Population Structure and Genetic Diversity of the Boll Weevil (Coleoptera: Curculionidae) on Gossypium in North America Adam P

Entomology Publications Entomology

2012 Population Structure and Genetic Diversity of the Boll Weevil (Coleoptera: Curculionidae) on Gossypium in North America Adam P. Kuester Iowa State University, [email protected]

Robert W. Jones Centro Universitario Queretaro

Thomas W. Sappington Iowa State University, [email protected]

Kyung Seok Kim Seoul National University

NFoorllomwa nthi Bs. Bandarr additional works at: https://lib.dr.iastate.edu/ent_pubs UnitPeda Srta oftes theDepaArtmgreniculturt of Agalric Sulctuierence Commons, Agriculture Commons, Agronomy and Crop Sciences Commons, Biology Commons, Entomology Commons, Genetics Commons, and the See next page for additional authors Systems Biology Commons The ompc lete bibliographic information for this item can be found at https://lib.dr.iastate.edu/ ent_pubs/198. For information on how to cite this item, please visit http://lib.dr.iastate.edu/ howtocite.html.

This Article is brought to you for free and open access by the Entomology at Iowa State University Digital Repository. It has been accepted for inclusion in Entomology Publications by an authorized administrator of Iowa State University Digital Repository. For more information, please contact [email protected]. Population Structure and Genetic Diversity of the Boll Weevil (Coleoptera: Curculionidae) on Gossypium in North America

Abstract Although the boll weevil, Anthonomus grandis grandis Boheman (Coleoptera: Curculionidae), is a devastating pest in the United States and Mexico, its population structure and genetic diversity in Mexico on wild and cultivated cotton hosts (genus Gossypium) is poorly understood. Past studies using morphology, host use, and distribution records suggest that A.grandis grandis comprises three forms with host-associated characteristics: the southeastern form (from domesticated Gossypium hirsutum L., southeastern United States and northeastern Mexico), the thurberia form (from Gossypium thurberi Todaro, Arizona and northwestern Mexico), and the Mexican form (from multiple Gossypium species and other malvaceous plant genera in the remainder of Mexico and Central America). However, the phylogenetic relationships, host preferences, and distributions of these forms are not completely understood. An alternative hypothesis of an eastern and western form of the boll weevil is suggested by the suspected phylogeographic range expansion from an ancestral distribution in the tropics northward along both Mexican coasts, culminating in the maximally contrasting phenotypes observed in the northeastern and northwestern arms of the current distribution. In this study, we sequenced one mitochondrial and four nuclear genes to gain insight into the evolutionary relationships among the putative forms and their distributions on wild and domesticated cotton hosts. Using models of evolution, we compared the three-form to the two-form classification and to two alternative classifications that incorporate geography and host use traits. The eg netic data at most loci provide stronger support for the two-form than the three-form hypothesis, with an eastern and western group separated by the Sierra Madre Occidental mountain range. They do not us pport separate taxonomic status for boll weevils developing onG. thurberi.

Keywords boll weevil, thurberia weevil, Gossypium, genetic marker

Comments This article is from Annals of the Entomological Society of America 105 (2012): 902, doi:10.1603/AN12072.

Rights Works produced by employees of the U.S. Government as part of their official duties are not copyrighted within the U.S. The onc tent of this document is not copyrighted.

Authors Adam P. Kuester, Robert W. Jones, Thomas W. Sappington, Kyung Seok Kim, Norman B. Barr, Richard L. Roehrdanz, and John D. Nason

This article is available at Iowa State University Digital Repository: https://lib.dr.iastate.edu/ent_pubs/198 Population Structure and Genetic Diversity of the Boll Weevil (Coleoptera: Curculionidae) on Gossypium in North America Author(s): Adam P. Kuester, Robert W. Jones, Thomas W. Sappington, Kyung Seok Kim, Norman B. Barr, Richard L. Roehrdanz, Patti Senechal, and John D. Nason Source: Annals of the Entomological Society of America, 105(6):902-916. 2012. Published By: Entomological Society of America URL: http://www.bioone.org/doi/full/10.1603/AN12072

BioOne (www.bioone.org) is a nonprofit, online aggregation of core research in the biological, ecological, and environmental sciences. BioOne provides a sustainable online platform for over 170 journals and books published by nonprofit societies, associations, museums, institutions, and presses. Your use of this PDF, the BioOne Web site, and all posted and associated content indicates your acceptance of BioOne’s Terms of Use, available at www.bioone.org/page/terms_of_use. Usage of BioOne content is strictly limited to personal, educational, and non-commercial use. Commercial inquiries or rights and permissions requests should be directed to the individual publisher as copyright holder.

BioOne sees sustainable scholarly publishing as an inherently collaborative enterprise connecting authors, nonprofit publishers, academic institutions, research libraries, and research funders in the common goal of maximizing access to critical research. GENETICS Population Structure and Genetic Diversity of the Boll Weevil (Coleoptera: Curculionidae) on Gossypium in North America

ADAM P. KUESTER,1 ROBERT W. JONES,2 THOMAS W. SAPPINGTON,3 KYUNG SEOK KIM,4 5 6 7 1,8 NORMAN B. BARR, RICHARD L. ROEHRDANZ, PATTI SENECHAL, AND JOHN D. NASON

Ann. Entomol. Soc. Am. 105(6): 902Ð916 (2012); DOI: http://dx.doi.org/10.1603/AN12072 ABSTRACT Although the boll weevil, Anthonomus grandis grandis Boheman (Coleoptera: Curcu- lionidae), is a devastating pest in the United States and Mexico, its population structure and genetic diversity in Mexico on wild and cultivated cotton hosts (genus Gossypium) is poorly understood. Past studies using morphology, host use, and distribution records suggest that A. grandis grandis comprises three forms with host-associated characteristics: the southeastern form (from domesticated Gossypium hirsutum L., southeastern United States and northeastern Mexico), the thurberia form (from Gos- sypium thurberi Todaro, Arizona and northwestern Mexico), and the Mexican form (from multiple Gossypium species and other malvaceous plant genera in the remainder of Mexico and Central America). However, the phylogenetic relationships, host preferences, and distributions of these forms are not completely understood. An alternative hypothesis of an eastern and western form of the boll weevil is suggested by the suspected phylogeographic range expansion from an ancestral distribution in the tropics northward along both Mexican coasts, culminating in the maximally contrasting phenotypes observed in the northeastern and northwestern arms of the current distribution. In this study, we sequenced one mitochondrial and four nuclear genes to gain insight into the evolutionary relationships among the putative forms and their distributions on wild and domesticated cotton hosts. Using models of evolution, we compared the three-form to the two-form classiÞcation and to two alternative classiÞcations that incorporate geography and host use traits. The genetic data at most loci provide stronger support for the two-form than the three-form hypothesis, with an eastern and western group separated by the Sierra Madre Occidental mountain range. They do not support separate taxonomic status for boll weevils developing on G. thurberi.

KEY WORDS boll weevil, thurberia weevil, Gossypium, genetic marker

The boll weevil, Anthonomus grandis grandis Boheman Belt in the southeastern United States (Burke et al. (Coleoptera: Curculionidae), has been a major pest of 1986, Allen 2008). This range expansion is believed to commercial cotton (genus Gossypium) in the United have been a result of increased cotton production in States for more than a century and is still of agricul- northeastern Mexico after the devastation of the cot- tural concern in southern Texas and parts of northern ton-growing industry in the United States during the Mexico and Central and South America (Burke et al. Civil War (Jones 2006, Allen 2008). An increase in 1986, Scataglini et al. 2006, Stadler and Buteler 2007). cotton agriculture in northern Mexico bridged a geo- The Swedish entomologist C. H. Boheman described graphical gap for boll weevil migration and establish- A. grandis grandis in 1843 from a specimen collected ment in areas previously devoid of a host for the pest. in Veracruz, Mexico. The insect crossed the Rio This resulted in ruinous consequences for U.S. cotton Grande in 1892 and quickly moved through the Cotton production including ϾUS$10 billion in crop damage and management costs within the United States alone 1 Department of Ecology, Evolution, and Organismal Biology, 251 (Allen 2008). Although the boll weevil has recently Bessey Hall, Iowa State University, Ames, IA 50011. been eradicated from much of the southeastern 2 Universidad Auto´noma de Quere´taro, Licenciatura en Biologõ´a, Centro Universitario, Quere´taro, 76010, Quere´taro, Me´xico. United States and parts of northern Mexico, many 3 USDAÐARS, Corn Insects and Crop Genetics Research Unit, Ge- areas continue to be active eradication zones in which netics Laboratory, Iowa State University, Ames, IA 50011. management and associated costs are signiÞcant (Al- 4 Conservation Genome Resource Bank for Korean Wildlife, Re- len 2008). search Institute and BK21 Program for Veterinary Science and Col- lege of Veterinary Medicine, Seoul National University, Seoul, Re- The boll weevil infests many species of wild cotton public of Korea. in addition to several noncotton hosts throughout the 5 Center for Plant Health Science and Technology, Mission Labo- Americas, although phylogenetic analyses of the A. ratory, USDAÐAPHIS, Moore Air Base, Edinburg, TX 78541. grandis grandis species group suggest it originated 6 USDAÐARS BRL, 1605 Albrecht Blvd., Fargo, ND 58102. 7 Biodesign Institute, Arizona State University, Tempe, AZ 85287. on Hampea (Malvaceae: Gossypieae) (Jones 2001), 8 Corresponding author, e-mail: [email protected]. found in southern Mexico and Central America. Re-

0013-8746/12/0902Ð0916$04.00/0 ᭧ 2012 Entomological Society of America November 2012 KUESTER ET AL.: BOLL WEEVIL GENETIC STRUCTURE AND DIVERSITY 903 search addressing phylogenetic or cladistic relation- portant in the management the pest on commercial ships of wild host-associated boll weevils by using cotton in eradicated areas. molecular data are sparse and limited thus far to stud- The detection of boll weevils by pheromone traps in ies of weevils obtained from a single wild cotton host an active or posteradication zone can trigger costly (Gossypium thurberi Todaro) (Roehrdanz 2001) and pesticide applications. The appropriate response by an to weevil samples from cultivated and wild vegetation eradication program depends in part on whether the in South America (Scataglini et al. 2000, 2006). captured weevils are from local populations infesting The boll weevil has been classiÞed into three forms, commercial cotton or long-distance migrants from one form associated with cultivated cotton (Gos- cultivated or wild hosts (Kim et al. 2010). Diagnostic sypium hirsutum L.) in the southeastern United States markers to indicate source areas would be an impor- and northeastern Mexico (the southeastern boll wee- tant tool in helping management ofÞcials determine vil [SE]), one form associated with the wild cotton G. appropriate control tactics (Roehrdanz 2001). As thurberi in the Sonoran Desert of the southwestern mentioned, suites of morphological characters could United States and northwestern Mexico (thurberia potentially serve in this regard (Warner 1966, Burke boll weevil [TW]), and one form associated with cul- 1968), but the differences are subtle, and a diagnostic tivated and wild cotton species (and a few other mal- system has yet to be developed. Molecular markers are vaceous hosts) elsewhere in Mexico (Mexican boll another potential avenue for diagnosing boll weevil weevil [MX]) (Burke 1968, Burke et al. 1986). Both forms. the SE and MX forms are pests of commercial cotton, In this study, we used DNA sequence data to infer whereas the TW form is not considered a pest, even evolutionary relationships among boll weevils sam- though it may be found on G. hirsutum late in the pled from across the geographical ranges of the tra- season (Fye 1968). Behavioral and morphological dif- ditional SE, TW, and MX forms. To date, intraspeciÞc ferences between TW and SE forms have been re- DNA sequence variation in boll weevil has been an- ported previously (zharvBurke 1968, Fye 1968, Burke alyzed using two loci, the internal transcribed spacer et al. 1986), with the MX form exhibiting traits inter- region II (ITSII) of the nuclear ribosomal RNA (Roeh- mediate to the other two forms (Warner 1966, Burke rdanz 2001, Roehrdanz et al. 2010) and mitochondrial 1968, Burke et al. 1986). However, the phenotypic cytochrome oxidase I and II (COI and COII, respec- variation in morphological traits characteristic of the tively) (Scataglini et al. 2006). Here, we used se- different boll weevil forms is not entirely genetically quence data from ITSII and COI plus three additional based, as revealed by larval diet experiments (Burke nuclear loci that have proven useful for population- 1968, Burke et al. 1986); therefore, they cannot be used level and shallow time phylogenetic analyses of other directly for phylogenetic reconstruction or for direct insect groups: nuclear elongation factor 1␣ (EF-1␣, taxonomic designation (Roehrdanz 2001). Neverthe- Hughes and Vogler 2004), arginine kinase (AK), and less, morphometric traits, especially in combination rudimentary gene (CAD; Wild and Maddison 2008). with genetic data, have shown potential for identifying We subjected these Þve loci to phylogenetic and co- the host plant or region of origin (Warner 1966, Burke alescent analyses to evaluate the extent to which they 1968, Burke et al. 1986) and thus for indirectly assign- support the classiÞcation of boll weevils into three ing an individual to a taxonomic form based on this forms (SE, TW, and MX) as opposed to only two information. forms, eastern and western. Molecular markers have been used to evaluate sup- The main objectives of this study are to 1) use port for the three weevil forms, and previous work by molecular markers to better understand the popula- Roehrdanz (2001) supports a genetic separation be- tion genetic structure of the boll weevil from Gos- tween TW and SE forms. This separation concurs with sypium in North America and 2) to identify diagnostic the scenario proposed by Burke et al. (1986) whereby molecular markers to differentiate forms from the the boll weevil dispersed northward along the eastern most probable boll weevil classiÞcation scheme. Fur- and western coasts of Mexico, isolated by MexicoÕs thermore, because boll weevils attack both domesti- central mountain ranges. Depending on the degree of cated cotton cultivars and a number of wild cotton isolation, genetic divergence of these populations species in western Mexico, we also examined data for through drift and local adaptation is expected. How- host-associated structure. The markers developed ever, because genetic analyses have been limited prin- here should prove valuable for determining the pop- cipally to the TW and SE forms, it is unknown whether ulations of origin of boll weevils invading eradication these two forms represent genetically distinct popu- zones, especially with increased genetic characteriza- lations or are extreme phenotypes of a south-to-north tion of boll weevil populations originating on wild gradient in genetic divergence from southern popu- cotton hosts in Mexico. lations. At present, the molecular genetic data needed to evaluate these hypotheses are lacking and the evo- Materials and Methods lutionary relationships among putative boll weevil forms are unclear. Analyses of boll weevil population Boll Weevil Species Sampling. Most boll weevils genetic structure across the species range in North from wild cotton species A. grandis adults and imma- America would be helpful in resolving these issues and tures were collected by hand in Mexico during falls are economically and environmentally important 2008Ð2010, including G. thurberi, Gossypium turneri given that the source of boll weevil outbreaks is im- Fryxell, Gossypium aridum (Rose and Standl.) Skovst, 904 ANNALS OF THE ENTOMOLOGICAL SOCIETY OF AMERICA Vol. 105, no. 6

Table 1. Locations of A. grandis grandis collections from cultivated and wild cotton species in Mexico and the United States

Site Collection Location Host Plant Form Region Na Yr Latitudeb Longitude no. method 1 Weslaco, TX G. hirsutum SE E 5 2000 Trap 2 Tlahualilo, Durango G. hirsutum SE E 5 2004 Trap 3 Ojinaga, Chihuahua G. hirsutum SE E 5 2004 Trap 4 Rosales, Chihuahua G. hirsutum SE E 5 2004 Trap 5 Lubbock, TX G. hirsutum SE E 5 2002 Trap 6 Childress, TX G. hirsutum SE E 5 2001 Trap 7 Artesia, NM G. hirsutum SE E 5 2001 Trap 8 Hobart, OK G. hirsutum SE E 5 2001 Trap 9 Little Rock, AR G. hirsutum SE E 5 2001 Trap 10 Brownsville, TN G. hirsutum SE E 5 2001 Trap 11 Malden, MO G. hirsutum SE E 5 2002 Trap 12 Cleveland, MS G. hirsutum SE E 5 2001 Trap 13 La Ventosa, Oaxaca G. aridum EMX E 3 2011 16 34.867Њ N 94 48.983Њ W Hand 14 Tecoman, Colima G. aridum WMX W 12 2011 19 05.017Њ N 103 46.644Њ W Hand 15 Tecoman, Colimac G. hirsutum WMX W 15 2011 18 54.600Њ N 103 42.900Њ W Trap 16 El Cajon Presa, Nayarit G. aridum WMX W 5 1998 21 26.032Њ N 104 27.697Њ W Hand 17 Cajeme, Sonora (Obregon) G. hirsutum WMX W 15 2011 27 25.700Њ N 109 53.400Њ W Trap 18 Rncho San Ramon, Sonora G. hirsutum WMX W 15 2011 28 22.415Њ N 111 18.085Њ W Hand/Trap 19 San Carlos, Sonora G. turneri WMX W 15 2011 27 58.677Њ N 111 07.725Њ W Hand 20 Baja California Surd G. davidsonii WMX W 20 2006Ð2010 22 55.498Њ N 109 50.488Њ W Hand 21 Laveen, AZ G. hirsutum WMX W 8 1989 Trap 22 Pima Co., AZd G. thurberi TW W 25 1996 31 47.900Њ N 110 47.800Њ W Hand 23 Santa Ana, Sonora G. thurberi TW W 3 2011 28 22.869Њ N 109 09.470Њ W Hand 24 Tampico, MXe G. hirsutum SE E 2 2001 22 14.500Њ N 97 51.500Њ W Trap 25 Brazos Co., TXe G. hirsutum SE E 1 2001 Trap

Listed are site number (also see Fig. 1), collection location, host plant species, boll weevil form (southeastern ͓SE͔, thurberia ͓TW͔, eastern Mexican ͓EMX͔, or western Mexican ͓WMX͔), geographical region (east ͓E͔ or west ͓W͔) of the central mountains in Mexico, sample size (N), collection year, lastitude and longitude, and collection method. Boll weevil samples from cultivated cotton in the United States and northeastern Mexico had no recorded latitude and longitude coordinates. a Not all samples were used for each locus and different individuals within a pop were sometimes used for different loci. b Latitude and longitude were unavailable at some site locations. c Collected by N. Barcenas in 1998. d Collections were made from multiple sites in relative proximity (Ͻ100 km). Latitude and longitude coordinates represent the center of the sampled locations. e Sequences were obtained from GenBank for these locations. and Gossypium davidsonii Kellogg, all previously iden- Polymerase Chain Reaction. Primers for sequenc- tiÞed as potential reproductive hosts of this insect ing the one mitochondrial (COI) and four nuclear (Cross 1973). Other host species, including Gossypium (AK, CAD, EF-1␣, and ITSII) loci were all obtained harknessii Bandegee and Gossypium gossypioides from the literature (Supp Table 1 [online only]). For Smith and Cothern, also were reassessed for presence the CAD locus, we ampliÞed a fragment of the car- of boll weevils (Cross 1973), but they were not found bamoylphosphate synthetase locus. Reaction condi- to support them. Boll weevil adults obtained from tions for nuclear genes included 40 ng of DNA tem- cultivated cotton in northeastern Mexico and the plate in 10-␮l reactions with 0.2 ␮M Promega dNTPs, ␮ ϫ southeastern United States were collected in phero- 2.5 M MgCl2 (Bioline, Taunton, MA), 1 NaCl buf- mone traps Kim and Sappington (2004a) and Kim et fer (Bioline), 0.4 U of Biolase TaqDNA Polymerase al. (2006, 2008). Weevils from locations 1 and 5Ð12 (Bioline), and 0.5 ␮M each primer. All primers were (Table 1; Fig. 1) were sampled by traps from areas synthesized by Integrated DNA Technologies (Cor- where no other hosts exist. Weevils from locations 2Ð4 alville, IA). A Programmable Thermal Controller-100 are within the range of G. thurberi, but they were thermocycler (MJ Research, St. Bruno, QC, Canada) sampled by traps placed in the center of cultivated was used for all polymerase chain reaction (PCR) cotton Þelds. In this case, we assumed sampled indi- reactions. For all nuclear genes, we used a touchdown viduals originated from cultivated G. hirsutum. Spec- method with thermocycler conditions: 95ЊC for 3 min, imens of Anthonomus hunteri (Burke & Cate), the 95ЊC for 30 s, 58ЊC for 60 s, and 72ЊC for 50 s. We proposed sister species of A. grandis (Burke et al. 1986, reduced the annealing temperature by 2ЊC every three Jones 2001), were included as an outgroup for phy- cycles through 42ЊC. In total, there were 27 cycles, logenetic analyses. followed by a Þnal extension at 72ЊC for 15 min. For DNA Extraction. We extracted DNA from boll wee- COI, we used thermocycling conditions of 94ЊC vil forelegs (and from boll weevil heads in cases of low for 3min, 30 cycles of 94ЊC for 60 s, 56ЊC for 60 s, 72ЊC DNA concentration) by using the Puregene Core kit for 60 s, and a Þnal extension at 72ЊC for 10 min. (QIAGEN, Valencia, CA), following the procedure Because of low DNA quality in samples from some described by Kim and Sappington (2004a). The DNA populations, we were only able to obtain sequence collections and vouchers are stored at Mission Labo- reads from a subset of individuals. The Tecoman sam- ratory, Edinburg, TX. ples were particularly affected by DNA degradation as November 2012 KUESTER ET AL.: BOLL WEEVIL GENETIC STRUCTURE AND DIVERSITY 905

Fig. 1. Map of A. grandis collection sites (location descriptions and site numbers correspond to Table 1). The cotton host species from which boll weevils were collected are indicated by three shapes: cultivated cotton, G. hirsutum, circle; G. thurberi, triangle; and other wild cotton host species, square. The dashed line demarcates eastern and western regions. a result of sample age, yielding limited data for the vergence were calculated with Mega v. 4.0 (Tamura et nuclear genes assayed. al. 2007), and the most parsimonious model of evolu- Sequencing. Sequencing was performed on a 3730 tion for each locus was determined using FindModel DNA Analyzer (Applied Biosystems) at the Iowa State (www.hiv.lanl.gov/content/sequence/Þndmodel/ University DNA Facility (Ames, IA) by using ABI Big Þndmodel.html). Dye v. 3.1 (Applied Biosystems), with 0.5 ␮M primer Genetic Evaluation of Boll Weevils Forms. We Þrst 1ϫ Big Dye buffer and 1ϫ Big Dye per 10-␮l reaction. addressed genetic support for four hypotheses of evo- Sequencing and PCR products were cleaned with Þne lutionary relationships among boll weevil forms (Fig. DNA grade Sephadex columns (GE Healthcare, Little 2, models 1Ð4) by identifying single nucleotide poly- Chalfont, Buckinghamshire, United Kingdom), fol- morphisms (SNPs) for each of the Þve loci that are lowing the manufacturerÕs instructions. We se- diagnostic of particular boll weevil forms. We also quenced COI from more individuals than other loci constructed maximum parsimony networks for each because we anticipated a higher mutation rate and locus with TCS v. 1.21 (Clement et al. 2000), by using resolution of intraspeciÞc genetic structure than for a 95% cutoff and resolving connection ambiguities the nuclear genes and because this locus was previ- (unique sequences that may have arisen from one of ously used for analysis of A. grandis (Scataglini et al. several possible ancestral sequences) following the 2006). Nucleotide sequences described in this article methods of Pfenninger and Posada (2002). Boll weevil have been deposited in GenBank under the follow- forms were mapped onto these networks to evaluate ing accession numbers: AK, JQ894407ÐJQ894430; support for models 1Ð4 (Fig. 2). CAD, JQ894431ÐJQ894452; COI, JQ894340ÐJQ894406; Analysis of molecular variance (AMOVA) and EF-1␣, JQ894471ÐJQ894490; and ITSII, JQ894453Ð Bayesian phylogenetic analyses were conducted for JQ894470, AY882992ÐAY83003, EF194205ÐEF194224, each individual locus and for a combined dataset of and EU215423. COI, AK, EF-1␣, and CAD. ITSII was omitted from the Data Analysis. Sequence Alignment and Evaluation. combined analysis because of small sample sizes re- All sequences were aligned using ClustalW (Thomp- sulting from problems amplifying homologous se- son et al. 1994) in Bioedit (Hall 1999) and translated quences within some samples from western Mexico. to detect stop codons and frame-shifts, which may We conducted AMOVA using GenAlEx (Peakall and indicate the presence of pseudogenes. Unique se- Smouse 2006) to quantify genetic differentiation quences were identiÞed with Collapse v. 1.2 (Posada ␾-statistics) for a three-level hierarchical model con- 2011) and used in subsequent evolutionary tree con- sisting of forms, populations within forms, and indi- struction. Sequences for each locus were analyzed viduals within populations. We also conducted separately. The number of polymorphic sites, number AMOVA to quantify differentiation among boll wee- of informative sites, number of unique sequences, se- vils collected from the Þve different host species in quence diversity, nucleotide diversity (and its vari- western Mexico (Table 1). For all AMOVA models, ance), and theta (4N␮ and its standard error) were signiÞcance of differentiation was determined by per- estimated with DnaSP v. 5 (Librado and Rozas 2009). mutation (1,000 replicates). Because the AMOVA cal- Frequencies of unique sequences and sequence di- culated in GenAlEx is based on Euclidean distances 906 ANNALS OF THE ENTOMOLOGICAL SOCIETY OF AMERICA Vol. 105, no. 6

To determine phylogenetic support for models 1Ð4 (Fig. 2), we constructed trees for each locus using Bayesian methods, where hypothesized groups of in- terest (forms or regional collections) were con- strained to monophylies. All Bayesian tree searches were performed using MrBayes v. 2.03 (Huelsenbeck et al. 2001) on the CIPRES Science Gateway (Miller et al. 2010). The best models of evolution for each locus were as follows: COI, General Time Reversible (GTR) ϩ Gamma; AK, Tamura-Nei; EF-1␣, Tamura- Nei ϩ Gamma; CAD, TamuraÐNei ϩ Gamma; and ITSII, GTR. Because computational time was not a concern given the size of our data set and the fact that the closest model to the TamuraÐNei model in Mr- Bayes was GTR, we used a GTR model with invariant sites (Tavare 1986) for all sequence types. Searches were run with four simultaneous chains sampling every 1,000 generations for 1,000,000. Temperature of chain swapping was Þrst assessed at 0.2, 0.5, and 1.0. Because we saw no differences in topology or support among the temperatures, we reported results only for 0.2. All trees were rooted with a sample of A. hunteri, a closely related species in the A. grandis species group (Jones 2001). For each locus and the combined data set (excluding ITSII), we evaluated the Þt of the data to models Fig. 2. Four hypotheses (models) of evolutionary rela- 1Ð4 (Fig. 2) by using Bayes factors that were calcu- tionships among cotton boll weevil forms. (a) Model 1: the lated as the ratio of marginal likelihood values of two three traditionally recognized forms SE, TW, and MX, where TW is sister to the clade of SE and MX, the latter comprising competing models. For interpretation purposes we individuals from both east and west of MexicoÕs central worked with the absolute value of twice the natural mountain ranges. (b) Model 2: eastern (E) versus western logarithm of the Bayes factor Kass and Raftery (1995): (W) boll weevils (E, combining SE and EMX are sisters to if between 0 and 2, the difference in support for the two distinct forms of western weevils (WMX and TW). (c) two competing models is “not worth more than a bare Model 3: three boll weevil forms of unclear relationship. (d) mention”; between 2 and 6, there is positive support Model 4: (E) and (W) forms without consideration of the for the model with the higher marginal likelihood; three traditional forms (SE, MX, and TW). between 6 and 10 there is strong support; or Ͼ10, there between individuals, we also compared the AMOVA is very strong support. results for each locus with those of the closest model Contingent on results of the AMOVA and Bayesian of evolution in Arlequin v. 3.1 (described below; Ex- analyses described herein, we conducted additional cofÞer et al. 2005). We found similar model support analyses framed with respect to the hypotheses of the with both methods, so we only reported AMOVA three- or two-form model best supported by the data. results from GenAlEx. We used partial Mantel tests using the Ecodist package To evaluate population genetic support for the four (Goslee and Urban 2007) in R v. 2.12.1 (R Core Team models of boll weevil diversiÞcation (Fig. 2), we com- 2010) to evaluate the joint effects on pairwise genetic pared coefÞcient of determination and AkaikeÕs in- distances between populations of 1) geographic dis- formation criterion with correction for small sample tance and geographical region (eastern versus west- size (AICc) values from associated AMOVA analyses. ern Mexico) and 2) boll weevil form (SE, MX, and To assess performance of models by R2 we simply TW) and geographical distance. SigniÞcance of tested ranked models based on the amount variation ex- effects on genetic distance was obtained from pairwise plained. Although comparison of R2 values is qualita- adjusted mean bootstrap values. We report this test for tive, the difference in AICc values between models COI, AK, CAD, and EF-1␣ (excluding ITSII because (⌬AICc) provides a more formal approach to model of small sample sizes. selection (Burnham and Anderson 2002) and was ap- Isolation by Distance. Population pairwise linear- Ϫ plied to AMOVA models following Halverson et al. ized FST estimates (FST/[1 FST]) (Rousset 1997) (2008). The larger AICc of two compared models is were calculated for each pair of populations in Arle- considered to have considerably less support when quin v. 3.1 (ExcofÞer et al. 2005) and regressed on the 4 Ͻ⌬AICc Ͻ 7. When AICc Ͼ10, the model with the natural log of interpopulation distance (km) to test for larger ⌬AICc is considered to have no support and the isolation by distance (IBD) using a Mantel test (1,000 smaller AICc is preferred. When ⌬AICc Ͻ2, the com- permutation replicates; Manly 1986). We reported peting models are considered to have comparable sup- results of this analysis only for COI because this locus port. had the highest haplotype diversity and provided the November 2012 KUESTER ET AL.: BOLL WEEVIL GENETIC STRUCTURE AND DIVERSITY 907

Table 2. Sequences represented in populations of boll weevil in Mexico and the United States for ﬁve loci (deﬁned in Supp Table 1 ͓online only͔)

COI AK CAD EF-1a ITSII Site no. Location Unique sequences Unique Unique Unique N (no.) N sequences N sequences N sequences N Unique sequences 1 Weslaco, TX 3 63, 51, 53 2 11 (2) 5 4 (2), 5 (2), 8 3 7 (3) 5 4 (2), 10, 24 (2) 2 Tlahualilo, Durango 5 61, 57, 60 (3) 2 1 (2) 2 9 (2) 2 7 (2) 3 10, 24 (2) 3 Ojinaga, Chihuahua 3 60 (3) 2 5 (2) 2 7 (2) 6 10, 21,22, 24 (3) 4 Rosales, Chihuahua 3 51, 56, 58 5 16, 23, 24 (3) 5 Lubbock, TX 5 63, 51 (2), 53, 56 2 11 (2) 3 2, 5, 11 3 7 (3) 3 24 (3), 26 6 Childress, TX 4 64, 66, 52, 54 7 Artesia, NM 4 51(2), 54, 68 2 11, 18 3 1, 2, 10 2 6, 7 3 10 (3) 8 Hobart, OK 4 51(2), 54(2) 2 6 (2) 2 9 (2) 9 Little Rock, AR 3 51 (3) 2 11 (2) 3 2 (2), 10 3 7 (2), 8 10 Brownsville, TN 2 64(2) 2 11 (2) 3 7 (3) 11 Malden, MO 4 51(2), 54, 64 2 2 (2) 2 7 (2) 12 Cleveland, MS 3 64 (3) 2 14 (2) 2 2, 3 4 10 (2) 24 (2) 13 La Ventosa, Oaxaca 3 55, 59, 62 3 2, 13, 22 2 6, 7 3 10, 13, 14 14 Tecoman, Colima 3 28, 39, 46 3 12, 15, 17 (G. aridum) 15 Tecoman, Colima 5 42, 43, 45, 47, 48 2 11 (2) 7 12 (7) 2 17 (2) 16 El Cajon Presa, Nayarit 5 43, 38, 29, 40, 49 2 19 (2) 2 12 (2) 17 Cajeme, Sonora 7 41, 43, 5, 29, 34, 3 3, 7, 11 4 12 (4) 2 17 (2) 1 4 35, 38 18 Rncho San Ramon, 6 43, 8, 9, 32 (2), 37 3 8 (2), 11 10 12 (8), 15, 21 2 10, 17 2 4, 15 Sonora 19 San Carlos, Sonora 11 4, 5, 6, 7, 12, 29, 2 11, 20 10 12 (7), 13, 14, 18 3 5, 13, 14 1 7 30, 31, 33, 36,43 20 Baja California Sur 15 13 (4), 15, 16 (2), 5 5, 8, 9, 10, 21 14 7, 12 (5), 13 (5), 8 1,2,3,4,5,11, 7 4, 3 (2), 5, 6, 8, 9 17, 18 (2), 19, 17 (2), 20 12, 15 20, 23, 24, 50 21 Laveen, Arizona 6 13, 14, 25, 26, 27, 5 4, 8, 11, 16, 3 4,6,12 2 16(2) 2 4,17 29 23 22 Pima Co, Arizona 10 29, 3, 10, 11, 21, 1, 2 19 (2) 6 5 (3), 12, 16, 19 3 9 (2), 19 24 1, 2, 3 (13), 16, 17 2, 7, 22, 44 (5), 18, 19, 20 23 Santa Ana, Sonora 2 7 (2) 2 11 (2) 2 18 (2) 2 11, 12 24 Tampico, MX 2 25 (2) 25 Brazos, TX 124

Indicated are the site number (corresponding to site locations in Fig. 1 and Table 1), numbers of samples sequenced per locus from each site (N), and the identiÞcation number of each sequence found at a site (number of times a sequence was found at a site is given in parentheses if Ͼ1). greatest resolution of IBD over a range of geographical Results scales. Coalescent Testing. We used the program IMa (Hey Data on sequences represented in each boll weevil and Nielsen 2007) to estimate effective migration rates sample population are summarized in Table 2. and divergence time between eastern (EMX) and Genetic Evaluation of Boll Weevils Forms. We western (WMX) boll weevil populations deÞned in found one Þxed SNP difference to completely distin- Fig. 2. The Hasegawa, Kishino, and Yano (HKY) guish each of the three traditionally identiÞed forms model of evolution was used for each locus because it (SE, TW, and MX; models 1Ð3) at alignment position was the closest model available in IMa. Posterior sam- 152 in COI but found no other SNPs to completely pling came after a burn-in of 100,000 iterations and differentiate these forms (Table 3). In addition, no lasted for 10,000,000 steps with two chains. By assess- other locus could differentiate TW from MX (Table ing the effective sample size, comparing results from 3). In contrast, we observed Þve Þxed SNPs in COI at least three independent runs, and inspecting trend (positions 152, 164, 227, 278, and 362) and one Þxed plots for each parameter, we concluded this runtime SNP in EF-1␣ (position 471) to distinguish eastern length to be sufÞcient to reach reproducible param- versus western forms (Table 3). We also found seven eter estimates. Because speciÞc mutation rates are Þxed SNPs to distinguish the SE versus MX forms (Þve unknown for most genes used in this study, we eval- COI, one CAD, and one EF-1␣), 11 SNPs to distinguish uated a range of mutation rates for nuclear genes ␣ (10Ϫ6,10Ϫ7, and 10Ϫ8 base pair substitution per gene SE versus TW (seven COI, one EF-1 , and three per year), as suggested by Carstens et al. (2005), to ITSII), and one SNP to differentiate MX versus TW assess the range of estimates of time because of pop- (COI) (Table 3). ulation divergence. For COI, we used a mutation rate We found sequences that were unique to and of 1.5 ϫ 10Ϫ8 base pair per gene per year (Papado- shared between the largely western TW and MX forms poulou et al. 2010). Although estimates of historical and absent from the exclusively eastern SE form. The demography are subject to large variances, we pre- number of unique sequences shared out of total num- sented these estimates of migration and time diver- ber of unique sequences per locus were as follows: gence to aid in assessing form divergence in the face COI, 1/52; AK, 2/15; CAD, 1/16; EF-1␣, 0/19; and of ancestral polymorphism and recurrent gene ßow. ITSII, 1/22 (Table 2). 908 ANNALS OF THE ENTOMOLOGICAL SOCIETY OF AMERICA Vol. 105, no. 6

Table 3. Fixed differences in nuclear and mitochondrial SNPs AMOVA models evaluating model 4 (Fig. 2d) found between the A. grandis grandis forms or regional classiﬁca- showed signiÞcant differentiation among eastern and tions deﬁned in Fig. 2 western boll weevil forms for all Þve loci (Supp Table Three traditional Eastern vs 4 [online only]). Models 2 (eastern versus western Aligned Locusa forms western boll weevils with distinct thurberia and western forms, position SE TW MX EBW WBW Fig. 2b) and 3 (traditional boll weevil forms with unclear evolutionary relationships, Fig. 2c) explained COI 152 A G C A G/C COI 164 A C/T C/T A C/T signiÞcant genetic variation for all loci but AK (Supp COI 227 C T T C T Table 4 [online only]). Model 1 (three forms [SE, TW, COI 253 G/C A A/G and MX] where TW is sister to a clade of SE and MX, COI 278 A T T A T Fig. 2a) showed signiÞcant differentiation only for COI 362 G T T G T COI 380 G A G/A ITSII (Supp Table 4 [online only]). AK NA Comparing AMOVA models by ⌬AICc (Table 5), CAD 75 C C/T T the east versus west model (model 4, Fig. 2d) was ␣ EF-1 465 T C C T C supported over, and explained more genetic variation ITSII 47b C T T/C ITSII 296b A G A/G than, models 1Ð3 in all cases except for ITSII and for ITSII 419b A C A/C the combined data set. Also, there was inconsistent support among loci when comparing models 1 and 3, a Locus names are as deÞned in Supp Table 1 ͓online only͔. b with the exception of ITSII and the combined dataset. Difference found in all samples except one sample from Rosales, In contrast, ITSII strongly supported model 1 over Mexico, here denoted SE. models 2 and 4. The combined data set, however, could not differentiate a preferred model. Consistent with the maximum parsimony networks, Haplotype and nucleotide diversity ranged from Bayesian phylogenetic analyses of COI and EF-1␣ 0.495 to 0.986 (mean ϭ 0.836) and 0.004Ð0.027 ϭ showed a clear distinction between eastern forms (mean 0.014), respectively (Supp Table 2 [online (EMX and SE) and western forms (WMX and TW) only]). The theta estimate in western populations was (Fig. 4a and d), with trees for the remaining three loci higher than eastern populations for three loci (COI: ␪ ϭ ␪ ϭ ␣ ␪ ϭ ␪ ϭ also containing some eastern and western speciÞc W 1.0544, E 0.137; EF-1 : W 0.677, E 0.003; clades (Fig. 4b, c, and e). In contrast to ⌬AICc eval- ␪ ϭ ␪ ϭ and ITSII: W 0.095, E 0.022) (Supp Table 2 uation of AMOVA models, evaluation of the support of [online only]) and lower than eastern populations for Bayesian phylogenies for models 1Ð4 using two times ␪ ϭ ␪ ϭ ␪ ϭ two loci (AK: W 0.023, E 0.032; CAD: W 0.031, the natural logarithm of the Bayes factor showed no ␪ ϭ E 0.074) (Supp Table 2 [online only]). distinction between any models across all loci and The maximum parsimony networks for COI and combined locus dataset (all values Ͻ2; Supp Table 5 ␣ EF-1 showed a clear distinction between eastern [online only]). (EMX and SE) and western forms (WMX and TW) Partial Mantel tests indicated a signiÞcant effect on (Fig. 3a and d), a distinction also largely supported by genetic distance of geographical distance taking into ITSII (Fig. 3e) but not the remaining two loci (Fig. 3b account eastern versus western region for CAD and and c). EF-1␣, and, conversely, of region given geographic With the exception of AK, AMOVA analyses re- distance for COI and CAD (Table 6). We also found vealed signiÞcant divergence between east and west a signiÞcant effect on genetic distance of traditionally ␾ ϭ ϭ ␾ ϭ regions (COI: -RT 0.567, P 0.001; AK: RT deÞned boll weevil form (SE, TW, and MX) given ϭ ␾ ϭ ϭ ␣ 0.041, P 0.025; CAD: RT 0.428, P 0.001; EF-1 : geographic distance for COI, CAD, and EF-1␣; and of ␾ ϭ ϭ ␾ ϭ ϭ RT 0.190, P 0.001; and ITSII: RT 0.213, P geographic distance accounting for form for CAD (Ta- 0.001) (Table 4). We also found signiÞcant differen- ble 6). tiation among populations within the eastern region Isolation by Distance. We observed a minor but ␾ ϭ ϭ for COI, AK, and CAD (COI: PR 0.269, P 0.001; signiÞcant correlation between population-level ge- ␾ ϭ ϭ ␾ ϭ ϭ AK: PR 0.574, P 0.005; and CAD: PR 0.400, P netic differentiation and geographic distance by ␣ ␾ ϭ ϭ 2 0.001) (Table 4) but not EF-1 ( PR 0.177, P means of the Mantel test over all samples (R ϭ 0.029, 0.164) (Table 4). Similarly, we found signiÞcant dif- P ϭ 0.001), all western samples (R2 ϭ 0.029, P ϭ 0.014), ferentiation among western populations for COI, AK, and western samples omitting those from the Baja ␣ ␾ ϭ ϭ ␾ ϭ 2 and EF-1 (COI: PR 0.274, P 0.001; AK: PR Peninsula (R ϭ 0.076, P ϭ 0.005). In contrast, there ϭ ␣ ␾ ϭ ϭ 0.191, P 0.003; and EF-1 : PR 0.181, P 0.016) was no signiÞcant IBD pattern over the range of south- ␾ ϭ ϭ 2 (Table 4) but not CAD ( PR 0.064, P 0.059) eastern boll weevil (R ϭ 0.003, P ϭ 0.220) or among (Table 4). We also found signiÞcant differentiation mainland Sonoran Desert sites (R2 ϭ 0.081, P ϭ 0.193). among boll weevils collected from different host spe- Coalescent Results. Although we observed genetic cies within the Sonoran Desert (populations: 14Ð20) evidence of vicariance between boll weevils from ␣ ␾ ϭ ϭ at COI, AK, and EF-1 (COI: HT 0.231, P 0.001; eastern and western Mexico, we also detected low ␾ ϭ ϭ ␣ ␾ ϭ AK: HT 0.141, P 0.050; and EF-1 : HT 0.157, rates of bidirectional gene ßow between the two re- P ϭ 0.019) (Supp Table 3 [online only]), with weaker gions using IMa. Eastward and westward migration ␾ ϭ but signiÞcant differentiation at CAD ( HT 0.051, rates averaged 1.23 and 0.61 individuals per genera- P ϭ 0.033) (Supp Table 3 [online only]). tion, respectively. The estimate of divergence time November 2012 KUESTER ET AL.: BOLL WEEVIL GENETIC STRUCTURE AND DIVERSITY 909

Fig. 3. Maximum parsimony sequence networks of A. grandis grandis constructed from DNA sequence data from genes COI (a), AK (b), CAD (c), EF-1␣ (d), and ITSII (e). Networks were constructed using 95% conÞdence levels; connections within networks outside of these conÞdence intervals are indicated with dotted lines in the networks where they occur. Sequences are indicated by the identifying numbers given in Table 2. Node size is proportional to the number of sample individuals represented by a sequence. Small black circles and ticks through connecting lines represent hypothetical sequences indicating the number of mutational changes between sampled sequences. Shading and hatching of nodes indicates geographical region or boll weevil form: SE form (light gray), western MX (white), eastern MX (vertical lined), and TW form (dark gray). since the original vicariance event ranged between tify a SNP in COI (position 152) distinguishing TW 0.22 and 141.39 million yr ago, representing a 95% and MX (Table 3), and ⌬AICc support for model 1 and conÞdence envelope. its three traditional forms (including a distinct TW) for ITSII. Although Roehrdanz (2001) and Roehrdanz et al. (2010) identiÞed mitochondrial (mt)DNA and Discussion ITSII markers, respectively, that are diagnostic among Overall, model tests using AMOVA and ⌬AICc sug- the three forms, precise discernment between trapped gest that model 4 (Fig. 2), representing eastern and TW and SE forms was not possible. In our analysis, western forms, best explains the historical and evolu- ITSII was a poor diagnostic of the TW and MX forms, tionary relationships within the boll weevil (Table 5). yet adequate for distinguishing the TW and SE forms. Although we found TW to be genetically indistin- Our genetic data support the idea that boll weevils guishable at AK, CAD, and EF-1␣ from other western traditionally classiÞed as the SE form differ from pop- samples, including the putative MX form, we did iden- ulations traditionally classiÞed as the TW form and the 910 ANNALS OF THE ENTOMOLOGICAL SOCIETY OF AMERICA Vol. 105, no. 6

Fig. 3. Continued.

MX form in the west. Thus, our data best support only ginning of the past century. Results presented here two forms, eastern and western, with the caveat that strongly reßect these events in the relatively low ge- more samples from southern Mexico, Central Amer- netic diversity found in all sequences analyzed from ica, and other hosts may reveal other forms. From the southeastern United States compared with popu- limited sampling in Oaxaca, however, we do observe lations on wild and cultivated cotton from Mexico. genetic similarity between the SE form and samples We found signiÞcant differentiation among popu- from southeastern Mexico. Populations from eastern lations from different host species within western and western sites of North America showed signiÞcant Mexico; yet, little genetic distinction between sym- differentiation, especially at the EF-1␣ and COI loci, patric populations of boll weevil collected from G. with a number of Þxed SNPs by region (Table 3). The thurberi and those from other wild and cultivated question of whether the traditional SE form differs cottons in Sonora (Fig. 3a, b, and e). Behavioral dif- from the MX form east of the mountains was not ferences between the traditional TW and MX weevil completely addressed and would beneÞt from more forms, in particular in host preference, support the samples. The step cline in variation in morphological notion that there are distinct forms of boll weevil. For characters between the SE form and eastern MX form example, cultivated cotton Þelds in Arizona that are that occurs near the Lower Rio Grande Valley recog- located near natural populations of G. thurberi remain nized by Burke (1968) and Burke et al. (1986) is unaffected by TW populations posteradication striking and form the basis for hypothesizing an SE (Roehrdanz 2001). Our genetic data, however, do not form. The uniform morphology of the SE form found support a thurberia form phylogenetically distinct north of the Rio Grande Valley is evidence of the from the traditional MX form, at least those from our initial founder event and subsequent rapid dispersal western sample locations. Burke et al. (1986) postu- throughout the southeastern United States at the be- lated that temporal isolation between the two forms November 2012 KUESTER ET AL.: BOLL WEEVIL GENETIC STRUCTURE AND DIVERSITY 911

Table 4. Hierarchical AMOVA analyses for each locus under the three-form (SE versus MX versus TW) hypothesis of boll weevil diversiﬁcation; differentiation among regions (eastern versus western); among populations in the eastern region; among populations in the western region; among populations in the western region excluding samples from the Baja Peninsula; and among samples obtained from different Gossypium host species in the western region

Locus Variance componenta df P -statistics COI Among weevil forms 2 FT 0.010 0.498 Among regions (eastern vs western) 1 RT 0.001 0.567 Among pops eastern 14 PR 0.001 0.269 Among pops western 9 PR 0.001 0.274 Among pops western (exclude Baja) 8 PR 0.002 0.175 Among species within western 4 SR 0.001 0.231 AK Among weevil forms 2 FT 0.032 0.059 Among regions (eastern vs western) 1 RT 0.025 0.041 Among pops eastern 9 PR 0.005 0.574 Among pops western 9 PR 0.003 0.191 Among species within western 4 SR 0.141 0.051 CAD Among weevil forms 2 FT 0.001 0.47 Among regions (eastern vs western) 1 RT 0.001 0.428 Among pops eastern 10 PR 0.001 0.4 Among pops western 7 PR 0.059 0.064 Among species within western 4 SR 0.033 0.051 ␣ EF-1 Among weevil forms 2 FT 0.01 0.153 Among regions (eastern vs western) 1 RT 0.001 0.19 Among pops eastern 6 PR 0.164 0.177 Among pops western 6 PR 0.016 0.181 Among species within western 4 SR 0.019 0.157 b ITSII Among weevil forms 2 FT 0.010 0.292 Among regions 1 RT 0.010 0.264

a The variance component among pops western (exclude Baja) is tested only for COI, because our sample of the COI locus for Baja is unusually large. b Within-region sample sizes for ITSII were too small to perform within-region analyses. was sufÞcient to produce behavioral barriers to sub- Desert and Colima, Mexico (Supp Table 3 [online stantial gene ßow but was not long enough to prevent only]), likewise do not support the classiÞcation of a all interbreeding when sympatric. Maintenance of be- TW form distinct from the MX form; rather, our se- havioral differences between distinct forms of a spe- quence data support a western form that subsumes the cies despite hybridization and introgression at neutral TW. The distinct mtDNA restriction fragment-length loci is certainly possible, as in the case of pheromone polymorphism haplotypes that Roehrdanz (2001) re- races of European corn borer, Ostrinia nubilalis (Hu¨ b- ported for TW were compared with SE populations, ner) (Lepidoptera: Crambidae) (Dopman et al. 2005). not MX populations, and are consistent with our pro- Our Þndings do not support the designation of three posal of eastern and western forms. It will be impor- forms (TW, SE, and MX), in part because there is no tant to compare those same markers between boll sequence distinction between populations of TW and weevil populations on G. thurberi and cultivated cot- adjacent populations on cultivated or other wild cotton, although this is no longer straightforward given tons in the western region, classically deÞned as the the successful eradication of the latter from Arizona MX form, with the exception of alignment position 152 and California. However, populations from further in the COI locus. Lack of Þxed nucleotide differences south in western Mexico would provide relevant com- and of substantial differentiation among boll weevils parisons. Our prediction is that, like the markers re- collected from speciÞc cotton hosts in the Sonoran ported in this study, the restriction fragment length

Table 5. Evaluation of the four models of evolutionary relationship among boll weevil forms presented in Fig. 2 by using hierarchical AMOVA

Model Model Model Model Best Locus 1 vs 2 1 vs 3 1 vs 4 2 vs 3 2 vs 4 3 vs 4 1 2 3 4 model COI 9.53 (2) 0.14 67.58 (4) 9.39 (2) 58.06 (4) 67.45 (4) 0.371 0.419 0.371 0.641 4 AK 0.88 0.35 39.14 (4) 0.53 38.25 (4) 38.79 (4) 0.0718 0.081 0.0718 0.558 4 CAD 4.46 (1) 0.21 35.03 (4) 4.67 (3) 39.49 (4) 34.82 (4) 0.336 0.297 0.336 0.565 4 EF-1␣ 1.79 0.41 19.6 (4) 166.82 (3) 17.8 (4) 19.18 (4) 0.144 0.17 0.144 0.437 4 ITSII 40.46 (1) 204.19 (1) 190.95 (1) 163.72 (2) 150.49 (2) 13.24 (4) 0.256 0.25 0.22 0.131 1 Combined 0.26 0.12 1.42 0.13 1.68 1.55 0.28 0.27 0.23 0.16 NA

Shown for each of Þve loci (deÞned in Supp Table 1 ͓online only͔) are sample sizeÐcorrected ⌬AICc values for each pairwise comparison between competing models 1Ð4 (with the preferred model indicated in parentheses), R2 values for each model, and the overall best model based on ⌬AICc. Descriptions of ⌬AICc can be found under Data Analysis. Combined results reßect the combined analysis of four loci: COI, AK, CAD, and EF-1␣. 912 ANNALS OF THE ENTOMOLOGICAL SOCIETY OF AMERICA Vol. 105, no. 6

Fig. 4. Bayesian phylogenetic tree analysis of cotton boll weevil samples collected in the United States and Mexico. Results are presented for unique sequences from each of Þve loci: COI (a), AK (b), CAD (c), EF-1␣ (d), and ITSII (e). Branch labels are posterior probability support values for each clade. Branches with support Ͻ0.50 were collapsed into a polytomy. Sequence IDs are as in Table 2 and are followed by the sample location east or west of the central mountainous divide in Mexico and the United States. Also shown is the form of weevil (MX, TW, and SE) represented by each unique sequence. Each tree was rooted using A. hunteri as outgroup (not shown). Information on which populations and boll weevil forms are represented by each unique sequence is described in Table 2. November 2012 KUESTER ET AL.: BOLL WEEVIL GENETIC STRUCTURE AND DIVERSITY 913

Table 6. Results from partial Mantel tests regressing pairwise genetic distance (substitutional differences) on geographic distance (geography) and region (eastern versus western), and genetic distance on geographic distance and boll weevil form (SE, TW, and MX)

Geography region Region geography Form geography Geography form Locus rPrPrPrP COI 0.028 0.339 0.925 Ͻ0.001 0.367 0.001 Ͻ0.001 0.296 AK 0.026 0.337 0.027 0.351 0.025 0.889 Ͻ0.001 0.812 CAD 0.294 Ͻ0.001 0.318 Ͻ0.001 0.990 0.001 0.425 0.019 EF-1␣ 0.243 0.013 Ϫ0.01 0.470 0.457 0.028 Ͻ0.001 0.696

Shown for each locus and model are partial correlations (r) and associated P values (P) for the effect of the Þrst factor listed in a column, taking into account the second-listed factor. polymorphism markers will not indicate substantial boll weevils have been residing on cultivated cotton in differentiation between TW and western Mexico pop- this region for longer than the SE form has been on ulations. cultivated cotton in the southeastern United States Although the three form classiÞcation models (Escobar-Ohmstede 2004) that may explain the (models 1 and 3, Fig. 2) are not as well supported as greater sequence and nucleotide diversity in the west- the east versus west classiÞcation, distinction of TW ern boll weevil clade (Table 2; Supp Table 2 [online from WMX may still be useful and warranted in an only]). eradication context, as long as observations continue It is still unclear what events initiated the radiation to support the behavioral difference in host prefer- of A. grandis from Hampea populations onto cotton. ence. Development of an assay based on morpholog- The expansion of cotton cultivation in Mexico may ical characters (Burke 1968, Burke et al. 1986) com- have established a new niche for A. grandis from Ham- bined with a molecular assay developed to genotype pea on cultivated G. hirsutum, with subsequent host the COI locus at position 152 (Table 3) could be shifts to adjacent wild host species. Alternatively, the particularly powerful and mutually supporting when a Þrst host switch away from Hampea could have been positive identiÞcation of form is needed for an indi- a single event to one of several cotton species endemic vidual captured in an erstwhile eradicated zone. The to south central Mexico, followed by subsequent former would indicate the host history, and the latter spread to other wild cotton species and cultivated would indicate the phylogenetic form and thus its cotton. The COI sequence most closely related to the presumed host preference. Together, the dual ap- outgroup was from a western boll weevil (Fig. 3a), proach would help mitigate the uncertainties arising whereas the most common COI sequence (51) was from imperfect isolation of the two forms in areas of restricted to eastern samples. These results suggest close sympatry between G. thurberi and G. hirsutum. that more ancestral genetic diversity remains in the Weevil Radiation and Dispersal. Our results are in western region and that variation was lost as the boll accordance with the Burke et al. (1986)Gossypium weevil expanded its range northward to northeastern host than by geographical region (east versus west). Mexico and the United States (Fryxell and Lukefahr MexicoÕs central Sierra Madre Occidental seems to be 1967, Burke et al. 1986, Jones 2001), an observation a signiÞcant barrier to gene ßow between these two reported across many marker systems (allozymes, Ter- regions with migration occurring instead south to ranova et al. 1990; mitochondrial DNA restriction frag- north along the respective coastlines in either region. ment length polymorphisms, Kim and Sappington The G. aridum-associated population in Oaxaca, 2004a; randomly ampliÞed polymorphic DNA, Kim southern Mexico, has sequences similar to those found and Sappington 2004b, and simple sequence repeats, in northeastern Mexico (Fig. 3a, d, and e), supporting Kim and Sappington 2006). the hypothesis that the cotton-host origin of the boll We found fewer haplotypes in eastern populations weevil likely occurred in southern Mexico. than in the west for several loci (Fig. 3a, b, and d), The historical host shift of boll weevil from Hampea which may reßect founder effects associated with the to Gossypium is believed to have occurred east of the rapid range expansion of eastern populations out of mountains in southern Mexico, in Chiapas or the Oax- northeastern Mexico and through the Cotton Belt in acan Valley (Warner and Smith 1968, Jones 2001). the twentieth century. In contrast, geographical ex- Although the genotypes of the eastern boll weevil pansion northward and host shifts onto additional cot- subgroup seem derived from those of the western ton species in western Mexico probably occurred subgroup relative to the A. hunteri outgroup (Fig. 4d), much earlier and over a greater time span, allowing for this result may reßect inadequate sampling in eastern the evolution of greater diversiÞcation. Northern ex- Mexico. Boll weevils collected from cultivated cotton pansion and diversiÞcation of the boll weevil in west- in Arizona seem to fall within a western clade (Figs. ern Mexico also may have been aided by connectivity 2 and 3), which further supports the Burke et al. of populations on several wild western cotton species (1986) hypothesis of a two-coast northern expansion. with overlapping distributions (Burke et al. 1986). Anecdotal evidence of boll weevil sightings in the More recently, this effect may have been enhanced by eighteenth century on cultivated cotton in northwest- the prevalence and spatial homogeneity of small-scale ern Mexico also suggests that populations of western cotton cultivation in the west compared with the east. 914 ANNALS OF THE ENTOMOLOGICAL SOCIETY OF AMERICA Vol. 105, no. 6

Although Kim and Sappington (2004a,b; 2006) been several host shifts among cottons that subse- found signiÞcant patterns of IBD across the entire quently led to rapid expansion. Because G. aridum is southeastern boll weevil range and within regional a known contemporary host of boll weevil (Cross clusters of populations by using COI, we detected no 1973; A.P.K., personal observation) and has a large IBD for southeastern boll weevil. Our failure to detect range from southern to northwestern Mexico, this signiÞcant IBD patterns is not surprising given our species may have been especially important for boll lower sampling within the southeastern United States. weevil spread northward to the Sonoran Desert. Be- Although Kim and Sappington (2004a,b; 2006) did cause local boll weevil populations on G. aridum are detect signiÞcant IBD, the magnitude of genetic dif- apparently small (A.P.K., personal observations), con- ferentiation between populations was actually low to tinuous pre-Colombian cultivation of cotton through- modest given the extensive geographical area encom- out Mexico (Rodriguez-Vallejo 1976) may have con- passed by their study. However, we detected low but tributed to expansion through the west, although our signiÞcant IBD among the western samples, including Þnding of an SNP distinguishing TW and MX forms populations from mainland Mexico alone, and for the and signiÞcant IBD in western Mexico indicate lower combined data set of samples from mainland Mexico contributions of cotton cultivation in the northward and the Baja Peninsula. expansion of boll weevils on the western coast. Coalescent models indicate historically greater mi- Our results suggest that a reassessment of intraspe- gration from east to west (east to west, 1.23 migrants ciÞc boll weevil classiÞcation may be necessary. per generation; west to east, 0.61 migrants per gener- Rather than forms deÞned by host-speciÞcity (TW) ation) The historical expansion northward along the and a vague region type (MX), we have found that the west coast likely occurred before cotton cultivation mountainous region running north to south through and involved exploiting wild cottons as hosts. The boll the center of Mexico likely represents a major barrier weevil seems to have been historically more wide- to gene ßow, resulting in vicariance and differentia- spread in western Mexico, and we might have antic- tion of eastern and western boll weevil forms. Though ipated more gene ßow from west to east. However, we do detect some connectivity between the east and cotton cultivation allows the buildup of large boll west through bidirectional migration, genetic differ- weevil populations and thus would have generated entiation is high and even Þxation of unshared SNPs at greater propagule pressure from the east than from the multiple loci were observed between populations west. from the two regions. These differences may form the Host Adaptation and Pest Status. Although we did basis for diagnostic markers to distinguish boll weevils not detect Þxed differences among host-associated native to eastern and western parts of Mexico where boll weevil populations in the western region, we did cotton is cultivated. Greater insight into the origin of Þnd signiÞcant genetic differentiation among popula- the boll weevil and its association with cotton will tions collected from different host species in Sonora. require additional sampling in southern parts of Mex- However, we did not Þnd a signiÞcant correlation ico. between genetic distance and host species having taken into account geographic proximity (R ϭ 0.03, P ϭ 0.118). Acknowledgments Boll weevils are differentially attracted to plant volatiles, leaf color, and gossypol content among races We thank N. Barcenas, K. Day, D. Gates, H. Yu, and students in the Department of Biology, Universidad Au- of G. hirsutum (McKibben et al. 1977, Hedin and to´noma de Quere´taro, for help with Þeld collections. We also McCarty 1995, Allen 2008). It is possible that local thank F. Janzen, M. OÕNeal, J. Wendel, H. Burke, and an populations have adapted to characteristics of speciÞc anonymous APHIS reviewer for helpful comments to im- hosts and may be preferentially attracted to local cot- prove the manuscript. Funding for this work was obtained ton species for feeding and oviposition. This in turn from Cotton Incorporated, the National Cotton Council, and could affect the degree of connectivity with popula- National Science Foundation grant DEB-0543102 (to J.D.N. tions developing on other hosts. In addition, wild cot- and R. Dyer, Virginia Commonwealth University). tons vary in seasonal timing of ßower and fruit production (Ulloa et al. 2006), and boll weevils may be adapted to different phenologies of particular Sonoran References Cited hosts, attacking plants at different times. Such differ- Allen, C. T. 2008. Boll weevil eradication: an areawide pest ences have been reported for TW on wild cotton management effort, pp. 467Ð559. In O. Koul, G. Cuperus, versus boll weevils adapted to cultivated Þelds in Ar- and N. Elliot (eds.), Areawide pest management theory izona, accounting for the purported lack of economic and implementation. CAB International, Cambridge, MA. importance for the former (Fye 1968, Bottrell 1983). Bottrell, D. G. 1983. The ecological basis of boll-weevil (An- Although the particular ancestral cotton host from thonomus grandis Boheman) management. Agric. Eco- which all boll weevils originated remains unclear, our syst. Environ. 10: 247Ð274. Burke, H. R. 1968. Geographic variation and taxonomy of data may suggest (Fig. 4a), in accordance with pre- Anthonomus grandis Boheman. Technical report. Texas vious literature, that the most likely origin is in south Agricultural Experiment Station Department of Entomol- central Mexico. Because several cotton hosts are dis- ogy, Texas A&M, College Station, TX. tributed in the area (G. aridum, Gossypium barbadense Burke, H. R. 1976. Bionomics of anthonomine weevils. L., G. hirsutum, and G. gossypioides), there might have Annu. Rev. Entomol. 21: 282Ð303. November 2012 KUESTER ET AL.: BOLL WEEVIL GENETIC STRUCTURE AND DIVERSITY 915

Burke, H. R., W. E. Clark, J. R. Cate, and P. A. Fryxell. 1986. Kim, K. S., and T. W. Sappington. 2004a. Boll weevil (An- Origin and dispersal of the boll weevil. Bull. Entomol. Soc. thonomus grandis Boheman) (Coleoptera: Curculion- Am. 32: 228Ð238. idae) dispersal in the southern United States: evidence Burnham, K. P., and D. R. Anderson. 2002. Model selection from mitochondrial DNA variation. Environ. Entomol. 33: and multimodel inference: a practical information-theo- 457Ð470. retic approach, 2nd ed. Springer, New York. Kim, K. S., and T. W. Sappington. 2004b. Genetic structur- Carstens, B., J. D. Degenhardt, A. Stevenson, and J. Sullivan. ing of boll weevil populations in the US based on RAPD 2005. Accounting for coalescent stochasticity in testing markers. Insect Mol. Biol. 13: 293Ð303. phylogeographical hypotheses: modelling Pleistocene Kim, K. S., and T. W. Sappington. 2004c. Isolation and char- population structure in the Idaho giant salamander Di- acterization of polymorphic microsatellite loci in the boll camptodon aterrimus. Mol. Ecol. 14: 255Ð265. weevil, Anthonomus grandis Boheman (Coleoptera: Cur- Clement, M., D. Posada, and K. Crandall. 2000. TCS: a com- culionidae). Mol. Ecol. Notes 4: 701Ð703. puter program to estimate gene genealogies. Mol. Ecol. 9: Kim, K. S., and T. W. Sappington. 2006. Molecular genetic 1657Ð1660. variation of boll weevil populations in North America Cross, W. H. 1973. Biology, control, and eradication of the estimated with microsatellites: implications for patterns boll weevil. Annu. Rev. Entomol. 18: 17Ð46. of dispersal. Genetica 127: 143Ð161. Dopman, E. B., L. Perez, S. M. Bogdanowicz, and R. G. Kim, K. S., P. Cano-Rıós, and T. W. Sappington. 2006. Using Harrison. 2005. Consequences of reproductive barriers genetic markers and population assignment techniques to for genealogical discordance in the European corn borer. infer origin of boll weevils (Coleoptera: Curculionidae) Proc. Natl. Acad. Sci. U.S.A. 102: 14706Ð14711. unexpectedly captured near an eradication zone in Mex- Escobar-Ohmstede, A. 2004. Desastres Agrõćolas en Me´xico: ico. Environ. Entomol. 35: 813Ð826. Cata´logo Historico, Toma II, Siglo XiX. Fondos de Cultura Kim, K. S., C. T. Allen, and T. W. Sappington. 2008. Genetic Econońmica, Me´xico, D.F. proÞling to determine potential origins of boll weevils Excoffier, L., G. Laval, and S. Schneider. 2005. Arlequin ver. (Coleoptera: Curculionidae) captured in a Texas eradi- 3.0: an integrated software package for population genet- cation zone: endemicity, immigration, or sabotage? J. ics data analysis. Evol. Bioinform. 1: 47Ð50. Econ. Entomol. 101: 1729Ð1736. Fryxell, P. A., and M. J. Lukefahr. 1967. Hampea schlecht: Kim, K. S., G. D. Jones, J. K. Westbrook, and T. W. Sapping- possible primary host of the cotton boll weevil. Science ton. 2010. Multidisciplinary Þngerprints: forensic recon- 155: 1568Ð1569. struction of an insect reinvasion. J. R. Soc. Interface 7: Fye, R. 1968. The thurberia weevil in Arizona J. Econ. En- 677Ð686. tomol. 61: 1264Ð1268. Librado, P., and J. Rozas. 2009. DnaSP v5: a software for Goslee, S., and D. Urban. 2007. The Ecodist package for comprehensive analysis of DNA polymorphism data. dissimilarity-based analysis of ecological data. J. Stat. Bioinformatics 25: 1451Ð1452. Softw. 22: 1Ð19. Manly, B. 1986. Randomization and regression methods for Hall, T. A. 1999. BioEdit: a user-friendly biological se- testing for associations with geographical, environmental quence alignment editor and analysis program for Win- and biological distances between populations. Res. Popul. dows 95/98/NT. Nucleic Acids Symp. Ser. 41: 95Ð98. Ecol. 28: 201Ð218. Halverson, K., S. B. Heard, J. D. Nason, and J. O. Stireman. McKibben, G., E. Mitchell, W. Scott, and P. Hedin. 1977. 2008. Origins, distribution, and local co-occurrence of Boll weevils are attracted to volatile oils from cotton polyploid cytotypes in Solidago altissima (Asteraceae). plants. Environ. Entomol. 6: 804Ð806. Am. J. Bot. 95: 50Ð58. Miller, M., W. Pfeiffer, and T. Schwartz. 2010. Creating the Hedin, P., and J. McCarty. 1995. Boll weevil Anthonomus CIPRES Science Gateway for inference of large phylo- grandis Boh. oviposition is decreased in cotton Gossypium genetic trees, pp. 1Ð8. In Proceedings, Gateway Comput- hirsutum L. lines lower in anther monosaccharides and ing Environments Workshop, 14 November 2010, New gossypol. J. Agric. Food Chem. 43: 2735Ð2739. Orleans, LA. Institute of Electrical and Electronics En- Hey, J., and R. Nielsen. 2007. Integration within the Felsen- gineers, Piscataway, NJ. stein equation for improved Markov chain Monte Carlo Papadopoulou, A., I. Anastasiouu, and A. Vogler. 2010. Re- methods in population genetics. Proc. Natl. Acad. Sci. visiting the insect mitochondrial molecular clock: the U.S.A. 104: 2785Ð2790. mid-Aegean trench calibration. Mol. Biol. Evol. 27: 1659Ð Hillis, D., and M. Dixon. 1991. Ribosomal DNA: molecular 1672. evolution and phylogenetic inference. Q. Rev. Biol. 66: Peakall, R., and P. E. Smouse. 2006. GENALEX 6: genetic 411Ð453. analysis in Excel. population genetic software for teach- Huelsenbeck, J. P., F. Ronquist, R. Nielsen, and J. P. Bollback. ing and research. Mol. Ecol. Notes 6: 288Ð295. 2001. Bayesian inference of phylogeny and its impact on Pfenninger, M., and D. Posada. 2002. Phylogeographic history evolutionary biology. Science 294: 2310Ð2314. of the land snail Candidula unifasciata (Poiret 1801) (He- Hughes, J., and A. Vogler. 2004. The phylogeny of acorn licellinae, Stylommatophora): fragmentation, corridor mi- weevils (genus Curculio) from mitochondrial and nuclear gration and secondary contact. Evolution 56: 1776Ð1788. DNA sequences: the problem of incomplete data. Mol. Posada, D. 2011. Collapse: describing haplotypes from se- Phylogenet. Evol. 32: 601Ð615. quence alignments v. 1.2 computer program. (http:// Jones, R. W. 2001. Evolution of the host plant associations of www.softpedia.com/get/Science-CAD/Collapse.shtml). the Anthonomus grandis species group (Coleoptera: Cur- R Core Team. 2010. R: a language and environment for culionidae): phylogenetic tests of various hypotheses. statistical computing. R Foundation for Statistical Com- Ann. Entomol. Soc. Am. 94: 51Ð58. puting, Vienna, Austria. (http://www.r-project.org.). Jones, R. W. 2006. March of the weevils. Nat. Hist. 115: Rodriguez-Vallejo, J. 1976. Ixcatl, el algodoń mexicano. 30Ð35. Fondo de Cultura Econo´mica, Me´xico, D.F. Kass, R., and A. Raftery. 1995. Bayes factors. J. Am. Stat. Roehrdanz, R. L. 2001. Genetic differentiation of southeast- Assoc. 90: 773Ð795. ern boll weevil and thurberia weevil populations of An- 916 ANNALS OF THE ENTOMOLOGICAL SOCIETY OF AMERICA Vol. 105, no. 6

thonomus grandis (Coleoptera: Curculionidae) using mi- Terranova, A. C., R. G. Jones, and A. C. Bartlett. 1990. The tochondrial DNA. Ann. Entomol. Soc. Am. 94: 928Ð935. southeastern boll weevil: an allozyme characterization of Roehrdanz, R., L. Heilmann, P. Senechal, S. Sears, and P. its population structure. Southwest. Entomol. 15: 481Ð Evenson. 2010. Histone and ribosomal RNA repetitive 495. gene clusters of the boll weevil are linked in a tandem Thompson, J. D., D. G. Higgins, and T. J. Gibson. 1994. array. Insect Mol. Biol. 19: 463Ð471. CLUSTAL W. Improving the sensitivity of progressive Rousset, F. 1997. Genetic differentiation and estimation of multiple sequence alignment through sequence weight- gene ßow from F-statistics under isolation by distance. ing, position- speciÞc gap penalties and weight matrix Genetics 145: 1219Ð1228. choice. Nucleic Acids Res. 11: 4673Ð4680. Scataglini, M. A., V. A. Confalonieri, and A. A. Lanteri. 2000. Ulloa, M., J. Stewart, E. Garcia, S. Godoy, A. Gaytan, and S. Dispersal of the cotton boll weevil (Coleoptera: Curcu- Acosta. 2006. Cotton genetic resources in the western lionidae) in South America: evidence of RAPD analysis. states of Mexico: in situ conservation status and germ- Genetica 108: 127Ð136. plasm collection for ex situ preservation. Genet. Resour. Scataglini, M. A., A. A. Lanteri, and V. A. Confalonieri. 2006. Crop Evol. 53: 653Ð668. Diversity of boll weevil populations in South America: a Warner, R. E. 1966. Taxonomy of the subspecies of Antho- phylogeographic approach. Genetica 126: 353Ð368. nomus grandis (Coleoptera: Curculionidae). Ann. Ento- Simon, C., T. Buckley, F. Frati, J. Stewart, and A. Becken- mol. Soc. Am. 59: 1073Ð1088. bach. 2006. Incorporating molecular evolution into phy- Warner, R. E., and C. Smith, Jr. 1968. Boll weevil found in logenetic analysis and a new compilation of conserved pre-Columbian cotton from Mexico. Science 162: 911Ð polymerase chain reaction primers for animal mitochon- 912. drial DNA. Annu. Rev. Ecol. Evol. Syst. 37: 545Ð579. White, T., T. Burns, S. Lee, and J. Taylor. 1990. AmpliÞca- Stadler, T., and M. Buteler. 2007. Migration and dispersal of tion and direct sequencing of fungal ribosomal RNA Anthonomus grandis (Coleoptera: Curculionidae) in genes for phylogenetics, pp. 315Ð322. In M. Innis, H. South America. Rev. Soc. Entomol. Argent. 66: 205Ð217. Gelfand, J. Sninsky, and T. White (eds.), PCR protocols: Tamura, K., J. Dudley, M. Nei, and S. Kumar. 2007. MEGA4: a guide to methods and applications. Academic Press, molecular evolutionary genetics analysis (MEGA) soft- New York, NY. ware version 4.0. Mol. Biol. Evol. 24: 1596Ð1599. Wild, A., and D. Maddison. 2008. Evaluating nuclear pro- Tavare, S. 1986. Some probabilistic and statistical prob- tein-coding genes for phylogenetic utility in beetles. Mol. lems in the analysis of DNA sequences, pp. 57Ð86. In Phylogenet. Evol. 48: 877Ð891. R. M. Miura (ed.), Lectures on mathematics in the life sciences, vol. 17. American Mathematical Society, Providence, RI. Received 12 June 2012; accepted 2 September 2012.