both, taxon, and at least one heterozy-gote (typical of the BP1 and BP2 classes, re-spectively); and category S consists of atleast one homozygote for each taxon (re-stricted to the F2 class). All genotypic cate-gories except S can contain members ofmultiple genealogical classes; category S isthe only category restricted to a single classand acts as a signature for those F2 individ-uals (Epifanio and Philipp 1997). Using thecategory definitions above, Epifanio and Phi-lipp (1997) determined the expected distri-bution of category assignments for the sixgenealogical classes based on the number ofcodominant, diagnostic loci (see their Table2). This allowed them to quantify the ex-pected misclassification error when mixedhybrids are assigned to genealogical classesbased on genotype inspection.

Dominant Markers

With diagnostic dominant markers (i.e.,fixed for presence in one taxon and fixedfor absence in the other taxon), the ge-notypic categories described above mustbe redefined because heterozygous locicannot be distinguished from homozygousdominant-taxon loci. Therefore multilocusphenotypes, based on band presence orabsence, must be considered. For locipresent in taxon A (LA), individuals fromthe P1, F1, and BP1 classes produce bandsat all loci and a phenotypic category, callit presence (1), incorporates the codom-inant categories A, H, and AI. The P2 classdoes not produce bands at any loci and iscategorized as absence (o), correspondingto the codominant category B. Finally,both the BP2 and F2 class are typified byindividuals having at least one locus withbands and at least one without (1/o). Thiscategory corresponds to BI and S. As ob-served in the codominant case, BP2 and F2

can also produce some individuals with 1or o for all loci, creating a potential formisclassification. For loci present in taxonB (LB), category 1 incorporates codomi-nant categories B, H, and BI; category ocorresponds to A; and category 1/o in-cludes AI and S.

One can determine the expected pro-portion of each genealogical class havingmultilocus phenotypes characteristic ofphenotypic categories (1, o, or 1/o)based on the number of dominant, diag-nostic loci from a single taxon. A critical


Table 1. Expected proportion of each genealogical class (P1, P2,. . ., F2) having multilocus phenotypescharacteristic of phenotypic categories (A, B, . . ., S) for unlinked dominant loci with at least one locusfixed for presence in taxon 1 (LA) and one locus fixed for presence in taxon 2 (LB)

Phenotypiccategory

Genealogical class

P1 P2 F1 BP1 BP2 F2

A 1 (1/2)LB (3/4)L

A(1/4)LB

B 1 (1/2)LA (1/4)L

A(3/4)LB

H 1 (1/2)LB (1/2)L

A (3/4)LA(3/4)L

B

AI 1–2(1/2)LB (3/4)L

A(4LB23L

B21)/4LB

BI 1–2(1/2)LA (3/4)L

B(4LA23L

A21)/4LA

S [(4LA23L

A)/4LA][12(3/4)L

B]

Table 2. Probability of assigning the multilocusphenotype of an F2 individual to phenotypiccategory S based on the number of unlinkeddominant or codominant loci

Dominant

Number of loci

LA LB Total S

Codominant

Numberof loci

S

1 1 2 0.06 1 0.003 3 6 0.33 3 0.286 6 12 0.68 6 0.669 9 18 0.86 9 0.85

13 13 26 0.95 13 0.9516 10 26 0.9319 7 26 0.8622 4 26 0.6825 1 26 0.2511 Infinite Infinite 0.961 Infinite Infinite 0.25

conclusion is that when multiple domi-nant loci are all present in a single taxon,that is, all LA or all LB, there is not a one-to-one correspondence between genealog-ical classes and phenotypic categories‘‘typical’’ of those classes. An infinite num-ber of LA loci can do no better than cate-gorize P1, F1, and BP1’s as A, H, or AI (1),and BP2 and F2’s as BI or S (1/o). Also,there is no category that is restricted to asingle genealogical class, so that no indi-vidual can be definitively classified basedon its category (i.e., a signature).

When dominant loci from each taxonare combined, however, the one-to-onecorrespondence of categories to classesand the F2 signature of the S category arerecovered. Six mutually exhaustive phe-notypic categories can be defined: cate-gory A 5 all LA 1 and all LB o; categoryB 5 all LA o and all LB 1; category H 5all LA and LB 1; category AI 5 all LA 1and at least one LB 1 and at least one LBo; BI 5 at least one LA 1 and at least oneLA o and all LB 1; and category S 5 atleast one LA o and at least one LB o. Then,as for codominant loci, the expected fre-quencies of phenotypes in each genealog-ical class can be calculated based ontransmission probabilities from taxon Aand B assuming Mendelian inheritance ofunlinked loci (Table 1) (Nason and Ell-strand 1993). Note that only the F2 classcan contain individuals ‘‘absent’’ (o) forboth LA and LB loci, again making the S cat-egory a signature for F2 individuals.

Misclassifying GenealogicalOrigins: Dominant versusCodominant Markers

Epifanio and Philipp (1997) highlightedthe potential for error when assigning off-spring in a hybrid mixture to a genealogi-cal class solely by genotypic inspectionusing codominant markers. They specifi-cally examined the probability of misclas-sifying F2’s, the class most likely to be in-correctly assigned (Figure 1 in Epifanio

and Philipp 1997). I examined the powerof dominant markers to classify F2’s (Table2). Based on the probability of assigningan F2 individual to category S, and thuscorrectly concluding that it is an F2, thereis an approximate equivalence betweentwo dominant loci, one from each taxon,and a single codominant locus. Thirteenpairs of dominant loci (i.e., 26 total loci)are needed to categorize F2’s as S withprobability greater than .95, equaling thenumber of single codominant loci needed(Epifanio and Philipp 1997). With fewerloci, and lower power, pairs of dominantloci result in higher probabilities of cor-rect assignment than equivalent numbersof single codominant loci. This is becauseany two dominant loci from alternate taxacan categorize an individual as an S if theyare both phenotype o. No single codomi-nant locus is sufficient because two ho-mozygotes, one from each taxon, must beobserved to assign to category S. Thispower difference dissipates as the numberof loci increases (Table 2).

Efficiency (power per locus) of dominantmarkers to classify F2’s is best increased byhaving similar numbers of loci from eachtaxon (Table 2). With a single locus fromone of the taxa and an infinite number fromthe other, only 0.25 F2’s are expected to beplaced into category S. Even maintaining theinfinite number of loci from one taxon, 11are needed from the other to increase pow-er to greater than 0.95. Less extreme markerdifferences also demonstrate this principle(Table 2). Contrast this with the probabilityof assigning backcross classes, BP1 and BP2,into their ‘‘typical’’ categories, AI and BI. Ta-ble 1 shows that the categorization of back-crosses to one parental taxon depends en-tirely on the number of loci from the othertaxon (e.g., BP1 are categorized based on thenumber of LB) (see also Boecklen and How-ard 1997). This is because a backcross toone taxon will produce the 1 phenotype forall dominant loci from that taxa (Table 2);making the phenotypic distributions ofclasses P, F1, and BP indistinguishable. A lo-

cus from the other taxon is needed to sep-arate these classes.

Implications

The main emphasis of this article has beento address sources of misclassifying gene-alogical origins of individuals in hybridpopulations using dominant markers, inthe manner of Epifanio and Philipp (1997).Their key conclusion was that many co-dominant loci are needed to minimize mis-classification based on genotypic inspec-tion and assignment of individuals, even ifall loci are diagnostic. I have shown thatmore dominant than codominant loci areneeded and that it is important that locicome from both parental taxa if F2’s andbackcrosses to both parental taxa may bepresent. My results reinforce support forusing maximum-likelihood methods for es-timating class contributions (Nason andEllstrand 1993) as an alternative to assign-ment by genotypic inspection.

My results also provide the basis for ap-plying maximum-likelihood methods to di-agnostic dominant loci. Following the pro-cedures of Nason and Ellstrand (1993),individuals are first placed into genotypiccategories according to Table 1. The prob-ability of observing a genotypic categoryis then equated to its observed frequencyin the sample. Six linear equations for thegenealogical class estimates can then besolved, which consist of category frequen-cies and conditional probabilities of as-signing a multilocus phenotype (geno-type) of an individual in a given class to agiven genotypic category (based on Table1). These equations were shown to pro-duce unbiased estimates of genealogicalclass frequencies assuming that model as-sumptions were met, particularly that no


advanced generation crosses are present(Nason and Ellstrand 1993).

The maximum-likelihood method of Na-son and Ellstrand (1993) has limitations(Epifanio and Philipp 1997). First, genea-logical class frequencies in populations,rather than individual class membership,are estimated. For example, any one indi-vidual in genotype category A could be aP1, BP1, or F2. Therefore the common ques-tion, ‘‘Is this individual unhybridized, pastor present?’’ is not answered by this meth-od. Only those individuals of the F2 classthat fall into category S can be classifiedwith certainty. Second, the model cannotaccommodate crosses beyond the F2 gen-eration. Individuals from advanced gener-ations will have multilocus genotypes thatplace them in one of the six categories,thereby biasing estimates of genealogicalclass frequencies. An extreme example ofthis is continuous backcrosses to thesame parent (e.g., taxon A). By the fifthbackcross (BC-5), more than 66% of indi-viduals will likely be placed into parentcategory A using 13 codominant loci ordominant loci LB (Boecklen and Howard1997). The remaining individuals will al-most certainly be placed into AI. These au-thors showed that more than 70 markersare needed before the probability of as-signing a BC-5 to a hybrid class would ex-ceed .95, although under this framework itwould be incorrectly considered a first-generation backcross (BP1).

Despite these limitations, maximum-like-lihood methods for estimating genealogi-cal origins in mixed hybrid populationsshould still be useful. If hybridization is arecent event, for example, due to popula-tion transfers (e.g., stocking), or if suc-cessful hybridization is restricted to thefirst or second generation of interbreeding(Avise 1994 and references therein), thenthe model of Nason and Ellstrand (1993)is applicable. Furthermore, Nason andEllstrand (1993) showed that their modelwill often suggest the presence of ad-vanced-generation hybrids by producingimpossible estimates (i.e., frequencies ,0or .1.0) for some classes. In this case,class frequency estimates will be biasedbut advanced-generation hybrids will bedetected, which may often be the goal ofthe research. The potential to significantlyincrease the number of diagnostic loci us-ing dominant marker techniques will in-crease the power of maximum-likelihoodmethods and should advance the use ofmolecular data in hybrid studies.

From the Department of Fisheries and Wildlife, Univer-

sity of Minnesota, 1980 Folwell Ave., St. Paul, MN 55108.I thank Wansuk Senanan for her contributions to thisarticle and William Ardren, William Eldridge, RaymondNewman, and John Epifanio for comments on themanuscript. This work is the result of research spon-sored by the Minnesota Sea Grant College Programsupported by the NOAA Office of Sea Grant, U.S. De-partment of Commerce, project no. R/A-12, under grantno. NOAA-NA86-RG0033; journal reprint no. 459. TheU.S. Government is authorized to reproduce and dis-tribute reprints for government purposes, not with-standing any copyright notation that may appear here-on. This is article 984410022 of the MinnesotaAgricultural Experiment Station Scientific Journal Se-ries. Address correspondence to Loren M. Miller at theaddress above or e-mail: [email protected].


References

Avise JC, 1994. Molecular markers, natural history, andevolution. New York: Chapman & Hall.

Avise JC and van den Avyle MJ, 1984. Genetic analysisof reproduction of hybrid white bass 3 striped bass inSavannah River. Trans Am Fish Soc 113:563–568.

Beismann H, Barker JHA, Karp A, and Speck T, 1997.AFLP analysis sheds light on distribution of two Salixspecies and their hybrid along a natural gradient. MolEcol 6:989–993.

Boecklen WJ and Howard DJ, 1997. Genetic analysis ofhybrid zones: numbers of markers and power of reso-lution. Ecology 78:2611–2616.

Campton DE, 1990. Application of biochemical and mo-lecular markers to analysis of hybridization. In: Elec-trophoretic and isoelectric focusing techniques in fish-eries management (Whitmore DH, ed). Boca Raton, FL:CRC Press; 241–264.

Crawford DJ, Brauner S, Cosner MB, and Stuessy TF, 1993.Use of RAPD markers to document the origin of the inter-generic hybrid Margyracaena skottsbergi (Rosaceae) onthe Juan Fernadez Islands. Am J Bot 80:89–92.

Epifanio JM and Philipp DP, 1997. Sources for misclas-sifying genealogical origins in mixed hybrid popula-tions. J Hered 88:62–65.

Fritsch P and Rieseberg LH, 1996. The use of randomamplified polymorphic DNA (RAPD) in conservationgenetics. In: Molecular genetic approaches in conser-vation (Smith TB and Wayne RK, eds). New York: Ox-ford University Press; 54–73.

Greene BA and Seeb JE, 1997. SINE and transposon se-quences generate high-resolution DNA fingerprints,‘‘SINE prints,’’ that exhibit faithful Mendelian inheri-tance in pink salmon (Oncorhynchus gorbuscha). MolMar Biol Biotechnol 6:328–338.

Gupta M, Chyi Y-S, Romero-Severson J, and Owen JL,1994. Amplification of DNA markers from evolutionarilydiverse genomes using single primers of simple-se-quence repeats. Theor Appl Genet 89:998–1006.

Nason JD and Ellstrand NC, 1993. Estimating the fre-quencies of genetically distinct classes of individualsin hybridized populations. J Hered 84:1–12.

Philipp DP, Childers WF, and Whitt GS, 1983. A bio-chemical genetic evaluation of northern and Floridasubspecies of largemouth bass. Trans Am Fish Soc 112:1–20.

Vos P, Hogers R, Bleeker M, Reijans M, van de Lee T,Hornes M, Frijters A, Pot J, Peleman J, Kuiper M, andZabeau M, 1995. AFLP: a new technique for DNA fin-gerprinting. Nucleic Acids Res 23:4407–4414.

Welsh J and McClelland M, 1990. Fingerprinting ge-nomes using PCR with arbitrary primers. Nucleic AcidsRes 18:7213–7218.

Williams DJ, Kazianis S, and Walter RB, 1998. Use ofrandom amplified polymorphic DNA (RAPD) for iden-tification of largemouth bass subspecies and their in-tergrades. Trans Am Fish Soc 127:825–832.

Williams JGK, Kubelik AR, Livak KJ, Rafalski JA, andTingey SV, 1990. DNA polymorphisms amplified by ar-bitrary primers are useful as genetic markers. NucleicAcids Res 18:6531–6535.

Wolfe AD, Xiang Q-Y, and Kephart SR, 1998. Assessinghybridization in natural populations of Penstemon(Scrophulariaceae) using hypervariable intersimple se-quence repeat ( ISSR) bands. Mol Ecol 7:1107–1125.

Zietkiewicz E, Rafalski A, and Labuda D, 1994. Genomefingerprinting by simple sequence repeat (SSR)-an-chored polymerase chain reaction amplification. Gen-omics 20:176–183.

Received January 4, 1999Accepted September 14, 1999

Corresponding Editor: Bruce S. Weir

Is There Really NaturalSelection Affecting the lFrequencies (Long Hair) inthe Brazilian Cat Populations?

M. Ruiz-Garcia

The scientific literature on cat geneticscontains a presumed typical example ofnatural selection affecting l frequencies(long hair) in 16 Brazilian cat populations.It has been observed that the hotter andmore tropical the climate in Brazil, the low-er the values of l frequencies in the catpopulations. Nevertheless, this study ofsome new cat populations in Latin Ameri-ca showed that all of them, independentof the climate, had high or very high l fre-quencies. I postulate that an alternativemigrational-historical hypothesis existsthat explains the correlation between the lfrequencies and climate characteristics(which are correlated with the latitude)without using natural selection explana-tions concerning the appearance of the lallele in Brazil.

Seven of 12 genes coding for coat charac-ters such as color, tabby, and length, aswell as certain skeletal anomalies havebeen studied in many domestic cat popu-lations (Felis catus) worldwide (e.g., Ah-mad et al. 1980; Lloyd 1985; Lloyd andTodd 1989; Ruiz-Garcia 1991, 1994, 1997b;Ruiz-Garcia et al. 1995, 1998, 1999). Theseloci have significant spatial patterns in dif-ferent areas of the world (Ruiz-Garcia1994, 1997b), showing a close relationshipto historical and commercial human mi-grations (Ruiz-Garcia and Alvarez 1996;Todd 1977). Artificial human selection hasbeen shown to have had little influence onthe genetic profiles of the stray cat popu-lations studied in different cities. Clark(1975) showed, for instance, that the hu-


Table 1. Some mutant allele frequencies of European (Spain, Portugal, and Italy), Hispanic settlementsin the United States, and Latin American cat populations (of Spanish and Portuguese origin)

Populations

Locus

n O a tb d l S W

SpainBarcelona (1989) 709 0.16 0.70 0.27 0.27 0.14 0.27 0.004Granollers 158 0.17 0.71 0.16 0.08 0.21 0.23 0.003Girona 159 0.22 0.67 0.24 0.12 0.15 0.29 0.003L’Estartit 100 0.19 0.81 0.23 0.00 0.15 0.31 0.000Llansa 75 0.26 0.74 0.15 0.24 0.00 0.20 0.000R Castell (1989) 319 0.24 0.80 0.17 0.22 0.15 0.28 0.007R Castell (1994) 228 0.20 0.78 0.26 0.26 0.13 0.27 0.009Sitges 204 0.13 0.74 0.32 0.25 0.20 0.25 0.006Vilanova 118 0.20 0.72 0.22 0.23 0.09 0.35 0.005Tarragona 216 0.21 0.68 0.38 0.15 0.00 0.29 0.000Benidorm (1990) 207 0.23 0.70 0.55 0.16 0.00 0.28 0.005Alicante 335 0.23 0.78 0.50 0.24 0.00 0.31 0.006Murcia 171 0.18 0.74 0.62 0.08 0.08 0.38 0.012Cadiz 64 0.16 0.71 0.45 0.25 0.00 0.23 0.010Mahon (Balearics) 475 0.30 0.80 0.18 0.38 0.12 0.14 0.003Villacarlos (Balearics) 226 0.24 0.75 0.13 0.44 0.17 0.18 0.002Ciudadela (Balearics) 510 0.21 0.73 0.22 0.35 0.08 0.27 0.004P.Majorca (Balearics) 475 0.20 0.72 0.36 0.35 0.27 0.23 0.004Ibiza (Balearics) 273 0.24 0.76 0.30 0.23 0.19 0.29 0.010Vigo 228 0.23 0.80 0.36 0.14 0.19 0.35 0.004Santiago Compostela 120 0.24 0.75 0.40 0.16 0.16 0.37 0.000Tenerife (Canary) 146 0.29 0.80 0.32 0.34 0.25 0.41 0.003Puerto Cruz (Canary) 126 0.16 0.83 0.33 0.38 0.13 0.32 0.000

Portugal and IslandsLisbon 371 0.07 0.65 0.45 0.27 0.09 0.21 0.020Porto 256 0.14 0.77 0.43 0.30 0.23 0.29 0.010Terceira (Azores) 160 0.31 0.66 0.37 0.34 0.16 0.45 0.030Faial/Pico (Azores) 109 0.25 0.57 0.38 0.34 0.17 0.34 0.010S. Miguel (Azores) 159 0.23 0.63 0.50 0.34 0.11 0.29 0.000Madeira 133 0.18 0.78 0.35 0.41 0.00 0.39 0.010Mindelo (C Verdes) 206 0.12 0.77 0.60 0.19 0.00 0.34 0.000Praia (C Verdes) 168 0.23 0.83 0.47 0.11 0.00 0.44 0.010

ItalyRome 480 0.09 0.66 0.49 0.34 0.10 0.31 0.010Venice (1991) 145 0.10 0.56 0.27 0.34 0.19 0.20 0.014San Remo 148 0.04 0.58 0.48 0.32 0.29 0.27 0.000Rimini 518 0.13 0.68 0.38 0.41 0.22 0.26 0.012Riccione 130 0.11 0.59 0.44 0.42 0.27 0.28 0.008

Hispanic settlements in the United StatesDenver (Colorado) 286 0.20 0.84 0.26 0.38 0.35 0.29 0.010Lubbock (Texas) 265 0.31 0.79 0.36 0.33 0.44 0.24 0.000Dallas (Texas) 311 0.25 0.67 0.27 0.32 0.44 0.18 0.000Denton (Texas) 311 0.25 0.81 0.27 0.33 0.46 0.22 0.010Mineral Wells (Texas) 311 0.31 0.73 0.34 0.29 0.52 0.20 0.000Houston (Texas) 294 0.25 0.69 0.29 0.29 0.35 0.19 0.000Richmond (California) 107 0.19 0.77 0.27 0.33 0.35 0.28 0.030San Francisco (California) 195 0.27 0.79 0.33 0.32 0.36 0.31 0.030Humboldt County (California) 238 0.27 0.75 0.51 0.35 0.30 0.22 0.030

Hispanic settlements in Latin AmericaLos Mochis (Mexico) 141 0.31 0.71 0.39 0.24 0.31 0.32 0.010Mexico City (Mexico) 170 0.16 0.62 0.23 0.29 0.57 0.29 0.020Caracas (Venezuela) 164 0.13 0.79 0.32 0.10 0.33 0.33 0.009Willemstadt (Curacao) 151 0.14 0.84 0.19 0.19 0.28 0.34 0.020Havana (Cuba) 334 0.30 0.72 0.24 0.14 0.62 0.39 0.022Bogota (Colombia) 1105 0.19 0.86 0.18 0.35 0.34 0.21 0.009Ibague (Colombia) 147 0.24 0.82 0.11 0.37 0.32 0.23 0.003Bucaramanga (Colombia) 240 0.16 0.87 0.11 0.32 0.29 0.24 0.002Cali (Colombia) 258 0.23 0.84 0.18 0.47 0.33 0.32 0.005Pasto (Colombia) 210 0.20 0.82 0.10 0.37 0.41 0.29 0.000Santiago (Chile) 126 0.13 0.76 0.42 0.51 0.59 0.33 0.036Buenos Aires (1992) 295 0.27 0.82 0.31 0.45 0.40 0.28 0.021Buenos Aires (1996) 675 0.21 0.79 0.29 0.43 0.41 0.29 0.016

man taste for specific mutant colors wasnot reflected in the genetic profile of thecat population of Glasgow. Nevertheless, ithas been particularly difficult to establishexact geographical patterns of and influ-ences on one gene, l ( long hair). Todd etal. (1974) put forward that ‘‘ . . . popula-tions which are unarguably related in oth-er aspects, show great disparity in their lfrequencies.’’ At first glance one wouldthink that this allele would be favorablyselected for in cold climates, and con-versely would be negatively selected for inhot climates. It would therefore corre-spond to our expectations to find high fre-quencies of l in populations in very coldclimates, such as Leningrad [q(l) 5 0.64]and Alma Ata [q(l) 5 0.56] in the formerSoviet Union, or Inverness [q(l) 5 0.52] inScotland, for example. It would be para-doxical, however, that populations in plac-es such as Cyprus [q(l) 5 0.50], Jericho,Israel [q(l) 5 0.42], and Phoenix, Arizona[q(l) 5 0.51], with very high recorded tem-peratures, would have high frequencies ofl. They are, in fact, much higher than otherpopulations with notably colder climates,such as Poznan [q(l) 5 0], Bialowieza [q(l)5 0.21], and Wroclaw [q(l) 5 0] in Polandor in Iceland [q(l) 5 0.14]. This has moti-vated the study of this characteristic indifferent areas of the world by several in-vestigators. Lloyd (1983, 1985), for exam-ple, showed that a significantly negativecorrelation existed between the averageminimum temperatures in 35 populationsfrom the Atlantic coast of North Americaduring January (winter) and the frequen-cies of l (r 5 20.44). However, there wasno significant correlation (r 5 20.08) withrespect to the average maximum temper-ature in July (summer). Still, the most fa-mous example where the existence of nat-ural selection affecting l frequencies waspostulated was that described by Wata-nabe (1984) in Brazil. The author ob-served that the hotter and more tropicalthe climate, the lower the values of q(l) for16 cat populations studied in Brazil (r 520.95), concluding that l was strongly un-favorable in tropical climates. This findinghas been recognized as a clear example ofnatural selection affecting a genetic char-acter for a morphological trait in cat pop-ulations [see, e.g., Klein (1993), Lloyd(1987), Lloyd and Todd (1989), and Toddand Lloyd (1984), among others]. Howev-er, the analysis of other Latin Americancat populations has allowed me to postu-late an alternative hypothesis that wouldexplain the correlation between q(l) andclimate characteristics according to lati-

tude, leaving aside the possible existenceof selection in Brazil.

Materials and Methods

In order to demonstrate the absence ofnatural selection affecting the locus L in

cats in Brazil, the allele frequencies of sev-en loci controlling fur color, tabby, andlength were studied in the cities of La Ha-vana (n 5 334) in Cuba; Bogota (n 51105), Ibague (n 5 147), Bucaramanga (n5 240), Cali (n 5 257), and Pasto (n 5 210)in Colombia; Santiago (n 5 126) in Chile;


Table 1. Continued

Populations

Locus

n O a tb d l S W

BrazilPorto Alegre 489 0.16 0.68 0.26 0.28 0.27 0.31 0.020Curitiba 327 0.17 0.71 0.25 0.21 0.27 0.35 0.020Sao Paulo 1164 0.22 0.71 0.26 0.18 0.35 0.42 0.030Rio de Janeiro (1984) 1545 0.24 0.74 0.32 0.27 0.20 0.38 0.030Rio de Janeiro (1996) 232 0.14 0.74 0.23 0.30 0.31 0.37 0.012Belho Horizonte 859 0.18 0.71 0.26 0.25 0.22 0.36 0.020Campo Grande 512 0.19 0.64 0.27 0.20 0.24 0.29 0.020Brasilia 492 0.26 0.63 0.28 0.19 0.18 0.31 0.010Cuiaba 302 0.16 0.66 0.20 0.12 0.10 0.29 0.040Salvador 959 0.20 0.57 0.42 0.14 0.16 0.47 0.020Rio Branco 235 0.25 0.49 0.29 0.06 0.06 0.34 0.000J. Norte 503 0.33 0.60 0.24 0.21 0.05 0.39 0.000Teresina 994 0.18 0.69 0.33 0.26 0.01 0.35 0.030Fortaleza 1254 0.20 0.66 0.27 0.29 0.08 0.50 0.020S. Luis 1323 0.24 0.66 0.19 0.24 0.00 0.53 0.010Manaus 993 0.18 0.68 0.32 0.06 0.06 0.47 0.030Belem 909 0.22 0.65 0.25 0.14 0.05 0.42 0.010

a new sample of cats in Buenos Aires (n5 675) in Argentina; a new sample in Riode Janeiro (n 5 232) in Brazil; and twononoverlapping samples in the Canary Is-lands (one from Lloyd 1989, unpublished,in Tenerife City, n 5 146; one from Puertode la Cruz, Tenerife Island, n 5 126). Eachpopulation was extensively sampled tominimize local effects that could cause de-viations in the allele frequencies. The catssampled were alley cats, or ‘‘pseudowild.’’Previously reported samples that were in-cluded in this analysis were from Rio deJaneiro (Watanabe 1984) and Buenos Ai-res (Kajon et al. 1992). The phenotypes ofthe individuals were recorded from directobservation. The genetic nomenclatureused is in accordance with the Committeeon Standardized Genetic Nomenclature forCats (1968). The genetic characteristicsstudied included the sex-linked gene [O, o;Orange (epistatic to the observation of theA locus) versus non-orange] and the non-linked autosomal loci: A [A, a; agouti ver-sus non-agouti (epistatic to the observa-tion of the T locus)], T (t1, tb, Ta, stripedor mackerel tabby versus blotched tabbyversus Abyssinian tabby), D (D, d; nondi-lution versus dilution), L (L, l; short hairversus long hair), S (S, s; piebald whitespotting versus non-white spotting), andW [W, w; dominant white (epistatic to allthe other colors) versus normal color].For the characteristics of these genes seeRobinson (1977). The frequency of the al-lele orange was calculated using a differ-ential equation (Ahmad et al. 1980). Theautosomic recessive frequencies (q) werecalculated as the square roots of the ob-served phenotypic frequencies, while the

dominant frequencies (p) were taken as 12 q.

The genetic relationships between Bra-zilian populations reported by Watanabe(1981, 1984) were analyzed in this study,matching them against other populations.The populations were grouped in twoways: (1) The first group consisted of 40cat populations including the Latin Amer-ican (both of Spanish and Portuguese or-igins), southwestern United States, andtwo Canary Island populations. (2) A sec-ond group included 80 populations,among them the 40 populations of thegroup just described as well as a group ofNorth American cat populations from theUnited States and Canada (reported byLloyd and Todd 1989) of probable Britishorigin, and 20 European populations fromthe countries of origin of all these Ameri-can populations (Kajon et al. 1992; Lloydand Todd 1989; Ruiz-Garcia, 1990a–d, 1993,1994, 1997b). In order to analyze the rela-tionship between these cat populations,two kinds of analysis were carried out.The first was to obtain matrices of geneticdistances between pairs of populations.The genetic distances used were four: theNei standard genetic distance (Nei 1978),the Cavalli-Sforza and Edwards (1967)chord distance, the distance of Prevosti(1974), and the DA distance (Nei et al.1983). With these matrices, different den-drograms were constructed in order to ex-plain the overall genetic relationships be-tween all of these American and Europeanpopulations. The algorithms used werethe UPGMA (Sneath and Sokal 1973),WPGMA (using the recommendation ofPamilo 1990), COMPLETE, and neighbor-

joining (Saitou and Nei 1987). To deter-mine the reliability of the trees generated,three statistical methods were applied: theinterior branch and Rzhestsky and Nei(1992) tests (Li 1989), the Felsenstein(1985) bootstrap test, and the copheneticcorrelation coefficient (Sneath and Sokal1973). The trees that showed the best sta-tistics for reliability are shown here. All ofthe analyses were performed both includ-ing and excluding the locus L to determineits influence on the relationships found be-tween the populations studied. The sec-ond analysis was a canonical analysis ofpopulations. This separates groups ofpopulations along axes of high discrimi-nation power using the Mahalanobissquare distance, and is based on the ful-fillment of two hypotheses: (1) that thereis homogeneity between all covariancematrices corresponding to the populationgroups (maximum likelihood test), and (2)that the means of the k groups are signif-icantly different [Wilks’ L test, and the as-sociate value of the Fisher–Snedecor F testby means of the approximation of Rao(1951)]. Subsequently, a canonical trans-formation, the eigenvalues, the signifi-cance of the first canonical axes with theBartlett’s test, and the radius of the con-fidence regions (for a 90% level) were cal-culated. In this canonical population anal-ysis the following groups were employed:(1) Buenos Aires (two samples), (2) Mexi-co (three samples), (3) Venezuela and Cu-racao (two samples), (4) southern Brazil(six samples), (5) northern Brazil (10 sam-ples), (6) Colombia (five samples), (7) Ca-nary Islands (two samples), and separate-ly, La Havana, Santiago, and the secondsample from Rio de Janeiro.

Results and Discussion

Table 1 and Figures 1–3 show the basis forthe apparent association between the fre-quencies of l and climatic factors in Brazilwithout having to resort to the explana-tion of selection such as has been offeredup until now. Figure 1a shows the analysisbased on the application of the WPGMAalgorithm with Cavalli-Sforza and Edwardschord distance, including the L locus for40 populations. Figure 1b shows the treederived from the neighbor-joining methodwith DA distance, including the L locus for40 populations. Figure 1c shows the resultfrom the COMPLETE algorithm with Ca-valli-Sforza and Edwards chord distance,without the L locus for 40 populations.They show clearly that upon analyzing thegenetic relationships between the Latin


Figure 1. (a) WPGMA phenetic analysis from Cavalli-Sforza and Edwards (1967) chord distance of Hispanic settlements in the United States, Hispanic America, two CanaryIsland, and Brazilian cat populations (40 populations) with the inclusion of the L locus. Cophenetic correlation coefficient, r 5 0.71; approximate Mantel t test: t 5 8.79, P ,.0000; out of 1,000 random permutations: one-tail probability is p[random Z . observed Z] 5 0.001. (b) Neighbor-joining tree with DA distance (Nei et al. 1983) of 40populations with the inclusion of the L locus. The numbers in the figure are the bootstrap (1,000) percentages. (c) COMPLETE phenetic analysis from Cavalli-Sforza andEdwards (1967) chord distance of 40 populations without the inclusion of the L locus. Cophenetic correlation coefficient, r 5 0.69; approximate Mantel t test: t 5 7.17, P ,.0000; out of 1,000 random permutations: one-tail probability is p[random Z . observed Z] 5 0.001. The dendrograms shown are those with the better cophenetic correlationcoefficients, better percentages of Felsenstein bootstraps, and better Rzhestsky and Nei (1992) statistics.


Figure 2. Canonical analysis of populations. The first and second axes explain 90.09% of the variation.

American cat populations, the Brazilianpopulations do not form a homogeneousgroup. A group of Brazilian populationswas observed which showed more of a ge-netic similarity to the Hispanic popula-tions, such as Caracas (Venezuela), Wil-lemstadt (Curacao), and Los Mochis(Mexico), than to the other group of Bra-zilian populations. A second group hadmarked differences from the first, and hadno particular resemblance to the Hispanicpopulations. The composition of thesetwo groups of Brazilian populations is notgeographically random, however. Thegroup resembling the Hispanic popula-tions is located in southern Brazil (PortoAlegre, Curitiba, Sao Paulo, and Rio de Ja-neiro). These populations have the high-est l frequencies in Brazil (0.22–0.35). Onthe contrary, the other group of Brazilianpopulations was made up of coastal pop-ulations from northern Brazil and extend-ed inland into the Amazon regions, namelySalvador, Sao Luis, Rio Branco, J. Norte,Teresina, Fortaleza, Manaus, and Belem,and is characterized by very low or null lfrequencies (0–0.16). Populations fromCampo Grande, Brasilia, Bello Horizonte,and Cuiaba are found to cluster differentlywith both groups with regard to the algo-rithmic techniques and genetic distancesemployed.

The same analysis excluding the L locusshows a similar perspective, although it isslightly less clear. The northern Brazilianpopulations (Salvador, Manaus, Rio Bran-

co, Sao Luis, Fortaleza, and J. Norte) areless related to the Hispanic American pop-ulations than are the southern Brazilianpopulations (particularly Porto Alegre,Curitiba, and the two samples from Rio deJaneiro). A canonical analysis of popula-tions is shown in Figure 2. This analysisshowed a Wilks L 5 0.0002 and F 5 4.63with 70 and 100 df, being F 5 1.43 with a5 0.05. Consequently the hypothesis thatthe representative group means are equalwas rejected as expected. The two first ca-nonical axes explained 90.09% of the vari-ability. All the Hispanic American groupswere clearly related. The southern Brazil-ian group and the new sample from Rio deJaneiro were also highly related to the His-panic American groups. To the contrary,the northern Brazilian and Amazon groupswere isolated from the other groups ana-lyzed.

The main point which allows a newnonselectionist hypothesis of an historic-migrational character is the following.Ruiz-Garcia (1997a,c) and Ruiz-Garcia etal. (1998, 1999) observed that a constantfeature of the Hispanic American cat pop-ulations was the high, or very high, l fre-quencies [e.g., Bogota (0.34), Buenos Ai-res (0.41), Mexico City (0.57), Santiago,Chile (0.59), and La Havana (0.62)]. Thesevalues are much higher than those foundin Spanish populations, mostly with val-ues of q(l) between 0 and 0.20 (Ruiz-Gar-cıa 1990c,d, 1991, 1994, 1997b). Since inthe entire geographic range of the His-

panic American cat population, from Cal-ifornia, Colorado, and Texas, to Argentinaand Chile (more than 7000 km), the cli-matic characteristics are extremely di-verse, such high and constant q(l) valuescannot be attributed to the action of nat-ural selection, either in favor or against,after the formation of the original Hispan-ic American populations. Neither couldgenetic drift explain the systematic oc-currence of such high l frequencies in allof the Hispanic American populations.The Spaniards have a great admirationfor long-haired cats, a trait with low fre-quencies in Spain (Ruiz-Garcıa M, unpub-lished observations). It is quite likely thatduring the period when the Americaswere being colonized, the l frequencieswere lower still. A migrational selectioncould therefore have occurred due to thenovelty of this character ( Todd 1977,1978). This hypothesis is much more par-simonious than the existence of humanselection a posteriori at a time when thepopulations in question in Latin Americahad become important from a demo-graphic point of view. As shown by An-derson and Jenkins (1979), Morrill andTodd (1978), and Ruiz-Garcıa (1991), oncethe human population of a locality ap-proaches 30,000, the cat population maybe large enough to be refractory againstchanges in the allele frequencies.

The populations from southern Brazilresembled the Hispanic populations morethan those of the rest of Brazil. This couldhave originated from the establishment ofa gene flow of a certain magnitude be-tween those Brazilian populations andpopulations of Spanish origin near thesouthern frontier of Brazil. The commercebetween southern Brazil and the Hispaniccolonies in Uruguay, Paraguay, and thearea of Rio de La Plata in Argentina wasvery intense from 1713 onward. In fact, thePortuguese succeeded in setting up somecommercial colonies on the Rio de La Pla-ta. For instance, the first explorer in cur-rent Paraguay was the Portuguese AlejoGarcıa in 1525 searching for the ‘‘SilverMountain.’’ Later the area was conqueredby the Spaniards Juan de Salazar y Espi-nosa, Alvaro Nunez de Vaca, and DomingoMartinez de Irala. Juan de Salazar foundedAsuncion in 1537, and coming from Bue-nos Aires, Alvaro Nunez and DomingoMartınez founded small colonies in thesouth of today’s Brazil. In 1588 the Span-iard Jesuits founded numerous ‘‘Missions’’or ‘‘Reducciones,’’ where they congregat-ed hundreds of Guaranı Indian familiesand lodged Spaniard colonists. They trav-


Figure 3. (a) UPGMA phenetic analysis with DA distance with 80 populations with the inclusion of the L locus.Cophenetic correlation coefficient, r 5 0.62; approximate Mantel t test: t 5 32.66, P , .0000; out of 1,000 randompermutations: one-tail is p[random Z . observed Z] 5 0.001. The numbers in the figure are the bootstrap (1,000)percentages. (b) WPGMA phenetic analysis from Prevosti (1974) genetic distance of 80 populations without theinclusion of the L locus. Cophenetic correlation coefficient, r 5 0.68; approximate Mantel t test: t 5 25.81, P ,.0000; out of 1,000 random permutations: one-tail is p[random Z . observed Z] 5 0.001. The dendrograms shownare those which offered better cophenetic correlation coefficients, better Felsenstein bootstraps, and better Rzhes-tsky and Nei (1992) statistics.

eled throughout the Paraguay and Paranarivers as far as the frontier of southernBrazil. The tremendous ability of the Guar-anı Indians in various jobs encouraged thePortuguese to enslave entire populations,which were made to travel into southernBrazil (1690–1767). Also, numbers of poorPortuguese colonists in Paraguay and Uru-guay followed the Spanish Missions alongtheir way.

When the Jesuits were expelled from theSpanish Empire in 1767 the majority ofGuaranı Indian families, poor Portuguesecolonists, and one-sixth of the Spanish col-onists in Paraguay emigrated to the stateof Sao Paulo (southern Brazil) to work onthe rice and maize plantations and herd-ing cattle. In Uruguay there was also a re-lationship between the Spanish and thePortuguese populations. The first settle-ments in Uruguay were established by thePortuguese in Colonia in 1680. Later Spainestablished another colony in this area,Montevideo in 1726, and the area was fi-nally controlled by the Spanish in 1777. In1811 and 1816, the Portuguese invaded alarge portion of Uruguay and relations be-tween this country and southern Brazilhave been very important ever since.There was another Hispanic–Portugueseconnection in southern Brazil: In 1680–1710, a small and as yet undeveloped har-bor, Rio de Janeiro, started to gain impor-tance. The finding of gold was of keyimportance to this little harbor, and manySpanish ships arrived from the Canary Is-lands. By 1763 Rio de Janeiro had growninto a very important city. The Canary Is-lands were and remain a key part of thecommercial routes between Spain and Lat-in America. During the last decades of thecentury, more than 25,000 ships sailed tothese islands.

One more point is very important tobetter explain the Spanish relationshipwith Rio de Janeiro. In 1650 Potosı (Boliv-ia) was the second largest city of the West-ern World (160,000 inhabitants), after Lon-don. The reason was the mining of silver.The Spaniards arrived in Potosı followingthis route: southern Spain–Canary Islands–Dominican Republic–Panama (Puerto deDios, Porto Belo, and Panama City)–Gua-yaquil (Ecuador)–Peru (Callao and Lima).Once the silver had been loaded in Potosı,one of the most important routes back toSpain was through Bolivia (previouslycalled Alto Peru)–northern Argentineandeserts–the Argentinean cities of Salta, Ju-juy, Cordoba, Tucuman, La Plata (in Span-ish, silver) and finally, to the Port of Bue-nos Aires. All of these cities were founded


Figure 3. Continued.


during the years of the silver trade. Manyof the ships carrying silver sailed fromBuenos Aires to Spain via Rio de Janeiro(Fernandez de Oviedo 1944; Perrottet1990). It could have been that this intensecontact with Hispanic populations, wherethe frequencies of l were high was gener-ated by migrational selection, based onnovelty, from their original populations inSpain, promoting an increase in the fre-quencies of l in the populations of south-ern Brazil.

In northern Brazil the process of colo-nization was, in contrast, very different.Discovered by the Portuguese Pedro Al-vares Cabral in 1500, the most importantpopulations founded were located innorthern Brazil, with Salvador de Bahıabeing the first capital of Brazil (one of thecat populations studied). The importantcrop behind this development was sugar,for which the shoots for the plantationshad been brought from the Madeira Is-lands (African coast). This activity wassustained for nearly two centuries, duringwhich southern Brazil remained forgotten.The populations of the north coast and in-land Brazil were not affected by the His-panic influence, since their q(l) frequen-cies are null or very low. These Brazilianpopulations are therefore probably consti-tuted almost exclusively of Atlantic Por-tuguese influences, coming from the Is-lands of Madeira, Cape Verde, and theAzores, where the l frequencies are null orvery low. Alternatively they could havetheir origin in some continental Portu-guese populations which have not yetbeen studied.

The tree displayed in Figure 3a showsthe analysis based on the application ofthe UPGMA algorithm with DA distance,including the L locus, for 80 populations,and Figure 3b shows the correspondingWPGMA algorithm with Prevosti dis-tance, excluding the L locus, for the 80populations, revealing that some Spanishpopulations were mainly clustered withthe populations from southern Brazil, es-pecially some Catalan populations suchas Barcelona, Sitges, and Girona, whichhad previously been postulated to beSpanish populations with probable an-cient genetic profiles ( Kajon et al. 1992;Ruiz-Garcia 1988, 1990a–d, 1991, 1994).The southern Brazilian populations aretherefore more similar to certain Spanishpopulations than to recently sampledPortuguese populations such as Lisbonand Porto, confirming the proposed hy-pothesis.

From the historical and genetic evi-

dence gathered so far, the apparent cor-relation between the frequencies of l andgeographic latitude in Brazil in fact re-veals a geographical correlation with themigrational and historical factors thatwere involved in the formation of the co-lonialsettlements in Brazil. This correla-tion there-fore does not result from theoccurrence of natural selection, but rath-er from the particular location of the mi-grational and historical relationships,which by chance, in this case, coincidewith the location, according to latitude,of the Brazilian populations and their ge-netic constitution.

One more group of results supports theconclusion that natural selection doesnot necessarily act against the l allele intropical populations: data for the l allelein the cat population of La Havana(Cuba), a place of tropical climate, is oneof the highest worldwide (0.62). Also,data from other recently sampled tropicalCaribbean populations, such as Santo Do-mingo [Dominican Republic, q(l) 50.534]; Veracruz [Mexican Carib, q(l) 50.472], and Acapulco [Mexico, q(l) 50.441], have q(l) values greater than 0.40,which could not be expected under theselection hypothesis (Ruiz-Garcia M, un-published data). Additional studies onthe subject would, however, be necessaryin order to explain why there is appar-ently no migrational selection favoring lin the case of the Portuguese, as seemsto have occurred for the Spanish migra-tion. It could have been that the l allelewas not systematically present in somePortuguese cities which were the sourcesof cats during the colonization of theAmericas. Another possibility would bethat in Portugal there were at least twodifferent gene pools, one very similar oridentical to certain Spanish populationspostulated to be ancient, and which werethe origin of southern Brazilian popula-tions studied, and another(s) of differentcharacteristics, which might have beenthe origin of certain Portuguese islanderand northern and Amazonian Braziliancat populations. Only a thorough study ofthe Portuguese cat populations would al-low a more precise historical hypothesis.

I conclude that the l allele can in fact befavorably selected for in cold or very coldclimates (Lloyd 1983, 1985), but its behav-ior is neutral in temperate, hot, or veryhot climates, such as the tropics.

From Unidad Genetica (Genetica de Poblaciones-Biol-ogıa Evolutiva), Departamento de Biologıa, PontificiaUniversidad Javeriana, Cra 7a No 43-82, Bogota DC, Co-lombia, and CIGEEM, Barcelona, Spain. The author

thanks Diana Alvarez (Bogota DC, Colombia) and Drs.A. Kajon and S. Diaz for their help. This study was par-tially supported by the Convenios no. 139-94 and no.140-96 (Decreto 1742 de 1994) between COLCIENCIASand the author. Address correspondence to the authorat the address above or e-mail: [email protected].


References

Ahmad M, Blumenberg B, and Chaudhary MF, 1980. Mu-tant allele frequencies and genetic distance in cat pop-ulations of Pakistan and Asia. J Hered 71:323–330.

Anderson MM and Jenkins SH, 1979. Gene frequenciesin the domestic cats of Reno, Nevada: confirmation ofa recent hypothesis. J Hered 70:267–269.

Cavalli-Sforza LL and Edwards AWF, 1967. Phylogeneticanalysis: models and estimation procedures. Evolution 21:550–570.

Clark JM, 1975. The effects of selection and human pref-erence on coat colour gene frequencies in urban cats.Heredity 35:195–210.

Committee on Standardized Genetic Nomenclature forCats, 1968. Standardized genetic nomenclature for thedomestic cat. J Hered 59:39–40.

Felsenstein J, 1985. Confidence limits on phylogenies:an approach using the bootstrap. Evolution 39:783–791.

Fernandez de Oviedo J, 1944. Historia general y naturalde las Indias. Ed. Guaranıa. Asuncion.

Kajon A, Centron D, and Ruiz-Garcia M, 1992. Gene fre-quencies in the cat population of Buenos Aires, Argen-tina, and the possible origin of this population. J Hered83:148–152.

Klein K, 1993. Population genetics and gene geography.In: Genetikii koshkii (Ruvinki A and Borodin PM, eds).Moscow: Russian Academy of Science; 118–150.

Li WH, 1989. A statistical test of phylogenies estimatedfrom sequence data. Mol Biol Evol 6:424–435.

Lloyd AT, 1983. Population genetics of domestic cats(Felis catus L.) in New England and the Canadian Mar-itime Provinces: an investigation of the historical im-migration hypothesis (PhD dissertation). Boston: Bos-ton University.

Lloyd AT, 1985. Geographic distribution of mutant al-leles in domestic cat populations of New England andthe Canadian Maritimes. J Biogeog 12:315–322.

Lloyd AT, 1987. Cats from history and history from cats.Endeavour 11:112–115.

Lloyd AT and Todd NB, 1989. Domestic cat gene fre-quencies. A catalogue and bibliography. Newcastleupon Tyne: Tetrahedron Publications.

Morrill RB and Todd NB, 1978. Mutant allele frequen-cies in the domestic cats of Denver, Colorado. J Hered69:131–134.

Nei M, 1978. Estimation of average heterozygosity andgenetic distance from a small number of individuals.Genetics 89:583–590.

Nei M, Tajima F, and Tateno Y, 1983. Accuracy of esti-mated phylogenetic trees from molecular data. Genefrequency data. J Mol Evol 19:153–170.

Pamilo P, 1990. Statistical tests of phenograms basedon genetic distances. Evolution 44:689–697.

Perrottet T, 1990. South America. Singapore: HoferPress.

Prevosti A, 1974. La distancia genetica entre pobla-ciones. Misc. Alcobe. Publicacions de l’Universitat deBarcelona; 109–118.

Rao CR, 1951. Advanced statistical methods in biomet-ric research. Darien, CT: Hafner Publishing.

Robinson R, 1977. Genetic for cat breeders. Oxford:Pergamon Press.


Ruiz-Garcia M, 1988. Frecuencias alelicas mutantes enuna poblacion de gatos domesticos urbanos (Barce-lona) y en una poblacion de gatos rurales (Castelldefelsrural) en Cataluna, Espana. Genet Iber 40:157–187.

Ruiz-Garcia M, 1990a. Frecuencias alelicas en la pobla-cion de gatos domesticos de la isla de Menorca (Bal-eares): diferentes modelos de evolucion colonizadora.Evol Biol 4: 307–342.

Ruiz-Garcia M, 1990b. Frecuencias alelicas en la pob-lacion de gatos domesticos de Palma de Mallorca e Ibi-za y relaciones geneticas con otras poblaciones de ga-tos Europeos y Norteafricanos. Evol Biol 4:189–216.

Ruiz-Garcia M, 1990c. Mutant allele frequencies in do-mestic cat populations in Catalonia, Spain, and geneticrelationships between Spanish and English colonial catpopulations. Genetica 82:209–214.

Ruiz-Garcia M, 1990d. Mutant allele frequencies in do-mestic cat populations on the Spanish Mediterraneancoast and genetic distances from other European andNorth African cat populations. Genetica 82:215–221.

Ruiz-Garcia M, 1991. Mas sobre la genetica de pobla-ciones de Felis catus en la costa Mediterranea Espan-ola: un analisis de la estructura genetica de las pobla-ciones naturales de gatos. Evol Biol 5:227–283.

Ruiz-Garcia M, 1993. Analysis of the evolution and ge-netic diversity within and between Balearic and Iberiancat populations. J Hered 84:173–180.

Ruiz-Garcia M, 1994. Genetic profiles from coat genesof natural Balearic cat populations: an eastern Medi-terranean and North African origin. Genet Sel Evol 26:39–64.

Ruiz-Garcia M, 1997a. Caracterısticas geneticas de laspoblaciones de gatos domesticos (Felis catus) en lasAmericas. I. Estructura espacial inducida por coloni-zacion. Brazil J Biol (in press).

Ruiz-Garcia M, 1997b. Genetic relationships amongsome new cat populations sampled in Europe: a spatialautocorrelation analysis. J Genet 76:1–24.

Ruiz-Garcia M, 1997c. Perfiles geneticos de las pobla-ciones de gatos domesticos de La Habana (Cuba) y deBogota (Colombia) y posibles origenes Europeos deesas poblaciones. I: no existencia de paralelismo conel modelo colonizador britanico. Misc Zool (in press).

Ruiz-Garcia M and Alvarez D, 1996. The use of the do-mestic cat as an extragenic marker of the historical andcommercial human movements. Brazil J Genet 19:184.

Ruiz-Garcia M, Barrera MI, and Alvarez D, 1998. Geneticrelationships between Caribbean and South-Americancat populations: confirmation of the ‘‘Historical Migra-tion Hypothesis.’’ Genet Sel Evol (submitted).

Ruiz-Garcia M, Campos HA, Alvarez D, Kajon A, andDiaz S, 1999. Mutant allele frequencies of cat popula-tions in Latin America, Havana (Cuba), Bogota (Colom-bia), Ibague (Colombia), Santiago (Chile) and BuenosAires (Argentina): genetic relationships with otherAmerican and European cat populations. J Hered (sub-mitted).

Ruiz-Garcia M, Ruiz S, and Alvarez D, 1995. Perfiles ge-neticos de poblaciones de gatos domesticos (Felis ca-tus) de la provincia de Girona (Catalunya, NE, Espana)y posibles relaciones geneticas con otras poblacioneseuropeas occidentales. Misc Zool 18:169–196.

Rzhestsky A and Nei M, 1992. A simple method for es-timating and testing minimum-evolution trees. Mol BiolEvol 9:945–967.

Saitou N and Nei M, 1987. The neighbor-joining meth-od: a new method for reconstructing phylogenetictrees. Mol Biol Evol 4:406–425.

Sneath PH and Sokal RR, 1973. Numerical taxonomy.San Francisco: W. H. Freeman.

Todd NB, 1977. Cats and commerce. Sci Am 237:100–107.

Todd NB, 1978. An ecological, behavioural genetic mod-el for the domestication of the cat. Carnivore 1:52–60.

Todd NB and Lloyd AT, 1984. Mutant allele frequencies

in the domestic cats of Portugal and the Azores. J He-red 75:495–497.

Todd NB, Robinson R, and Clark JM, 1974. Gene fre-quencies in domestic cats of Greece. J Hered 65:227–231.

Watanabe MA, 1981. Mutant allele frequencies in thedomestic cats of Sao Paulo, Brasil. Carnivore GenetNewslett 4:168–177.

Watanabe MA, 1984. Estudo populacional da cor de pe-lagem de gato domestico (Felis catus L.) em dezesseislocalidades do Brasil (PhD dissertation). Sao Paulo:Universidade de Sao Paulo.

Corresponding Editor: Stephen J. O’Brien

Improved Estimation of theProportion of Triploids inPopulations with Diploid andTriploid Individuals

M. S. Ridout

We consider the estimation of the propor-tion of triploids in populations of plants oranimals in which diploid and triploid indi-viduals coexist, using data from electro-phoretic analysis of isozyme or microsa-tellite markers. Individuals that have threedistinct alleles at a locus are unambigu-ously triploid. However, other individualscannot be classified with certainty as dip-loid or triploid, unless allelic dosage canbe determined reliably. This is impossiblefor microsatellite markers, and for manyisozyme markers. We therefore present amaximum likelihood method of estimatingthe proportion of triploids based only onthe presence or absence of different al-leles.

Populations that include both diploid andtriploid individuals occur in various plantand animal species. Estimation of the pro-portion of triploid individuals within suchmixed populations is important for popu-lation genetic studies and for ecologicalstudies (Krieger and Keller 1998). One ap-proach to this problem is to use pheno-typic data from electrophoretic analysis ofisozyme or microsatellite markers. Thesetechniques identify the distinct alleles thatare present. However, determination of al-lelic dosage is generally impossible withmicrosatellite markers. Dosage can some-times be determined for isozyme markers,but is often unreliable. Clearly an individ-ual that has three distinct alleles at a locusis necessarily triploid. However, if dosagecannot be determined reliably, an individ-ual that has only two distinct alleles maybe diploid, or it may be a triploid that hastwo identical alleles. Similarly, an individ-

ual that has only one distinct allele maybe either diploid or triploid. Thus unam-biguous classification of individuals asdiploid or triploid is impossible in general.Krieger and Keller (1998) showed that it ispossible, nonetheless, to estimate the pro-portion of triploids from phenotypic data,provided that the same alleles occur withthe same frequencies in diploids and trip-loids and provided that genotype frequen-cies are in Hardy–Weinberg equilibrium.

In this article we demonstrate that thereare advantages in using likelihood meth-ods to address this problem. We show thatthe maximum likelihood estimate of theproportion of triploids is almost alwayspreferable to the estimator of Krieger andKeller (1998). Another advantage of thelikelihood method is that it leads naturallyto a test of one of the key assumptions,namely that the alleles occur at the samefrequency in diploids and triploids.

Triploidy Estimates

Single Locus with Three AllelesWe consider a locus with three alleles (A,B, and C) in a mixed population of diploidsand triploids in which the proportion oftriploids is t. The allele frequencies, a, b,and c, are assumed to be the same for dip-loids and triploids and genotype frequen-cies are assumed to be related to allele fre-quencies according to the Hardy–Weinberglaw. There are seven phenotypic classes, asshown in Table 1 [identical to Table 1 inKrieger and Keller (1998)]. ABC individualsare necessarily triploid and may thereforebe termed ‘‘overt triploids.’’ Other individ-uals may be diploid or triploid.

The method of Krieger and Keller (1998)involves the following steps

1. Estimate allele frequencies a, b, and con the assumption that all individuals,apart from the overt triploids, are diploid.

2. Estimate t by equating the observednumber of overt triploids (n7) to the ex-pected number, based on the current es-timates of a, b, and c.

3. Calculate revised estimates a, b, andc, based on the current estimate of t.

Steps 2 and 3 are alternated until the esti-mates converge, or until the estimate of texceeds one, in which case the estimate istaken as one. Krieger and Keller (1998)show that convergence is usually achievedin a few iterations.

An alternative method of estimating t isby maximum likelihood. Given a sample ofN independent individuals, the distribu-tion of the numbers of individuals in the


Table 1. Frequencies of phenotypes and genotypes for a locus with three alleles

Phenotype AA AB AC BB BC CC ABC

Genotype 2n AA AB AC BB BC CC —Frequency a2(1 2 t) 2ab(1 2 t) 2ac(1 2 t) b2(1 2 t) 2bc(1 2 t) c2(1 2 t) —Genotype 3n AAA AAB ABB AAC ACC BBB BBC BCC CCC ABCFrequency a3t 3a2bt 3ab2t 3a2ct 3ac2t b3t 3b2ct 3bc2t c3t 6abct

Figure 1. Relative efficiency (RE, %) of the Krieger and Keller (1998) estimator of t compared to the maximumlikelihood estimator for different values of t and for different sample sizes [N 5 50 (solid circles), N 5 100 (opencircles), N 5 250 (solid triangles)] for two sets of allele frequencies. Each relative efficiency value is based on datafrom 5,000 simulations.

different phenotypic classes is multinomi-al. Ignoring combinatorial terms that areirrelevant to parameter estimation, thelog-likelihood function is

7

L(a, b, t) 5 n ln(p ),O i ii51

where ni denotes the number of individu-als in phenotypic class i (i 5 1,…,7), pi isthe probability that a randomly selectedindividual belongs to the ith phenotypicclass, and where the allele frequency c iseliminated from the likelihood by writingc 5 1 2 a 2 b. Numerical methods areneeded to maximize this function andthereby obtain maximum likelihood esti-mates of a, b, and t. We used an imple-mentation of the Nelder–Mead simplex al-gorithm (Nelder and Mead 1965). Analternative approach would be to use theEM algorithm (e.g., Lange 1997, chap. 2).

To compare the Krieger and Keller (KK)estimator of t with the maximum likeli-hood (ML) estimator, we simulated data-sets with N 5 50, 100, or 250 and with var-ious sets of allele frequencies. Fivethousand datasets were simulated foreach combination of parameter values.Krieger and Keller (1998) note that theirestimator of t is unbiased when the geno-type frequencies within the sampled dataare in exact Hardy–Weinberg equilibrium.However, this condition will seldom bemet, because of sampling variation, evenwhen the population frequencies are inHardy–Weinberg equilibrium. Therefore ingeneral the estimator is biased. For ex-ample, with N 5 100, a 5 b 5 c 5 1/3 andt 5 0.5, the mean of the 5000 KK estimateswas 0.508, a bias of 0.008 (SE 5 0.0020).Although the bias is small, there is strongstatistical evidence that it is not zero. Forthese parameter values, the estimatedbias of the ML estimator was 0.010 (SE 50.0018). Thus the ML estimator is also bi-ased. However, these biases are small inrelation to the standard deviations of theestimates (0.143 for KK, 0.126 for ML).Larger biases arose for some other com-binations of parameter values (see below),but in most instances the bias remainedsmall in relation to the variability of theestimates.

The efficiency of the KK estimator rela-tive to the ML estimator (RE) was calcu-lated using the formula

MSE(ML estimator)RE(%) 5 100 3 ,

MSE(KK estimator)

where MSE denotes the mean squared er-ror, the average squared difference be-


Table 2. Sample sizes needed to obtain the sameprecision as KK or ML estimators whenunambiguous classification of diploids andtriploids is possible

t

a 5 1/10, b 5 1/10,c 5 4/5

KK ML

a 5 1/3, b 5 1/3,c 5 1/3

KK ML

0.1 8 8 51 520.3 5 9 40 470.5 4 8 30 400.7 3 8 19 310.9 2 6 10 17

Values are rounded to the nearest integer. The preci-sion of the KK and ML estimates is based on the sim-ulated mean square error in samples of size 250.

Table 3. Phenotype frequencies simulated assuming different allele frequencies in diploids andtriploids

Phenotype AA AB AC BB BC CC ABC

Observed 26 86 119 15 108 95 51Fitted, model 1 27.8 66.9 139.0 21.1 120.1 78.7 46.4Fitted, model 2 25.7 84.0 120.4 18.0 102.8 96.3 52.8

The fitted frequencies are from two models. Model 1 is the standard model, assuming equal allele frequencies indiploids and triploids. Model 2 allows the frequencies to differ.

Table 4. Parameter estimates for two models fitted to the simulated data shown in Table 3

Parameter t a b c

Model 1 0.462 0.288 0.254 0.458(0.0543) (0.0138) (0.0132) (0.0153)

Model 2 0.489 Diploid 0.198 0.209 0.593(0.0732) (0.0864) (0.0541) (0.0507)

Triploid 0.401 0.304 0.295(0.0894) (0.0780) (0.0315)

Figures in parentheses are standard errors. Model 1 is the standard model, assuming equal allele frequencies indiploids and triploids. Model 2 allows the frequencies to differ.

tween the parameter estimate and thetrue parameter value. The mean squarederror incorporates the bias and the vari-ance of the estimates.

Figure 1 shows the values of RE for twosets of allele frequencies. In the first set(Figure 1a), one allele was much morecommon than the other two. The KK esti-mator was more efficient than the ML es-timator for small values of t when the sam-ple size was 50 or 100, but otherwise theML estimator was preferable. The relativeefficiency of the KK estimator declinedrapidly as t increased and as the samplesize increased. In the second set of simu-lation runs (Figure 1b) the allele frequen-cies were equal. The KK estimator was al-most always less efficient than the MLestimator. The decline in efficiency withincreasing t was less rapid than in the firstset of simulations and the effect of thesample size (N) was much smaller.

The poor relative performance of theML estimator for small values of t andsmall sample sizes in the first set of sim-ulations was due partly to a large positivebias, though the estimator was also morevariable than the KK estimator. This illus-trates that the ML estimator, though hav-ing desirable theoretical properties inlarge samples, need not be optimal forsmall samples. However, in these circum-stances the KK estimator, though prefera-ble to the ML estimator, was itself a poorestimator. For example, with N 5 50 and t5 0.1, the KK estimate exceeded 0.5 in 583of the first set of 5,000 simulations(11.7%), and in 222 instances (4.4%) theestimate was one, suggesting that all in-dividuals were triploid. For comparison, inthe second set of simulations, when theallele frequencies were equal and the twoestimation methods performed similarly,only four of the KK estimates (0.08%) ex-ceeded 0.5, the largest being 0.63. Our con-clusion is that the ML method is as good

as, and usually better than, the KK meth-od, except in some circumstances inwhich neither method gives reliable esti-mates.

Although our primary interest is in theparameter t, it is of interest to comparebriefly the estimates of allele frequenciesobtained by the two methods. For the firstset of simulations, with one allele muchmore common than the other two, the re-sults were similar to those for parametert. The ML method was more efficient, oftensubstantially so, except when the samplesize and the value of t were small. For thesecond set of simulations, with all allelefrequencies equal, there was little differ-ence in efficiency when t was small but, atlarge values of t, the KK method was moreefficient, particularly when the samplesize was small. When t 5 0.9, the relativeefficiency was about 120% when N 5 50,and was still about 110% when N 5 250.

Comparison with Estimation WhenTriploids Can Be IdentifiedAs we have discussed earlier, the need forthese rather elaborate approaches to es-timating the proportion of triploids in thepopulation arises because triploids can-not be identified reliably unless they carrythree distinct alleles. Suppose, however,that such identification were possible, andthat t could be estimated directly as theproportion of triploids in a sample of sizen. This estimator is unbiased and its MSEis therefore equal to its variance, which ist(1 2 t)/n. We can then ask what value ofn would give the same MSE as the esti-mators considered earlier. For example, inthe simulations with t 5 0.1, a 5 1/3, b 51/3, and c 5 1/3, the MSE of the ML esti-

mator based on a sample of size 250 was0.00172. This same value could have beenobtained by taking n 5 52.2, that is, witha sample of only 52 individuals, if triploidscould be identified reliably. As can be seenfrom Table 2, even smaller samples wouldbe needed if the true value of t were larger,or particularly if the true allele frequen-cies were a 5 1/10, b 5 1/10, and c 5 4/5and, although the ML estimator is prefer-able to the KK estimator, both estimatorsare very poor. Although the mean squarederror of the ML and KK estimators can bereduced by increasing the sample size, amore effective approach, discussed below,will be to utilize data from several un-linked loci.

Goodness-of-FitOnce the parameters have been estimat-ed, it is sensible to check the goodness-of-fit to the observed phenotypic frequen-cies. A lack of fit suggests that theunderlying assumptions are invalid. Sincethere are seven genotypes, and three pa-rameters have been estimated, three de-grees of freedom remain to assess good-ness-of-fit. One assumption that can betested specifically is that allele frequen-cies are the same in diploids and triploids.To do this we fit a more complex model tothe data in which the allele frequenciesare different in diploids and triploids. Thelikelihood ratio test statistic is then twicethe difference in the maximized log-likeli-hood values of these two models. Underthe null hypothesis that the allele frequen-cies are the same, this statistic has, ap-proximately, a chi-square distribution withtwo degrees of freedom.

To illustrate the procedure, Table 3


gives simulated data for 500 individualswith t 5 0.5. For diploids we took a 5 b 50.2, c 5 0.6, and for triploids a 5 b 5 c 51/3. Parameter estimates under the twomodels are shown in Table 4 and the fittedvalues are shown in Table 3. For the stan-dard model, the chi-square goodness-of-fitstatistic is 15.38 (3 df, P , .01), but for theextended model with different allele fre-quencies in diploids and triploids, this isreduced to 0.91 (df. 5 1, P 5 .34). If, in-stead of the usual chi-square statistic, wecalculate the deviance ( likelihood ratio)statistic using the formula

D 5 2 observedO3 log(observed/expected),

we obtain very similar values of 15.09 forthe standard model and 0.93 for the ex-tended model; the difference betweenthese values, 14.16 (df. 5 2, P , 0.001), isthe likelihood ratio test for comparing thetwo models. Thus, as one might hope witha sample of size N 5 500, the analysis isable to detect the difference in allele fre-quencies of diploids and triploids veryconvincingly.

Single Locus with More than ThreeAllelesThe likelihood approach extends in astraightforward manner to loci with morethan three alleles. For m alleles the num-ber of phenotypes is m(m2 1 5)/6. Thereare m parameters to be estimated, leaving(m3 2 m 2 6)/6 degrees of freedom to as-sess goodness-of-fit. As with three alleles,a likelihood ratio test with m 2 1 degreesof freedom can be constructed to test theassumption that allele frequencies in dip-loids and triploids are equal.

Several LociThe likelihood approach can also be ex-tended to analyze several loci, providedthat they are unlinked and therefore seg-regate independently. In these circum-stances, the log-likelihood functions fordifferent loci can be added together togive an overall log-likelihood function.This function can be maximized to esti-mate separate allele frequencies for eachlocus, together with a single pooled valueof t. The likelihood ratio statistic for test-ing the homogeneity of t is twice the dif-ference between the maximized value ofthis function and the sum of the maxi-mized log-likelihood values for each locusanalyzed separately. This can be com-pared with the chi-square distributionwith degrees of freedom one less than the

number of loci. A significant value wouldcast doubt on the reliability of the pooledestimate.

Although the likelihood approach forseveral loci is straightforward conceptu-ally, the amount of computation increasesconsiderably with each new locus. A sim-pler method of obtaining a pooled esti-mate is to form a weighted average

r rt 1it 5 ,O Opool @V Vi51 i51i i

where r is the number of independent loci,ti is the estimate of t for the ith locus, andVi is the variance of this estimate. The es-timate tpool can be shown to be fully effi-cient, in the sense that its theoretical vari-ance is the same as the theoreticalvariance of the maximum likelihood esti-mator. Moreover, the statistic

2r ˆ ˆ(t 2 t )i pool2X 5 OVi51 i

may be used to test the homogeneity ofthe estimates {ti}. Under the null hypoth-esis of homogeneity, X2 is distributed aschi-square with r 2 1 degrees of freedom(e.g., Bailey 1961, Appendix A1.5).

Concluding Remarks

Likelihood methods provide a general ap-proach to statistical inference that isknown to be optimal in many senses forlarge sample sizes. However, as indicatedby this example, maximum likelihood es-timates may often be as good as, or betterthan, other ad hoc estimates, even withsamples of small or moderate size. Anoth-er advantage of using the likelihood ap-proach is that it allows simple tests ofsome of the underlying assumptions. Al-though the computations are somewhatmore complex than the method of Kriegerand Keller (1998), they can be completedvery rapidly on a modern personal com-puter, and we therefore suggest that thelikelihood approach should be the methodof choice for this problem. A program fordoing the basic calculations for a singlelocus, with up to nine distinct alleles, isavailable from the author as an executableprogram for IBM-compatible PCs.

From Horticulture Research International, East Malling,West Malling, Kent ME19 6BJ, UK.


References

Bailey NJT, 1961. Introduction to the mathematical the-ory of genetic linkage. Oxford: Oxford University Press.

Krieger MJB and Keller L, 1998. Estimation of the pro-portion of triploids in populations with diploid andtriploid individuals. J Hered 89:275–279.

Lange K, 1997. Mathematical and statistical methodsfor genetic analysis. New York: Spring-Verlag.

Nelder JA and Mead R, 1965. A simplex method forfunction minimization. Comp J 7:303–313.

Received September 23, 1998Accepted August 18, 1999

Corresponding Editor: James L. Hamrick

Inheritance of Unique Fruitand Foliage Color Mutationin NuMex Pinata

E. J. Votava, C. Balok, D. Coon,and P. W. Bosland

The inheritance of mature fruit color inpeppers (Capsicum spp.) is controlled byseveral genes. However, the inheritance ofthe transition of colors the fruit undergoduring ripening has not been describedextensively. The authors describe the in-heritance of a unique gene which affectsfoliage color and fruit color transition oc-curing in the jalapeno cultivar NuMex Pi-nata. The gene responsible is designatedthe tra gene.

The jalapeno cultivar NuMex Pinata wasreleased in 1998 (Votava and Bosland1998). It originated spontaneously in thecultivar Early Jalapeno, released byPetoSeed in 1977 (Tigchelaar 1980). Nu-Mex Pinata is unique in the transition ofcolors the fruit undergo as they mature.Immature fruit are light green, maturing toyellow, orange, and finally red. The fruitcolor of standard jalapeno cultivars, forexample, Early Jalapeno, changes fromdark green to red.

The inheritance of mature fruit color inCapsicum has been described by severalauthors (Hurtado-Hernandez and Smith1985; Kormos 1962; Kormos and Kormos1954, 1960; Schifriss and Pilovsky 1992;Smith 1950). It is generally considered tobe controlled by the combination and in-teraction of the following genes: c1 and c2(carotene pigment inhibitors), cl (chloro-phyll maintainer in fruit), and y (yellowmature fruit color) (Daskalov and Poulos1994). NuMex Pinata ripens to a final redmature fruit color, indicating, according togeneral consensus, that the genotype formature color is y 1 c1. However, it is thetransition of colors the fruit undergo asthey ripen as well as luteous foliage thatmakes NuMex Pinata interesting both


Table 1. Chi-square analysis for goodness-of-fit to a 3:1 ratio of inheritance of the Early Jalapenophenotype compared to the NuMex Pinata phenotype in the F2 generation

No. of plants

EarlyJalapenophenotype

NuMexPinataphenotype

Expectedratio x2 P

Early Jalapeno P1 4 0NuMex Pinata P2 0 4Early Jalapeno 3 NuMex Pinata F1 96 0NuMex Pinata 3 Early Jalapeno F1 96 0Selfed F2 521 175 3:1 0.0076 .9305

commercially and genetically, and whichsuggests that a separate genetic system isinvolved.

The inheritance of foliage color and fruitcolor transition in NuMex Pinata was stud-ied. The mode of inheritance of this traitis described and the gene responsible isdesignated.


Reciprocal hybridizations were made be-tween Early Jalapeno and NuMex Pinata.Hybridizations were performed in the en-vironmentally controlled insect-proof green-house of the New Mexico State UniversityChile Pepper Breeding Program in the ear-ly summer of 1996. A total of 192 F1 seed,representing eight distinct crosses, wereplanted in the greenhouse in late summer.Plants were evaluated for phenotype asthey grew. In the spring of 1997, F2 seed-lings from 12 randomly chosen F1 plantswere transplanted to the field at the Lyen-decker Plant Science Research Center, LasCruces, New Mexico. A total of 696 F2

plants were evaluated for phenotype asthe first-set fruit reached maturity. Datafrom the F2 segregating population wereanalyzed using the chi-square goodness-of-fit method.

Results and Discussion

The foliage and fruit color transition phe-notypes of the entire F1 population werethat of Early Jalapeno, regardless ofwhether Early Jalapeno was the female ormale parent, indicating the NuMex Pinataphenotype is recessive and displays nomaternal effects. The segregating F2 pop-ulation was scored for either of two ob-served phenotypes. One phenotype wasthat typical of the foliage and fruit colortransition associated with NuMex Pinata;the other phenotype was that typical ofthe foliage and fruit color transition of Ear-ly Jalapeno. No other phenotypes were ob-served. A single gene appears to have apleiotropic effect on the foliage of NuMex

Pinata. The foliage of Early Jalapeno andother jalapeno cultivars is dark green,while NuMex Pinata foliage has luteous fo-liage. There was an absolute associationbetween fruit color and foliage coloramong all 696 F2 progeny. No recombina-tion was observed that produced a NuMexPinata–type fruit color transition with anEarly Jalapeno–type foliage or, conversely,a normal Early Jalapeno fruit color tran-sition with a NuMex Pinata foliage pheno-type.

Data from the F2 segregating populationwas tested for goodness-of-fit to a 3:1 ratiousing chi-square analysis (Table 1). Theresults of the chi-square analysis indicatethat the progeny fit a 3:1 ratio. Thereforethe inheritance of the foliage color andfruit color transition phenotype of NuMexPinata is due to a single homozygous re-cessive gene.

No other Capsicum accession possessesthe same foliage color and fruit color tran-sition phenotype associated with NuMexPinata. Therefore true allele and comple-mentation tests were not performed.

In accordance with rules prepared bythe Capsicum and Eggplant Newsletter com-mittee for Capsicum gene nomenclature(CENL 1994), the gene responsible for theunique fruit color transition seen in Nu-Mex Pinata is designated tra for transition.

From the Department of Agronomy and Horticulture,MSC 3Q, New Mexico State University, Las Cruces, NM88003. This article is a contribution of the New MexicoAgricultural Experiment Station, New Mexico State Uni-versity, Las Cruces, New Mexico. Address correspon-dence to Eric J. Votava at the address above.


References

CENL Committee for Capsicum gene nomenclature,1994. Rules for gene nomenclature of Capsicum. Cap-sicum Eggplant Newsl 13:13–14.

Daskalov S and Poulos JM, 1994. Updated Capsicumgene list. Capsicum Eggplant Newsl 13:15–26.

Hurtado-Hernandez H and Smith PG, 1985. Inheritanceof mature fruit color in Capsicum annuum L. J Hered76:211–213.

Kormos J, 1962. Einige bemerkungen uber die karotin-

farben der paprikafruct. Acta Bot (Acad Sci Hung) 8:279–281.

Kormos J and Kormos K, 1954. Heredite des pigmentset des carotenoides chez le piment. Ann Inst Biol (Ti-hany) Hung Acad Sci 22:253–259.

Kormos J and Kormos K, 1960. Die genetischen typender carotinoid system der paprikafruct. Acta Bot (AcadSci Hung) 6:305–319.

Schifriss C and Pilovsky M, 1992. Studies of the inher-itance of mature fruit color in Capsicum annuum L. Eu-phytica 60:123–126.

Smith PG, 1950. Inheritance of brown and green maturefruit color in peppers. J Hered 41:138–140.

Tigchelaar EC (ed), 1980. New vegetable varieties listXXI. HortScience 15:565–576.

Votava EJ and Bosland PW, 1998. ’NuMex Pinata’ jala-peno chile. HortScience 33:350.

Received March 1, 1999Accepted August 18, 1999

Corresponding Editor: Kendall R. Lamkey

Population Genetics ofGeographically Restrictedand Widespread Species ofMyrica (Myricaceae)

Y.-P. Cheng, C.-T. Chien, andT.-P. Lin

Allozyme variation of 11 putative loci infive populations of the rare Myrica aden-ophora Hance, and four populations of itswidespread congeneric species, M. rubra(Lour.) Sieb. & Zucc. was studied. Amongthe 21 alleles studied, no unique allelewas detected for M. adenophora, whereasM. rubra had 3 alleles not found in the for-mer species. In terms of genetic diversity,populations of the rare species containedfewer alleles per locus (1.5 versus 1.7),fewer effective number of alleles per locus(1.12 versus 1.20), fewer number of allelesper polymorphic locus (2.14 versus 2.46),lower percentage of polymorphic loci(30.9 versus 40.9), and lower expectedheterozygosity (0.106 versus 0.163) thanpopulations of the widespread species.Genetic distances within species average0.043 for M. adenophora and 0.045 for M.rubra, and between species ranged from0.052 to 0.177, with a mean of 0.103,which agrees with the very similar grossmorphologies of these two species. Intra-population differentiation was similar inboth species: GST 5 0.152 for M. adeno-phora, and 0.146 for M. rubra, whereasestimated gene flow based on GST valueswere moderate in these two species (Nm5 1.39 versus 1.46). We inferred that M.rubra and M. adenophora are a progeni-tor-derivative species pair that emerged


Figure 1. Location of nine sample populations in Taiwan. Closed squares indicate M. adenophora (five popula-tions) and closed circles represent M. rubra (four populations).

before migrating into Taiwan during thelast glacial period. We consider theHengchun population (Chiupeng, Hsuhai,and Chufengpi) and Taitung population(Tienkuan and Lanshan) of M. adenophorawhich probably arose from two subsets ofthe genome of M. rubra. Genetic drift wasinferred to be one of the forces shapingthe observed genetic structure in M. ad-enophora and M. rubra.

Recent interest in conservation biologyhas directed attention to comparisons ofwidespread versus restricted congenericspecies, especially with respect to levelsof genetic variation. An understanding ofthe long-term demographic patterns islikely to be useful as an indicator of futureprospects for a rare species (Milligam etal. 1994). Enzyme electrophoresis hasbeen used to describe the levels and dis-tribution of genetic variation and the pop-ulation genetic structure of a variety ofgroups of plants. The use of allozymes hasbeen particularly fruitful in this respect,enabling identification of progenitor-deriv-ative species pairs in several plant genera(see reference in Kadereit et al. 1995). Thegeneral conclusion is that genetic similar-ity between progenitor and derivative isvery high, and that in virtually all casesthe derivative taxon contains a subset ofthe alleles present in the progenitor taxonand alleles unique to the derivative arerare or absent (Gottlieb 1973). Severalstudies on this subject were found in re-cent publications (Allen et al. 1991; Ed-wards and Wyatt 1994; Kadereit et al. 1995;Mayer et al. 1994; Purdy and Bayer 1995;Purps and Kadereit 1998).

Myrica has a wind-pollinated catkin in-florescence. Two species of Myrica occurin Taiwan, M. rubra and M. adenophora,both of which are favorites for landscapedesign. These two species are usually di-oecious, but may be monoecious as in M.adenophora. M. adenophora has an ex-tremely restricted distribution, only beingfound on grasslands (elevation of 50 m) ofthe east coast of the Hengchun Peninsula,and on mollisol of a mud stone area (ele-vation of 100 m) of the coastal mountainrange in Taitung County. The Hengchunpopulation is considered an endemic va-riety of M. adenophora Hance var. kusanoiHayata (Hayata 1911). Illegal removal forcommercial purposes has consistently re-duced the population size of M. adenopho-ra. M. adenophora was also reported inGuangdong and Fujian Provinces of China.M. rubra, known as bayberry or strawber-ry tree, is widespread but patchily distrib-

uted in Taiwan from low elevations up to2000 m. M. rubra is a species of east Asia,including Korea, Japan, Ryukyu, China,and the Philippines. These two species arevery similar in gross morphology and arenever found sympatrically. Tropical weath-er and a strong prevailing northeast win-ter monsoon may cause plants of M. ad-enophora to change their morphologywhen compared with M. rubra, which oc-cur throughout the island except Heng-chun Peninsula. In addition to the differ-ence in environmental factors, we as yetdo not know whether the morphologiesshown in M. adenophora and M. rubra rep-resent similar differences in the geneticcomposition of the two species.

For this study, we compare genetic vari-ation in two species of Myrica. This spe-

cies pair therefore should be well suitedfor a comparison of genetic variation. Thecomparison should be especially interest-ing to see if the results conform with theconclusions reported previously.


SamplingTwo species of Myrica were sampled inthis study. M. rubra is found as a big treein broadleaf forests and can be distin-guished from M. adenophora by the follow-ing characteristics: glabrous branches,staminate flowers with six to eight stamen,and spheroid fruit with a diameter of 1.4–2 cm. M. adenophora, on the other hand,is a shrub or small tree with smaller, thick-er leaves, pubescent young branches, sta-


Table 1. Allele frequencies and expected (He) and observed (Ho) heterozygosities for sevenpolymorphic loci in populations of M. adenophora and M. rubra

Locus/allele

M. adenophora

Chiupeng Hsuhai Chufengpi Tienkuan Lanshan

M. rubra

Puli ChiayangNan-chuang

Yangming-shan

Mdh-2abcHe

Ho

0.0001.0000.0000.0000.000

0.0200.9800.0000.0400.040

0.0100.9900.0000.0200.020

0.1510.8490.0000.2600.302

0.1790.8210.0000.3040.357

0.0500.7900.1600.3510.420

0.2270.7610.0120.3730.477

0.0500.9500.0000.0960.100

0.0290.9710.0000.0580.058

Pgi-1abHe

Ho

0.9900.0100.0200.020

0.9700.0300.0590.060

1.0000.0000.0000.000

0.8370.1630.2760.279

0.6790.3210.4520.500

0.9600.0400.0780.080

0.9770.0230.0450.045

1.0000.0000.0000.000

0.5690.4310.4970.472

Pgm-1abcHe

Ho

0.0000.0700.9300.1320.140

0.0000.0001.0000.0000.000

0.0000.0100.9900.0200.020

0.0000.1050.8950.1900.209

0.0000.0360.9640.0710.071

0.1000.5900.3100.5510.480

0.0000.2390.7610.3680.477

0.0000.5490.4510.5010.561

0.0000.6670.3330.4510.333

Skdh-2abHe

Ho

0.2300.7700.3580.340

0.2300.7700.3580.260

0.0100.9900.0200.020

0.3140.6860.4360.302

0.3930.6070.4950.357

0.7200.2800.4070.400

0.9090.0910.1670.182

0.9510.0490.0940.098

0.9860.0140.0280.028

6Pgd-1abHe

0.9000.1000.182

0.9400.0600.114

0.5800.4200.492

0.9880.0120.023

1.0000.0000.000

0.9400.0600.114

0.9550.0450.088

1.0000.0000.000

0.5830.4170.493

Ho 0.200 0.120 0.600 0.023 0.000 0.120 0.091 0.000 0.000

6Pgd-2abHe

Ho

1.0000.0000.0000.000

1.0000.0000.0000.000

1.0000.0000.0000.000

1.0000.0000.0000.000

1.0000.0000.0000.000

0.8100.1900.3110.340

0.8980.1020.1860.205

0.9880.0120.0240.024

1.0000.0000.0000.000

Per-1abcHe

Ho

0.0000.3500.6500.4600.420

0.0000.2900.7100.4160.420

0.0400.2800.6800.4620.440

0.0120.8950.0930.1920.163

0.0001.0000.0000.0000.000

0.0100.2200.7700.3620.280

0.0120.6020.3860.4920.477

0.2320.5610.2070.5960.537

0.0140.2920.6940.4390.250

Alleles a, b, and c were designated according to their migration distance from the origin. Mdh-1, Pgi-2, Idh-1, andEst-1 are monomorphic loci.

Table 2. Genetic diversity statistics for each population and the species level of M. adenophora and M.rubra

Population N A Ae AP Pa He (SD)b Ho (SD)b

M. adenophoraChiupengHsuhaiChufengpiTienkuanLanshanMean

505050431441

1.51.51.51.61.41.5

1.121.101.101.141.141.12

2.002.002.502.202.002.14

36.427.318.245.527.330.9

0.105 (0.050)0.090 (0.046)0.092 (0.057)0.125 (0.046)0.120 (0.059)0.106 (0.013)

0.102 (0.047)0.082 (0.042)0.100 (0.064)0.116 (0.041)0.117 (0.057)0.103 (0.012)

Species level 207 1.6 1.14 2.20 54.5 0.126 (0.051) 0.101 (0.036)

M. rubraPuliChiayangNanchuangYangmingshanMean

5044413643

1.91.81.51.61.7

1.251.181.141.221.20

2.672.602.332.252.46

54.545.527.336.440.9

0.198 (0.061)0.156 (0.054)0.119 (0.065)0.179 (0.070)*0.163 (0.047)

0.193 (0.058)0.178 (0.062)0.120 (0.065)0.104 (0.051)*0.149 (0.036)

Species level 171 1.9 1.24 2.57 63.6 0.191 (0.060) 0.153 (0.050)

a The frequency of the most common allele is less than 0.95.b Difference between He and Ho is significant: * P , .01.

minate flowers with two to four stamen,and ellipsoid fruit but slightly flattenedwith a diameter of 0.7–1 cm.

Fresh materials of M. adenophora werecollected from five natural populations atChiupeng (50 individuals), Hsuhai (50),and Chufengpi (50) on the Hengchun Pen-insula, and Tienkuan (43) and Lanshan(14) of Taitung County (Figure 1), for a to-tal of 207 individuals. Hengchun and Tai-tung are separated by a geographical dis-tance of about 100 km. Hengchun’spopulations usually flower from Decemberto March, but those in Taitung do so be-tween August and November. M. rubrasamples were collected from four popula-tions at Puli (50), Nanchuang (41), Chiay-ang (44), and Yangmingshan (36) for a to-tal of 171 individuals (Figure 1). Newshoots were carried back to the labora-tory in sealed PE bags and stored in a re-frigerator.

ElectrophoresisYoung leaves were ground with extractionbuffer according to procedures describedin Feret (1971) and absorbed onto What-man 3MM filter paper (4 mm 3 12 mm).Paper strips were arranged in a plastic Pe-tri dish and stored at 2708C until use. Hor-izontal starch-gel electrophoresis wasused for separating the following iso-zymes: IDH (isocitrate dehydrogenase, EC1.1.1.42), MDH (malate dehydrogenase, EC1.1.1.37), PER (peroxidase, EC 1.11.1.7),6PGD (phosphogluconate dehydrogenase,EC 1.1.1.44), PGM (phosphoglucomutase,EC 5.4.2.2), SKDH (shikimate 5-dehydroge-nase, EC 1.1.1.25), EST (carboxylesterase,EC 3.1.1.1), and PGI (glucose-6-phosphateisomerase, EC 5.3.1.9). Electrophoresisand staining followed the procedure de-scribed in Cheliak and Pitel (1984): Thezone specifying the most anodally migrat-ing variant was designated as 1, the nextas 2, and so on. Within each zone the mostanodally migrating variant was designatedas a, the next b, and so on. Genotype fre-quencies were inferred directly from iso-zyme phenotypes.

Data AnalysisAllozyme genotypes from young leaf tis-sue of individual trees from each popula-tion were used in conjunction with theBIOSYS-1 package (Swofford and Selander1989) for estimates of allele frequencies,mean number of alleles per locus (A), ef-fective number per locus (Ae), the numberof alleles per polymorphic locus (AP ), per-centage of polymorphic loci (P ), and meanobserved (Ho) and expected (He) hetero-


Table 3. F statistics, G statistics, and contingency chi-square test for polymorphic loci in populations ofM. adenophora and M. rubra

Locus

M. adenophora

x2a FIS FIT HT HS GST

M. rubra

x2a FIS FIT HT HS GST

Mdh-2Pgi-1Pgm-1Skdh-2

4.845*0.7451.6037.085**

20.15320.06020.088

0.185

20.0680.102

20.0480.252

0.1240.1880.0850.360

0.1250.1610.0830.333

0.0660.1400.0230.074

15.956**0.0900.0000.125

20.2160.023

20.00120.027

20.1200.3090.1150.085

0.2370.2160.5230.193

0.2200.1550.4680.174

0.0730.2840.1060.100

6Pgd-16Pgd-2Per-1

6.632*—0.915

20.179—

0.047

0.103—0.421

0.209—0.501

0.162—0.306

0.223—0.389

81.866**1.850

10.236*

0.69220.104

0.173

0.76820.010

0.305

0.2270.1400.556

0.1740.1300.472

0.2340.0730.150

Mean 0.011 0.226 0.246 0.195 0.152 0.078 0.219 0.299 0.256 0.146

Chi-square test according to Li and Horvitz (1953) and referred to as FIS.a Significance level: *P , .05, ** P , .01.

Table 4. Nei’s measure of genetic identity (I) (above the diagonal) and genetic distance (D) (below thediagonal) for populations of M. adenophora and M. rubra

Population 1 2 3 4 5 6 7 8 9

ChiupengHsuhaiChufengpiTienkuanLanshanPuliChiayangNanchuangYangmingshan

*****0.0010.0150.0370.0590.0700.0650.0960.134

0.999*****0.0170.0430.0650.0770.0700.1070.144

0.9850.983*****0.0700.0980.1220.1260.1670.177

0.9640.9580.932*****0.0020.1100.0520.0760.157

0.9430.9370.9070.998*****0.1360.0580.0870.163

0.9320.9260.8850.8960.873*****0.0410.0380.046

0.9370.9320.8820.9500.9440.960*****0.0170.070

0.9080.8980.8470.9260.9170.9630.983*****0.058

0.8750.8660.8380.8550.8500.9550.9330.944*****

zygosities. Deviations of observed fromexpected heterozygosities over all poly-morphic loci for each population were as-sessed with chi-square testing. Wright’s Fcoefficient was used to estimate devia-tions of observed and expected Hardy–Weinberg heterozygotic frequencies foreach polymorphic locus in each popula-tion (Wright 1965). The amount of geneflow among these populations was esti-mated as Nm 5 (1 2 GST)/4GST (Wright1951). Genetic diversity within and amongpopulations was partitioned using Nei’s(1973) genetic diversity statistics.

A UPGMA (Sneath and Sokal 1973) clus-ter analysis of identity values was gener-ated using BIOSYS-1 to examine genetic as-sociations among populations. Principalcomponent analysis (PCA) (Sneath andSokal 1973) using the NTSYS-pc 1.8 pro-gram (Rohlf 1992) helped evaluate thephenetic interpopulational relationshipsbased on allele frequency distributionsfrom polymorphic loci only. PCA is pre-sented with OTUs (populations) plottedonto the derived variables (principalaxes).

ResultsEleven loci in eight enzyme systems testedcould be resolved clearly enough for ge-netic diversity analysis: Mdh-1, Mdh-2, Pgi-1, Pgi-2, Pgm-1, Skdh-2, 6Pgd-1, 6Pgd-2, Idh-1, Est-1, and Per-1. Among them, Mdh-2,

Pgi-1, Pgm-1, Skdh-2, 6Pgd-1, 6Pgd-2, andPer-1 are polymorphic (95% criterion) in atleast one population, while Mdh-1, Pgi-2,Idh-1, and Est-1 are monomorphic. Skdh-1,Got-1, Got-2, Mr-1, and Mr-2 are probablypolymorphic but were not included foranalysis because they were not adequate-ly resolved on most gels. Observed allelefrequencies at each locus for each popu-lation, along with overall mean allele fre-quencies and observed and expected het-erozygosities are presented in Table 1.6Pgd-2 is monomorphic in M. adenophora,but polymorphic in M. rubra. Among 21 al-leles of 11 loci, Mdh-2c, Pgm-1a, and 6Pgd-2b do not occur in M. adenophora, andtherefore are unique to M. rubra. Pgm-1a isa rare allele and only found in the Pulipopulation. M. adenophora has no uniquealleles relative to M. rubra.

Genetic DiversityMeasures of genetic diversity are present-ed in Table 2. M. adenophora has six poly-morphic loci. At the species level, the av-erage number of alleles per locus (As) was1.6, effective number of alleles per locus(Aes) was 1.14, the number of alleles perpolymorphic locus (APs) was 2.2, the per-centage of polymorphic loci per individual(Ps) was 54.5, and expected heterozygositywas 0.126. Estimates of the genetic diver-sity parameter of M. adenophora at thespecies level or at the population level

were categorized between endemic andnarrowly distributed species as defined byHamrick et al. (1992). At the populationlevel, M. adenophora exhibits a lower levelof genetic diversity than M. rubra. The av-erage number of alleles per locus (A) was1.5; effective numbers of alleles per locus(Ae) varied from 1.10 to 1.14, with an av-erage of 1.12; the number of alleles perpolymorphic locus (AP ) varied from 2.0 to2.5, with an average of 2.14; the percent-age of polymorphic loci per individual (P )varied from 18.2% to 45.5%, with an aver-age of 30.9%; and the expected heterozy-gosity was 0.106.

M. rubra has seven polymorphic loci. Atthe species level, estimates of parametersAs, Aes, APs, Ps, and Hes were 1.9, 1.24, 2.57,63.6, and 0.191, respectively. This level ofdiversity is similar to that observed in re-gional and widespread species. At thepopulation level, A, Ae, AP, P, and He were1.7, 1.20, 2.46, 40.9, and 0.163, respectively,and fall between the narrow and regionalspecies. Differences between observedand expected heterozygosities were notsignificant for any population of M. aden-ophora and M. rubra, indicating that thepopulations are in Hardy–Weinberg equi-librium.

Genetic DifferentiationIn M. adenophora the FIS value for all en-zyme systems was 0.011 (Table 3). This in-dicates allele frequencies within the pop-ulation are close to random mating. M.rubra also has a positive FIS (0.078) whichis not significantly deviated from zero asverified by the formula of Li and Horvitz(1953) (x2 5 2.08, P . .05). Contingencychi-square tests showed that Mdh-2, Skdh-2, and 6Pgd-1 of M. adenophora have sig-nificant deviations from Hardy–Weinbergexpectations. In M. rubra, however, Mdh-2,6Pgd-1, and Per-1 showed significant differ-ences.

The extent of genetic differentiationamong the populations (GST) of M. adeno-phora averaged 0.152. Thus about 15% ofgenetic variation resides among popula-tions. Similar genetic differentiationamong populations was also observed inM. rubra (0.146). Gene flow per generationwas moderate in both M. adenophora (Nm5 1.39) and M. rubra (Nm 5 1.46). Unbi-ased genetic distances between pairs ofpopulations of M. adenophora ranged from0.001 to 0.098, with an average of 0.043(Table 4). Among them, Chiupeng andHsuhai showed very low genetic distance;the same observation occurred betweenTienkuan and Lanshan. Genetic distances


Figure 2. Dendrogram of cluster analysis was using the UPGMA method for populations of M. adenophora andM. rubra.

Figure 3. Principal component analysis (PCA) based on allele frequencies for populations of M. adenophora(populations 1–5) and M. rubra (populations 6–9). The first two principal components explain 54.8% and 28.5% ofthe variances, respectively. 1, Chiupeng; 2, Hsuhai; 3, Chufengpi; 4, Tienkuan; 5, Lanshan; 6, Puli; 7, Chiayang; 8,Nanchuang; 9, Yangmingshan.

among populations of M. rubra varied from0.017 to 0.058, with an average of 0.045. Atthe species level, the genetic distance be-tween M. adenophora and M. rubra was0.103.

The UPGMA dendrogram produced clus-ters of populations separating M. adeno-phora and M. rubra into two distinctgroups (Figure 2) with a genetic distanceof 0.103. PCA of allele frequencies revealedthat Skdh-2 and Pgm-1 were the loci thatdifferentiated the population of M. adeno-

phora from populations of M. rubra themost. PCA also indicated a close geneticrelationship between populations of M. ad-enophora where they co-occurred at Lan-shan and Tienkuan, and Hsuhai and Chiu-peng (Figure 3).

Discussion

The relationship between M. adenophoraand M. rubra agrees with that of otherclosely related congeners suggesting that

they are a progenitor-derivative speciespair. The genetic similarity of these twospecies is 0.897, which, in fact, is higherthan the 0.67 6 0.04 for congeneric spe-cies reported by Gottlieb (1981).

Even though rare, M. adenophora main-tains higher levels of genetic diversitythan many rare tree species in Taiwan,such as Amentotaxus formosana (He 50.004; Wang et al. 1996), Bretschmeidera si-nensis (He 5 0.062; Huang 1994), and Rho-dodendron kanehirai (He 5 0.068; Huang etal. 1995). This pattern may be due to therelationships between the specific progen-itor-derivative species pairs. The progeni-tors of A. formosana, B. sinensis, and R. ka-nehirai occurring in mainland China havenever been studied, though theoreticallythey should have higher genetic diversi-ties. Cunninghamia lanceolata, native toChina, and C. konishii, found only in Tai-wan, is another progenitor-derivative spe-cies pair; the former having a much highergenetic diversity than the latter (Lin et al.1998).

It was reported that during the latePleistocene, temperate forests dominatedthe lowlands of Taiwan in the Tali glacialstage from about 50,000 to 15,000 BP (Tsu-kada 1967). During that time, although gla-ciation did not advance into southern Chi-na, the climate did cool. Taiwan wouldthen likely have been connected withmainland China by a land bridge due to alowering of the sea level (Liu 1988) acrosswhich plants could have migrated towardthe warmer south without a barrier. M. ad-enophora being better adapted to a tropi-cal climate might have survived in refugiaon the Hengchun Peninsula. As glaciers re-treated northward around 15,000 BP,warming returned again to the lowlands.M. adenophora remained in southern Tai-wan probably because there was less com-petition due to the lack of open grasslandsbeyond Hengchun Peninsula and it beingsmall in size. M. rubra, a species moreadapted to a temperate climate, spreadfrom lowland forests up to elevations ofabout 2000 m. Regarding species diver-gence, we inferred that M. adenophoraemerged from M. rubra in temperate main-land China before migrating into Taiwan.One alternative scenario for speciation ofM. adenophora could be that it originatedfrom M. rubra in Taiwan. However, it is un-likely that M. adenophora could have ex-perienced the same speciation indepen-dently from two distant localities:Guangdong of China and Taiwan. Certainlycomparison between populations fromboth Hengchun and Guangdong will shed


light on our understanding of the origin ofM. adenophora.

Lower genetic diversity of M. adenopho-ra could have originated as a subset ofthat of the progenitor. M. adenophora alsoexhibits no unique alleles, probably due toits relatively recent origin and experienc-ing genetic drift. These two groups, theHengchun (Chiupeng, Shuhai, and Chu-fengpi) and Taitung populations (Tien-kuan and Lanshan), distinctly differed intheir gene frequencies, for example, allelesMdh-2, Pgi-1, Skdh-2, 6Pgd-1, and Per-1; thismay indicate they represent a differentsubset of the genetic variability of M. ru-bra. To further test this hypothesis, char-acterization of chloroplast DNA or mito-chondria DNA haplotype would provideadditional evidence (Kadereit et al. 1995).At present the individuals per populationof M. adenophora are less than 500. Illegalremoval for commercial purposes and ar-tificial fire are the major disturbances thatreduce the number of individuals. Geneticdrift, in fact, may have occurred since M.adenophora is only restricted to the southpart of the island. For example, Pgm-1b be-comes a rare allele, 6Pgd-2a is a fixed al-lele, and 6Pgd-1a is fixed in the Taitungpopulation. Taitung populations grow onmollisol of a mud stone area, the soil tex-ture being generally grayish in color, hav-ing little moisture content during the dryseason, having an abundant mineral com-ponent, and being alkaline or neutral in re-action, but lacking organic compounds.Reproductive isolation was found betweenthe Hengchun and Taitung populations ofM. adenophora. We suspect adaptation tomollisol of a mud stone area might causereproductive isolation, as has been re-ported in the evolution of many plant spe-cies to a serpentine habitat (Kruckeberg1957). An edaphic factor can be the chiefstimulus for the evolution of endemic taxa(Kruckeberg 1986; Mason 1946a,b; Proctorand Woodell 1975). This isolation shouldproceed for a long time, thus the Heng-chun and Taitung populations have differ-entiated to the same genetic distance asthat between populations of widespreadM. rubra (Figure 2). Within M. rubra, genefrequency distribution of Pgi-1, Pgm-1, and6Pgd-1 in Nanchuang, and Pgm-1 and 6Pgd-2 in the Yangmingshan population are con-sistent with the action of genetic drift.

M. adenophora and M. rubra both havelow FIS, which is consistent with the ab-solute outcrossing pollination behavior ofMyrica. A high FIT value for both species isprobably due to a spatial Wahlund effect.Differences in allele frequencies between

the populations may produce the appear-ance of a deficiency of heterozygotes. M.rubra is patchily distributed in dense for-est and also has a restricted neighbor-hood size preventing from gene flow, thusfavoring population substructuring andthe Wahlund effect.

Genetic differentiation between popula-tions of M. adenophora (GST 5 0.152) issimilar to that of M. rubra (GST 5 0.146).These values are higher than the mean re-ported for wind pollination outcrossingplant of long-lived woody species (GST 50.077) (Hamrick et al. 1992), and is slightlyhigher than that reported for tropicalwoody species (GST 5 0.135) (Hamrick1994). In general, not only is the dispersalof fruit restricted, differences in floweringperiod contribute to the genetic differen-tiation of the Hengchun and Taitung pop-ulations of M. adenophora. The same phe-nomenon was also observed within M.rubra but has not been well defined, es-pecially among localities at different ele-vations.

From the Taiwan Forestry Research Institute, 53 Nan-Hai Road, Taipei, Taiwan. This research was financiallysupported by the Taiwan Forestry Research Institute(TFRI). This paper is contribution no. 133 of the TFRI.We thank Chien-Lien Pan and Wan-Long Chang, Heng-chun Station of Taiwan Forestry Research Institute, forassistance with field sampling. Address correspon-dence to Yu-Pin Cheng at the address above or e-mail:[email protected].


References

Allen GA, Gottlieb LD, and Ford VS, 1991. Electropho-retic evidence for the independent origins of two self-pollinatiry subspecies of Clarkia concinna (Onagra-ceae). Can J Bot 69:2299–2301.

Cheliak WM and Pitel JA, 1984. Techniques for starchgel electrophoresis of enzymes from forest tree spe-cies. Information Report PI-X-42. Chalk River, Ontario:Petawawa National Forest Institute; 19–45.

Edwards AL and Wyatt R, 1994. Population genetics ofthe rare Asclepias texana and its widespread sister spe-cies, A. perennis. Syst Bot 19:291–307.

Feret PP, 1971. Isozyme variation in Picea glauca(Moench) Voss seedlings. Silvae Genet 20:46–50.

Gottlieb LD, 1973. Genetic differentiation, sympatricspeciation and the origin of diploid species of Stepha-nomeria. Am J Bot 60:545–553.

Gottlieb LD, 1981. Electrophoretic evidence and plantpopulations. Progr Phytochem 7:1–46.

Hamrick JL, 1994. Genetic diversity and conservationin tropical forests. In: Proceedings of the InternationalSymposium on Genetic Conservation and Production ofTropical Forest Tree Seeds, Ching Mai, Thailand, 1993(Drysdale RM, John SET, and Yapa AC, eds). Dordrecht,The Netherlands: Kluwer Academic; 95–124.

Hamrick JL, Godt MJW, and Sherman-Broyles SL, 1992.Factors influencing levels of genetic diversity in woodyplant species. New For 6:95–124.

Hayata B, 1911. Materials for a flora of Formosa. J CollSci Imp Univ Tokyo 30(1):255.

Huang S, 1994. Genetic structure of Bretschneidera si-

nensis Hemsl. of Taiwan [in Chinese]. Ann Taiwan Mus37:49–67.

Huang S, Chen CM, Hsu SL, and Lu SY, 1995. Populationgenetic of endangered species, Rhododendron kanehi-rai [in Chinese]. Biol Bull Nat Taiwan Normal Univ30(2):63–68.

Kadereit JW, Comes HP, Curnow DJ, Irwin JA, and Ab-bott RJ, 1995. Chloroplast DNA and isozyme analysis ofthe progenitor-derivative species relationship betweenSenecio nebrodensis and S. viscosus (Asteraceae). Am JBot 82:1179–1185.

Kruckeberg AR, 1957. Variation in fertility of hybridsbetween isolated populations of the serpentine spe-cies, Streptanthus glandulosus Hook. Evolution 11:185–211.

Kruckeberg AR, 1986. An assay: the stimulus of unusualgeologies for plant speciation. Syst Bot 11:455–463.

Li CC and Horvitz DG, 1953. Some methods of estimat-ing the inbreeding coefficient. Am J Hum Genet 5:107–117.

Lin TP, Wang CT, and Yang JC, 1998. Comparison of ge-netic diversity between Cunninghamia konishii and C.lanceolata. J Hered 89:370–373.

Liu KB, 1988. Quaternary history of the temperate for-est of China. Q Sci Rev 7:1–20.

Mason HL, 1946a. The edaphic factor in narrow endem-ism. I. The nature of environmental influences. Mad-rono 8:209–240.

Mason HL, 1946b. The edaphic factor in narrow endem-ism. II. The geographic occurrence of plants of highlyrestricted patterns of distribution. Madrono 8:241–272.

Mayer MS, Soltis PS, and Soltis DE, 1994. The evolutionof the Streptanthus glandulosus complex (Cruciferae):genetic divergence and gene flow in serpentine endem-ics. Am J Bot 81:1288–1299.

Milligam BG, Leebens-Mack J, and Strand E, 1994. Con-servation genetics: beyond the maintenance of markerdiversity. Mol Ecol 3:423–435.

Nei M, 1973. Analysis of gene diversity in subdividedpopulations. Proc Natl Acad Sci USA 70:3321–3323.

Proctor J and Woodell SRJ, 1975. The ecology of ser-pentine soils. Adv Ecol Res 9:255–365.

Purdy BG and Bayer RJ, 1995. Allozyme variation in theAthabasca sand dune endemic, Salix silicicola, and theclosely related widespread species, S. alaxensis. SystBot 20:179–190.

Purps DML and Kadereit JW, 1998. RAPD evidence fora sister group relationship of the presumed progenitor-derivative species pair Sencico nebrodensis and S. vis-cosus (Asteraceae). Plant Syst Evol 211:57–70.

Rohlf FJ, 1992. NTSYS-pc: numerical taxonomy and mul-tivariate analysis system, version 1.8. Setauket, NY: Ex-ter Software.

Sneath PH and Sokal RR, 1973. Numerical taxonomy:the principles and practice of numerical classification.San Francisco; W.H. Freeman.

Swofford DL and Selander RB, 1989. A computer pro-gram for the analysis of allelic variation in populationgenetics and biochemical systematics, version 1.7.Champaign, IL: Illinois Natural History Survey.

Tsukada M, 1967. Vegetation in subtropical Formosaduring the Pleistocene glaciations and the Holocene.Paleogeog Paleoclimat Paleoecol 3:49–64.

Wang CT, Wang WY, Chiang CH, Wang YN, and Lin TP,1996. Low genetic variation in Amentotaxus formosanaLi revealed by isozyme analysis and random amplifiedpolymorphic DNA markers. Heredity 77:388–395.

Wright S, 1951. The genetic structure of populations.Ann Eugen 15:323–354.

Wright S, 1965. The interpretation of population struc-ture by F-statistics with special regard to systems ofmating. Evolution 19:395–420.

Received December 28, 1998Accepted August 18, 1999

Corresponding Editor: James L. Hamrick

Turkish Honeybees: Genetic Variation and Evidence for a Fourth ...

Documents

Transcript of Turkish Honeybees: Genetic Variation and Evidence for a Fourth ...