Molecular Phylogenetics and Evolution 41 (2006) 472–495 www.elsevier.com/locate/ympev

Species diversiWcation patterns in the Polynesian jumping spider genus Havaika Prószyjski, 2001 (Araneae, Salticidae)

Miquel A. Arnedo ¤, Rosemary G. Gillespie 1

Division of Insect Biology, University of California-Berkeley, ESPM 201 Wellman Hall, Berkeley, CA 94720-3112, USA

Received 30 March 2006; revised 10 May 2006; accepted 13 May 2006 Available online 20 May 2006

Abstract

Hotspot archipelagoes provide exceptional models for the study of the evolutionary process, due to the eVects of isolation and topograph- ical diversity in inducing the formation of unique biotic assemblages. In this paper, we examine the evolutionary patterns exhibited by the jumping spider genus Havaika Prószyjski, 2001 in the Polynesian islands of the Hawaiian and Marquesas chains. To date, systematic research on Havaika has been seriously limited by the poor taxonomic knowledge on the group, which was based on a handful of specimens that showed continuous variability and lacked clear-cut diagnostic characters. Here, we circumvent this problem by inferring a phylogeny based on DNA sequences of several fragments including both mitochondrial (protein coding cytochrome oxidase I, NAD1 dehydrogenase, ribosomal 16S, and tRNA leu) and nuclear (internal transcribed spacer 2) genes, and a statistical morphological analyses of a large sample of specimens. Results suggest that the Marquesan and Hawaiian Havaika may be the result of independent colonizations. Furthermore, data provide little support for the standard “progression rule” (evolution in the direction of older to younger islands) in Hawaiian Islands. This may be explained by a recent arrival of the group: age estimates of the diVerent lineages suggest that Havaika colonized the Hawaiian Islands after most of the extant islands were already formed. The lack of clear-cut diagnostic characters among species may also be explained by the recent origin of the group since molecular data do not provide any evidence of hybridization among lineages. Quantitative morphological data coupled with the phylogenetic information allow us to reevaluate the current limitation of Havaika taxonomy. Molecular data support the existence of at least four diVerent evolutionary lineages that are further morphologically diagnosable. However, genealogical relationships are better predicted by geographical aYnity (i.e. island) than by morphological characters used in the original descriptions of the species. A pattern of size segregation linked to largely overlapping distributions of some of the species hints at a potential involvement of competition in generating morphological diversity. This study contributes to our understanding on the origin and shaping of the biodiversity of oceanic islands and sets the stage for more detailed studies on particular aspects of these previously overlooked spiders. © 2006 Elsevier Inc. All rights reserved.

Keywords: Salticidae; Hawaiian islands; Marquesas; Competition; Island evolution; Phylogeny

1. Introduction pie, 1998; Simon, 1987), and development of communities (Gillespie, 2004) and ecosystems (Vitousek et al., 1998). In The Hawaiian Islands are often considered a natural the context of evolutionary biology, studies have shown the laboratory for evolution, allowing studies of patterns of importance of isolation that the islands provide in allowing diversiWcation and species formation (Roderick and Gilles- ecological exploration and adaptive radiation (Carson and Sato, 1969; Gillespie, 2005; Vandergast et al., 2004). It has recently been shown that the proportion of species * Corresponding author. Present address: Departament de Biologia Ani- endemic to an oceanic island is related linearly to its species mal, Universitat de Barcelona, Av. Diagonal 645, 08028 Barcelona, Spain. richness, from which it was inferred that species diversity Fax: +34 93 403 5740. may drive diversiWcation (Emerson and Kolm, 2005). This E-mail addresses: [email protected] (M.A. Arnedo), gillespi@ nature.berkeley.edu (R.G. Gillespie). relationship suggests that competition not only plays a key 1 Fax: +1 510 642 7428. role in structuring biological communities (Gillespie, 2004),

1055-7903/$ - see front matter © 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.ympev.2006.05.012 M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495 473 but also may act as a selective agent triggering species (Prószyjski, 2002). Diagnostic characters are restricted to the diVerentiation. Indeed, character displacement, an outcome bulb of the male palp (Prószyjski, 2003) and the pattern, dis- of interspeciWc competition due to limiting resources, tribution, and color of setae around the chelicerae and eyes appears to be more widespread in nature than previously (Simon, 1900). Although these characters cannot be used for thought (Losos, 2000), and its key role has been implicated species identiWcation, they do serve to separate Hawaiian in the evolution of diversity in many adaptive radiations species into three phenetic groups: species with reddish and (Schluter, 2000a). white setae covering the proximal anterior part of the chelic- Several spider groups with Hawaiian endemic lineages erae and with long male palpal tibia [H. albociliata (Simon, have been used as models for the study of diVerent aspects 1900), H. canosa (Simon, 1900), H. pubens (Simon, 1900), and of the evolutionary process. Ecological shifts have played a H. valida (Simon, 1900)] and species with chelicerae covered major role in the diversiWcation of the spider genus Tetrag- with long white bristles forming lines, and with a short male natha (Blackledge and Gillespie, 2004; Gillespie, 2004), palpal tibia. The latter group is further divided into species while speciation in the endemic spider Orsonwelles has been with a long male bulb embolus [H. cruciata (Simon, 1900), H. mostly driven by inter-island colonization (Hormiga et al., jamiesoni Prószyjski, 2001, H. navata (Simon, 1900)] and 2003). In all spider groups examined to date, species on species with a short embolous [(H. senicula (Simon, 1900) younger islands appear to have been derived from ances- and H. verecunda (Simon, 1900)]. The absence of clear-cut tors on older islands, which is in accordance with prevailing limits separating species in Hawaiian Havaika could be “progression rule” found in the large majority of lineages explained by a recent diversiWcation of the group with insuY- within the Hawaiian Islands (Funk and Wagner, 1995). cient time for Wxation of diagnostic characters. Alternatively, The current study uses a unique system of jumping spi- species limits may have been secondarily obscured as a result ders: the genus Havaika Prószyjski, 2001 (Araneae: Saltici- of recurrent hybridization events. Indeed, occurrence of nat- dae) in the Hawaiian Islands. Havaika is one of the most ural hybridization has been documented in several Hawaiian species-rich salticid genera in the PaciWc region, only sur- arthropod taxa, including Drosophila (Carson, 1989), Lau- passed by Sobasina Simon, 1898 from the western PaciWc pala crickets (Shaw, 1996), and Megalagrion damselXies (Jor- (Berry et al., 1998). It currently comprises three species in dan et al., 2003). the Marquesas (Berland, 1933, 1934) and nine species in the An additional biogeographic puzzle presented by the Hawaiian Islands (Prószyjski, 2002; Simon, 1900) although genus Havaika is the inclusion of the Marquesas islands, in given the status of knowledge of other spiders groups (e.g. addition to Hawaii, in its distribution (Berland, 1933, 1934). Tetragnatha) prior to recent research, the possibility of Biological similarities across PaciWc islands, in particular many more undescribed species is evident (Gillespie, 1999). among snails (Pilsbry, 1900), certain insects (Meyrick, Recent research using molecular characters for the entire 1935a,b), spiders (Berland, 1942), and plants (Brown, 1921; family Salticidae, which included 81 genera broadly scat- Campbell, 1933; Guillaumin, 1928) led scientists in the ear- tered throughout the more than 500 known genera accord- lier part of the 20th century to propose the existence of an ing to previous notions of the phylogenetic diversity of the extensive land mass (submerged around the early Tertiary) family, has identiWed the continental genera Pellenes in the area currently occupied by the PaciWc depression Simon, 1876 and Habronattus F. O. P.-Cambridge, 1901 as (Gregory, 1930). However, considerable evidence now the closest relatives of Havaika (Maddison and Hedin, exists to support the hypothesis that the islands within the 2003b). These results are further supported by somatic and PaciWc depression are of volcanic origin and acquired their genitalic morphological similarities (Prószyjski, 2002). The faunas by overseas dispersal (Gregory, 1928): The Hawai- genus Habronattus (Masta and Maddison, 2002) is well ian archipelago originated from a volcanic hotspot (Wil- known for elaborate male secondary sexual characteristics, son, 1963), and the islands of the Marquesas were formed in with sexual selection having been demonstrated to play a a similar fashion, though from diVerent hotspots (Nunn, prominent role in species diversiWcation (Masta and Madd- 1994). Nevertheless, molecular evidence has recently indi- ison, 2002). However, Havaika males, very much like Pel- cated biogeographic connections between the remote lenes males, have only mild ornamentation that shows PaciWc islands. The Hawaiian plant genus Bidens is sister to moderate diversity. a radiation from the Marquesas Islands, and together this Hawaiian Havaika species are remarkably variable in clade is derived from continental America (Ganders et al., size, ranging from 2 to 10 mm in overall length. Although 2000). Similarly, Ilex anomala occurs in Hawaii and Tahiti about half of the species of Havaika are single-island and is related to continental American species. Crab spiders endemics, islands are generally inhabited by more than one (Thomisidae) appear to form a monophyletic Polynesian species that largely overlap in distribution. These observa- group (Hawaii + Marquesas + Society Islands) that would tions allow us to hypothesize a possible role of intraspeciWc also include American representatives (Garb and Gillespie, competition in species diversiWcation. 2006). However, this pattern of phylogenetic aYnity across Despite the recent taxonomic treatment by Prószyjski the far-Xung islands of Polynesia is not universal. For (2002), Havaika remains poorly deWned, with the author instance, endemic lineages of the spider genus Tetragnatha describing the group as a “cluster of similar looking species, within the diVerent archipelagoes appear to have arisen diVering by inconspicuous and intergrading characters” from independent sources (Gillespie, 2002). 474 M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495

Unfortunately, taxonomic understanding of the Mar- included in the analyses (see Table 1). The genus Thiodina, a quesan species is even worse than that of Hawaiian repre- member of the amycoids, was used to root the trees, accord- sentatives, with poor descriptions based on few specimens ing to the results of Maddison and Hedin (2003a). (in two species only females are known). One of the endemic species originally included in the group (H. rufes- 2.1.3. Molecular phylogenetic analyses cens Berland, 1934 from Nuku Hiva) has recently been Total DNA was extracted from one or two legs of transferred to the genus Habronattus (Prószyjski, 2002). freshly collected specimens Wxed in 95% ethanol. Extrac- However, this species does not share any of the synapomor- tion, ampliWcation, and sequencing followed the protocols phies currently assigned to Habronattus, namely a unique described in Arnedo et al. (2004). Partial fragments of the elbow on the tegular apophysis (TA) of the male palpus mitochondrial genes Cytochrome c oxidase 1 (CO1), the (according to drawings, H. rufescens does not even have a large ribosomal subunit (16S), the complete tRNA leucine well-developed TA) and the male third legs longer than the (tRANleu), and the NADH dehydrogenase subunit 1 Wrst (Maddison and Hedin, 2003a). (ND1), as well as the complete nuclear intron ITS-2 (Inter- In this paper, we use the information provided by a nal transcribed Spacer 2) were ampliWed and sequenced molecular phylogeny inferred from mitochondrial and using the following primer pairs: C1-J-1718 and C1-N-2191 nuclear markers, and morphological analyses of a large (Simon et al., 1994) (CO1, 472 bp), LR-N-13398 (Simon sample of specimens, to address some of the evolutionary et al., 1994), and N1-J-12261 (Hedin, 1997) or N1-J-12334 questions posed by the spider genus Havaika. SpeciWcally, (CCIATTAIAAGAATTGAATATGCTG, this study) we examine the relationship between Hawaiian and Mar- (16S-tRNAleu-ND1, »980 bp); in some cases, the former quesan representatives, the relationship between morpho- fragment had to be ampliWed and sequence in two frag- logical variability and species boundaries (and hence ments with the primer pairs LR-N-13398 and LR-J-12864 determine the basis for the diYculty in diagnosing species), (Hsiao, pers. comm.) (16S, »430 bp), and N1-N-12945 and and the potential role played by interspeciWc competition in N1-J-12261 or N1-J-12334 (tRNAleu-ND1, »550 bp); and the diversiWcation of Hawaiian Havaika. ITS-5.8S and ITS-28S (White et al., 1990) (ITS-2, »400 bp). ITS-2 sequences were only obtained for the Marquesan and 2. Materials and methods Hawaiian taxa. Sequences including fragments of the 16S, tRNAleu, and 2.1. Taxonomic sampling ITS-2 showed diVerences in length suggesting insertion or deletion events (indels) during the evolutionary history of 2.1.1. Ingroup these sequences. The assignment of positional homology (i.e. A total of 250 specimens of Hawaiian Havaika were ana- alignment) in these situations is not trivial. The most widely lyzed for morphology, and for 29 of these we also obtained used strategy to accommodate indel events in phylogenetic DNA sequences (Table 1, localities shown in Fig. 1). Speci- inference involves the alignment of sequences of diVerent mens were obtained through Weld collections by the authors length by adding gap characters to keep positional similarity. or loaned from public institutions (American Museum of The aligned matrix is then subject to phylogenetic analyses Natural History, Bishop Museum, Muséum national d’His- using any of the available inference methods. Alternatively, toire naturelle, and the Natural History Museum in London) the assignment of positional homology can be viewed as part and private collections (J. Berry). Type material of Hawaiian of the phylogenetic inference problem (Mindell, 1991; but see Havaika was loaned from the Bishop Museum (Honolulu), Simmons, 2004; Simmons and Ochoterena, 2000 for critical the Muséum national d’Histoire naturelle (Paris), and the reappraisal of this approach). The direct optimization British Natural History Museum (London) to help in identi- method (Wheeler, 1996) overcomes the alignment construc- Wcation. Specimens for sequencing were selected on the basis tion step by incorporating indel events as one of the possible of their variation in morphological characters used in classi- transformations during the tree evaluation process. This cal Havaika taxonomy (Prószyjski, 2002; Simon, 1900), their method has been considered to be superior because it oVers a sex, and island location. Marquesan representatives included more consistent treatment of indels and is less dependent on Habronattus rufescens and an undescribed species from Hiva initial conditions (Brocchieri, 2001; Thorne et al., 1992; Ving- Oa with a tegular apophysis on the male bulb, hence inferred ron and Von Haeseler, 1997; Wheeler, 1996). However, most to belong to the genus Habronattus (referred to as Habronat- of the tests and inference methods of common use in system- tus?). In addition to allowing assessment of phylogenetic atics require a Wxed alignment or have not yet been imple- aYnities, Marquesan species provided an external calibration mented for a dynamic optimization framework. For this point to estimate clade ages in Hawaiian Havaika. reason, the present study used both static alignment based analyses and direct optimization analyses. 2.1.2. Outgroup Outgroup selection relied heavily on a recently published 2.1.4. Dynamic optimization study on the molecular phylogeny of jumping spiders Analyses under dynamic optimization were performed (Maddison and Hedin, 2003b). Representatives of clades at with the computer program POY (Wheeler et al., 1996– successive levels of exclusivity in relation to Havaika were 2003). Although current versions of the program include M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495 475

Table 1 Taxonomic and locality information of the specimens included in the molecular analyses Clade Genus Species/Morph Locality Co1 16s-ND1 ITS Amycoids Thiodina sp. USA: AZ: Tucson AF327930 AF327958/AF328017 — Ballinae Attidops youngi USA: MO: Valley View Galde AF327990 AF327961/AF327990 — Dendryphantinae Messua limbata USA: AZ: Pinaleno Mts. AF328045 AF327986/AF328045 — Euophryinae Zenodorus microphtalmus HI: Oahu: Wai’anae DQ531816 DQ532097 — Unassigned Heratemita alboplagiata Philippines: Luzon AF327991 AF327962/AF328021 — Synagelinae Peckhamia sp. USA: AZ: Atascosa Peak AF327995 AF327966/AF328025 — Marpissoids Itata sp. Ecuador: Machalilla, NP AF327989 AF327960/AF328019 — Misc. Wx. emb. Hasarius adansoni HI: Oahu: Wai’anae DQ531785 DQ532067 — Marpissoids Platycryptus undatus USA: FL: Newnan’s Lake AF327992 AF327963/AF328022 — Marpissines Marpissa pikei USA: AZ: W of Nogales AF327993 AF327964/AF328023 — Heliophaninae Phintella versicolor HI: Oahu: Wai’anae DQ531815 DQ532096 — Pelleninae Sibianor aemulus Canada: Ontario — AY296675/AY297318 — Pelleninae Pellenes shoshonensis USA: CA: White Mts. AY297383 AF477252 — Pelleninae Pellenes cf. apacheus USA: AZ: Huachuca Mts. — AF477250 — Pelleninae Pellenes cf. longimanus USA: TX: Rio Grande V. — AF477251 — Pelleninae Habronattus sp. MACHAL Ecuador: Manabi, Salaite AY297380 AF477276 — Pelleninae Habronattus mexicanus USA: TX: Pecos River AY297381 AF477353 — Pelleninae Habronattus rufescens Marquesas: Nuku Hiva DQ531803 DQ532084 DQ531817 Pelleninae Habronattus? sp. Marquesas: Hiva Oa DQ531801 DQ532082 DQ531818 Pelleninae Havaika sp. “pubens” HI: Kauai, Kumuwela DQ531795 DQ532077 — Pelleninae Havaika sp. “pubens” HI: Kauai, Kumuwela DQ531796 DQ532078 DQ531820 Pelleninae Havaika sp. “pubens” HI: Oahu: Ko’olau, Aiea DQ531804 DQ532085 DQ531835 Pelleninae Havaika sp. “pubens” HI: Oahu: Ko’olau, Waimano DQ531806 DQ532087 DQ531841 Pelleninae Havaika sp. “pubens” HI: Oahu: Wai’anae, Palikea DQ531809 DQ532090 DQ531837 Pelleninae Havaika sp. “pubens” HI: Oahu: Wai’anae, Palikea DQ531808 DQ532089 (16S) DQ531836 Pelleninae Havaika sp. “pubens” HI: West Maui, Pu’u Kukui DQ531812 DQ532093 DQ531831 Pelleninae Havaika sp. “pubens” HI: West Maui, Pu’u Kukui DQ531813 DQ532094 DQ531830 Pelleninae Havaika sp. “pubens” HI: Maui: Haleakala, Waikamoi DQ531786 DQ532068 DQ531843 Pelleninae Havaika sp. “pubens” HI: Maui: Haleakala, Hanawai DQ531787 DQ532069 (16S) DQ531819 Pelleninae Havaika sp. “pubens” HI: Big Island, Kahaualea DQ531791 DQ532073 DQ531832 Pelleninae Havaika sp. “pubens” HI: Big Island, Waiakea DQ531793 DQ532075 DQ531834 Pelleninae Havaika sp. “pubens” HI: Big Island, Laupahoehoe DQ531794 DQ532076 DQ531833 Pelleninae Havaika morphotype D HI: Maui: Haleakala, Hana DQ531789 DQ532071 DQ531840 Pelleninae Havaika cruciata HI: Big Island, Kipukas Saddle Rd. DQ531790 DQ532072 DQ531838 Pelleninae Havaika cruciata HI: Big Island, Kipukas Saddle Rd. DQ531792 DQ532074 DQ531839 Pelleninae Havaika sp. “verecunda” HI: Kauai, Kauluaha’ula DQ531797 —— Pelleninae Havaika sp. “verecunda” HI: Kauai, La’au Ridge DQ531798 DQ532079 DQ531828 Pelleninae Havaika sp. “verecunda” HI: Kauai, Mt. Kahili DQ531799 DQ532080 DQ531827 Pelleninae Havaika sp. “verecunda” HI: Kauai, Makalehas Mts. DQ531800 DQ532081 DQ531826 Pelleninae Havaika sp. “verecunda” HI: Oahu: Ko’olau, Aiea DQ531805 DQ532086 DQ531825 Pelleninae Havaika sp. “verecunda” HI: Oahu: Ko’olau, Waimano DQ531807 DQ532088 DQ531823 Pelleninae Havaika sp. “verecunda” HI: Oahu: Wai’anae, Mt. Ka’ala DQ531810 DQ532091 DQ531842 Pelleninae Havaika sp. “verecunda” HI: Oahu: Wai’anae, Mt. Ka’ala DQ531811 DQ532092 DQ531824 Pelleninae Havaika sp. “verecunda” HI: Molokai, Kamakou DQ531802 DQ532083 DQ531821 Pelleninae Havaika sp. “verecunda” HI: West Maui, Pu’u Kukui DQ531814 DQ532095 DQ531829 Pelleninae Havaika sp. “verecunda” HI: Maui: Haleakala, Auwahi DQ531788 DQ532070 DQ531822 GenBank accession number for each gene fragment sequenced. New sequences generated during the present study in bold. settings for maximum likelihood analyses, both the novelty compromising results (Giribet, 2001). For these reasons, a of the method (results of this option have not yet been fully preliminary static alignment of the 16S–tRNAleu partition explored) and the elevated computational time led us to was built with the aid of the automatic alignment program limit the analyses to parsimony. ClustalX (Thompson et al., 1997) using default options. The The 16S and tRNAleu mitochondrial gene fragments fragment was then spliced into eight regions of about 100-bp were considered a single partition that was analyzed individ- Xanked by a series of 10 identical nucleotides to guarantee ually and in combination with CO1, ND1, and ITS-2. In all the homology of these regions across all the taxa. ITS-2 was the analyses, the protein coding genes were considered as pre- analyzed as a single fragment because internal homologous aligned, which avoids the use of indel transformations during regions were diYcult to identify. tree length calculations of these particular partitions. Addi- Sensitivity of the results to particular assumptions of the tional computation time can be saved by dividing gene frag- analyses were investigated using combinations of diVerent ments into small putative homologous pieces, without gap opening, gap extension, and transversion (tv)/transition 476 M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495

La'au Makalehas Mt. Ka'ala Mt. Kahili KAUAI Waimano 4.70 My Kauluaha´ula Kamakou 2.60 My Kumuwela 3.00 My Aiea Pu'u Kukui 1.32 My OAHU Waikamoi Palikea MOLOKAI Hanawai C 2.20 My LANAI Hana 1.28 My Auwahi 1.2 My MAUI Saddle Rd. Laupahoehoe Hatutu 4.75 My B Eiao 5.40 My Waiakea

Nuku Hiva 4.00 My Ua Huka Mt. Tekoa 2.65 My Ua Pou 3.65 My

2.01 My Hiva Oa

2.35 My Tahuata Fatu Hiva Kahaualea A 1.55 My 0 20 km. HAWAII 0.50 My

Fig. 1. (A) Map showing the location of the Marquesas and the Hawaiian Islands in the PaciWc region. (B) Map of the Marquesas with geological ages after Craig et al. (2001), and collection localities of specimens included in the molecular analyses. (C) Map of the Hawaiian Islands with volcano geological ages after Carson and Clague (1995) and Price and Clague (2002), and collection localities of specimens included in the molecular analyses.

(ts) costs (Wheeler, 1995). Maximum congruence among connected in parallel with PVM software and the parallel data partitions as measured by the ILD (Mickevich and version of POY (commands–parallel–controllers 3 in Farris, 1981) was used to select across the diVerent parame- eVect); or (2) the Centre de Supercomputació de Catalu- ter cost combinations assayed (Wheeler and Hayashi, nya (CESCA, http://www.cesca.es), run sequentially on 1998). The actual parameter combinations investigated are either an 8-node Compac HPC320 at 833 MHz or an 8- listed in Table 2. node Compaq Beowulf 600 MHz. The heuristic search Analyses were run at either (1) the Bioinformatics strategy involved “quick” building of 10 trees by random Centre of the University of Copenhagen (http://www.binf. addition of taxa (-buildsperreplicate 10 -buildspr -build- ku.dk), using a cluster of 236 P4 processors at 2.4 GHz tbr -approxbuild -buildmaxtrees 2), followed by spr and

Table 2 Summary of the analyses performed using direct optimization with diVerent parameter combinations Gap opening Gap extension Tv/Ts Combined mt 16S+tRNAleu CO1 ND1 ILD ITS-2 Combined g Combined mt ILD Length #Trees Length Length Length Length #Trees Length #Trees Length #Trees 1 1 1 2668 8 899 830 888 0.0191 182 14 963 3 760 4 0.0218 2 1 1 2731 16 966 830 888 0.0172 213 1 1003 3 768 2 0.0219 2 2 1 2742 16 974 830 888 0.0182 243 6 1052 6 769 2 0.0380 2 1 2 4013 18 1440 1205 1282 0.0214 262 6 1322 1 1037 1 0.0174 2 2 2 4171 2 1473 1205 1282 0.0506 301 50 1376 2 1038 1 0.0269 4 1 1 2811 16 1041 830 888 0.0185 261 8 1071 8 784 2 0.0243 4 2 1 2824 16 1051 830 888 0.0195 295 7 1128 6 785 2 0.0426 4 4 1 2839 8 1061 830 888 0.0211 357 50 1224 6 787 2 0.0654 4 1 2 4152 8 1587 1205 1282 0.0188 318 20 1398 1 1053 1 0.0193 4 2 2 4159 4 1591 1205 1282 0.0195 357 3 1454 1 1054 1 0.0296 4 4 2 4363 2 1603 1205 1282 0.0626 456 4 1548 1 1056 1 0.0233 4 1 4 6546 2 2409 1935 2057 0.0222 411 27 2028 1 1581 1 0.0178 4 2 4 6674 2 2538 1935 2057 0.0216 417 4 2087 1 1584 1 0.0412 4 4 4 6732 2 2591 1935 2057 0.0221 529 18 2190 1 1586 1 0.0342 ILD: diVerence between the length of the combined analyses and the sum of the partial analyses divided by the length of the combined analysis; Combined mt: all mitochondrial genes combined; Combined g: all genes combined. Lowest ILD values in bold. Gap opening, gap extension, and Tv/Ts (transversion transition ratio) refer to the actual costs used in each particular analyses. M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495 477 tbr branch swapping holding one cladogram per round WINCLADA v.1.00.08 (Nixon, 2002). Heuristic searches (-sprmaxtrees 1 -tbrmaxtrees 1). Two rounds of tree fusing involved 100–500 rounds of random addition of taxa, hold- (-treefuse -fuselimit 10 -fusemingroup 5) and tree drifting ing Wve trees per round and a total maximum of 1000. Best (-numdriftchanges 30 -driftspr -numdriftspr 10 -drifttbr - overall trees were further subjected to a new round of tbr numdrifttbr 10) were added to increase eYciency, holding branch swapping. Clade support was assessed using jack- up to Wve trees per round (-maxtrees 5) and using the com- kniWng, with 1000 jackknife pseudoreplicates. Individual mand -Wchtrees, which saves the most diverse cladograms searches consisted of 10 rounds of random addition of taxa, found for each island. Additionally, all cladograms found holding two trees per replicate and an overall maximum of within 0.5% of the minimum tree length (-slop 5 -check- 1000. slop 10) were examined to avoid Wnding of suboptimal The program Modeltest v.3.06 (Posada and Crandall, trees due to tree length miscalculations (POY uses some 1998) was used to select the model of evolution that shortcuts to speed-up tree evaluation). This strategy was explained the data signiWcantly better with fewer parame- repeated 100 times (-random 100) and a maximum of 50 ters using the AIC criterion (Buckley et al., 2002). Maxi- trees was retained (-holdmaxtrees 50). Clade support was mum likelihood analyses of the static alignments were assessed by means of Jackknife proportions using 100 performed with the computer program PAUP* v.4.0 (Swo- randomly resampled matrices, with a probability of char- Vord, 2001) using the model of evolution and parameter acter deletion of 1/e (default option). Individual search values suggested by Modeltest and a heuristic search with strategies involved taking the best tree from Wve rounds of 10 random stepwise additions of taxa and TBR branch random additions of taxa. swapping. Additional ML trees and bootstrap proportions (500 pseudoreplicates) were calculated with the aid of the 2.1.5. Static alignments computer program PHYML v.2.4 (Guindon and Gascuel, Alignments with Wxed homology statements, hereafter 2003). Bayesian inference was implemented with the com- referred to as static alignments, of the ribosomal genes and puter program Mr Bayes v.3.0 (Ronquist and Huelsenbeck, the ITS-2 were constructed following the method of Hedin 2003). Four simultaneous MCMCMC chains (one cold and and Maddison (2001), building multiple automatic align- three heated) were run for 1,500,000 iterations. Three inde- ments that correspond to diVerent combinations of gap pendent runs were performed to ensure convergence of the opening and gap extension costs. The resulting alignments results. Plots of the negative log likelihood against the num- range from relatively gappy alignments to more com- ber of generations were used to identify the point of sta- pressed ones. A particular gap opening/extension cost tionarity, and trees obtained from generations before this alignment is chosen based on topological congruence to the point were discarded as burn in. All analyses were run with elision matrix (Wheeler et al., 1995), which results from speciWc and unlinked GTR + I +  models for each gene appending all the alignments constructed for a given gene fragment, although all mitochondrial genes were combined fragment. The method was implemented by building multi- in a single data set. ple alignments with ClustalX (Thompson et al., 1997) under the following gap opening/extension cost values: 8/2, 8/4, 2.2. Data partition congruence and alternative hypothesis 20/2, 24/4, and 24/6. In all cases, transition weight was Wxed testing at 0.5. Topological congruence was measured by the parti- tion, triplet, and quartet symmetric diVerences as imple- Congruence among mitochondrial and nuclear parti- mented in the computer program Component 2.0 (Page, tions was assessed on the static alignments under the parsi- 1993). All the former analyses were run with gaps consid- mony criterion using the ILD test (Farris et al., 1994), as ered as absence/presence data as suggested by Simmons implemented in the program Winclada. Heuristic searches and Ochoterena (2000). The program GapCoder (Young on 1000 pseudoreplicates consisted of Wve random addition and Healy, 2002) facilitates the automatic recoding of gaps of taxa holding two trees per iteration and a maximum of using the simple indel coding version of Simmons and Och- 10 trees overall. In all cases, uninformative characters and oterena’s method. A mitochondrial combined static matrix incomplete taxa were removed from the matrices before was then constructed by adding the best alignment selected performing the test, as suggested in the literature (Arnedo for the ribosomal and tRNA genes plus the COI and ND1 et al., 2001; Cunningham, 1997). fragments. The same alignment selection protocol was fol- The ILD test has been shown to be excessively prone to lowed for the nuclear ITS-2. The combined static matrices type 1 error (i.e. it may indicate signiWcant incongruence were analyzed using three diVerent gap treatments (gaps as among partition when they are actually not) (Barker and missing data, gaps as 5th character-state and gaps as Lutzoni, 2002). For this reason, incongruence among parti- absence/presence characters) to assess their eVect on tree tions was investigated further using topology tests. The selection. Templeton (Templeton, 1983) and Winning-site (Prager Analyses of the static alignment matrices under parsi- and Wilson, 1988) tests were used for parsimony analyses mony were run with the computer programs NONA v.2 and the Shimodaira–Hasegawa (Shimodaira and Hase- (GoloboV, 1993) and TNT v.1 (GoloboV et al., 2003) and gawa, 1999) test for maximum-likelihood analyses. The the matrices and tree were manipulated with the program same tests were also applied to elucidate the diVerences 478 M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495 between alternative topological hypotheses. All tests were ConWdence intervals for each clade age were estimated run in PAUP 4.0. Finally, local incongruence was examined with PAUP* from 100 bootstrap replicates of the original by looking for well-supported but irreconcilable clades in data matrix and the preferred tree constrained. Mean and the trees derived from separate analyses of the partitions standard deviations of the bootstrapped branch lengths of (Wiens, 1998). This is the only method to assess incongru- each clade were calculated with r8s and used to construct ence among partitions currently available for analyses 95% conWdence intervals. under direct optimization. 2.4. Analysis of morphological variation 2.3. Estimate of lineage divergence time Characters traditionally used for diagnosing Hawaiian A likelihood ratio test rejected the presence of a strict Havaika species were scored for available specimens. These molecular clock in the molecular phylogeny of Havaika characters refer to the face (clypeus and frontal side of the species (p D 0.001, 2 D 56.89, 28 d.f.). Several methods chelicerae) ornaments and the structure of the male copu- have been proposed to estimate divergence times in the latory bulb. Three types of face patterns have been absence of a molecular clock (Huelsenbeck et al., 2000; reported in Hawaiian Havaika (Simon, 1900): (1) Yellow- Sanderson, 1997, 2002; Thorne and Kishino, 2002; Yoder ish, scale-like hairs on the frontal part of the basal segment and Yang, 2000). Here we use a semi-parametric of the chelicerae (Fig. 2A and F) in both male and female approach, the penalized likelihood method (Sanderson, specimens; (2) several rows of white bristles instead of 2002), which has been proven to be superior to both full scales on the chelicerae and a necked clypeus (Fig. 2D and and non-parametric approaches (Sanderson, 2002). Penal- G–J) in both males and females; and (3) female clypeus ized likelihood (PL) requires choice of a smoothing and basal-most anterior part of the chelicerae densely cov- parameter value (), which determines the variation in the ered with white bristles (Fig. 2B, C, and E). The male pedi- rates of evolution across branches. In his proposition of palp show clear diVerences in at least two characters the PL, Sanderson suggested cross-validation, a widely (Prószyjski, 2002): some bulbs have a long and slender used model-selection method based on comparing param- embolus emerging from the basal half of the tegulum eter estimates after sequential branch removal, as an (Fig. 2K, L, and N), while in others the embolus is short objective criterion for choosing ’s value or even for con- and stout and emerges from the distal half of the tegulum trasting other estimators. Time of divergence and cross- (Fig. 2M and O). The relative length of the palpal tibia, validation calculations were carried out with the Linux compared with the length of the cymbium, also show version of the computer program R8S v.1.5 (Sanderson, diVerences across individuals. In some cases, the tibia is 2003). At least one calibration point is required for esti- clearly shorter than the cymbium (Fig. 2L–N), while in mating absolute divergence times, regardless of the others it is equal in size or even longer (Fig. 2K and O). method of choice. Fossil data are rarely available when The diVerent kind of face ornaments were recorded for working at the species level, necessitating the use of bio- every specimen examined and in male individuals also the geographic information which must be treated with cau- ratio between the length of the palpal tibia and the cym- tion since association between cladogenesis and bium, the ratio between the distance from the base of the geological events is always diYcult to prove (Caccone tegulum to the base of the embolus and the total length of et al., 1997). However, volcanic “hotspot” archipelagoes, the tegulum. Finally, the length of the carapace, a surro- such as the Hawaiian Islands or Marquesas provide excel- gate for body length, was measured in all specimens to lent opportunities for calibrating rates of molecular account for the large variation in size reported in the change based on the geological age of the islands (Fig. 1). genus. All measurements were taken in millimeters using a Estimates of the date of an island’s formation can then be dissection scope either Leica MZ16A or Leica MZAPO used as a maximum age for taxa inhabiting the island. equipped with an ocular measuring graticule. Nevertheless, this assumption is based on certain premises, the most important ones being that (1) the 2.5. Statistical procedures branching pattern in the phylogeny parallels the timing of island formation and (2) that the divergence between sis- 2.5.1. Morphotype discrimination analysis ter taxa does not greatly predate the formation of the col- A canonical correspondence analysis (CCA) was per- onized younger island (Fleischer et al., 1998). Baldwin and formed to assess the discriminating power of morphologi- Sanderson (1998) have cautioned against the use of inter- cal characters based on measurements. Specimens were nal calibration points for dating evolutionary events in assigned to lineages as suggested by face ornament patterns hotspot archipelagoes because errors in the placement of and island (see Results). Analyses were run with the com- island age are greatly magniWed deeper within the tree. puter program CANOCO for Windows v.4.5 (Hill’s scaling Additional sources of error can arise from extinction of option on, Monte Carlo permutation test enabled using 999 lineages and inclusion of the divergence that was already permutations). Results were visualized by plotting the Wrst present in the common ancestral population (Emerson two discriminant axes and adding information on discrimi- et al., 2000). nating characteristics, the centroids of classes, and individ- M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495 479

Fig. 2. Morphological traits of Hawaiian Havaika. Female face ornamentation: (A) pubens lineage; (B) H. cruciata; (C) verecunda lineage; (D) morphotype D; (E) morphotype Necker and Nihoa. Male face ornamentation: (F) pubens lineage; (G) H. cruciata; (H) verecunda lineage; (I) morphotype D; (J) mor- photype Necker and Nihoa. Copulatory bulb: (K) pubens lineage; (L) H. cruciata: (M) verecunda lineage; (N) morphotype D; (O) morphotype Necker and Nihoa; (P) Habronattus? sp. Hiva Oa; (Q) Habornattus rufescens Nuku Hiva. Overall dorsal view, female: (R) H. cruciata; (S) morphotype D. ual’s canonical scores. Biplots and scatter plots were 0.6 my ago, Molokai split from Maui Nui, followed about constructed with the program CANOCO DRAW for Win- 0.4 my ago by Lanai (Price and Elliott-Fisk, 2004). dows v.4.0. 3. Results 2.5.2. Size diVerentiation Variation in size and its association with lineages, sex, 3.1. Phylogenetic analyses and/or island population was investigated by means of standard statistical techniques. Individuals from each 3.1.1. Dynamic optimization island were grouped into diVerent lineages based on Results of the multiple analyses performed under diVer- somatic characters discussed above coupled with the infor- ent parameter values are summarized in Table 2. For the mation derived form the molecular phylogeny. For each mitochondrial data set, the parameter combination with island, a two-way analysis of variance was performed with gap opening twice the value of the uniformly weighted gap factors (1) lineage and (2) sex, and dependent variable “car- extension and transition/transversion ratio maximized apace length.” When applicable, post hoc multiple compar- character congruence across partitions as shown by the isons were run using the Student–Newman–Keuls (SNK) ILD (Fig. 3). Pellenine monophyly is supported with low test (Zar, 1984). For some islands, representatives of both jackknife values. Hawaiian Havaika are shown as mono- sexes were not available for all lineages. In these cases, sig- phyletic with 100 jackknife support, and its inclusion into a niWcant diVerences between lineages were assessed using a t clade containing mainland Habronattus and Pellenes spe- test. Additionally, diVerences in size among island popula- cies as well as the Marquesan representatives is also well tions of each lineage were investigated by means of one- supported. Hawaiian Havaika are divided into several well- way ANOVA. Normality and homogeneity of variance of supported clades that correspond to species groups diag- the data were assessed by Kolmogorov–Smirnov and Bart- nosed by face ornaments and male bulb characters. The lett tests, respectively (Zar, 1984). Only data from Oahu did Wrst clade includes male and female specimens from Kauai, not pass the homogeneity test and data from pubens (see Oahu, Maui, and Hawaii, all of which bear yellowish scales below) within islands did not pass normality test. These on the chelicerae and males have long palpal tibia (cym- data were subject to rank transformation prior to the anal- bium/tibia length < 1.3). Island populations of this clade are ysis (Potvin and RoV, 1993). shown as monophyletic. We will refer to this monophyletic SigniWcant size diVerences among lineages from a partic- group as the pubens clade. In the second clade, which is sup- ular volcano were also investigated for Oahu and the Maui ported by very high jackknife values, males bear rows of Nui complex. The last name refers to the group of islands long white bristles on the chelicerae and female specimens comprising Molokai + Lanai + Maui, which constituted a have a hairy clypeus. All males in this clade have short pal- single land mass until about 1.2 million years ago. About pal tibia (cymbium/tibia length > 1.3). Island populations 480 M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495

Thiodina sp. Phintella versicolor Hasarius adansoni Heratemita alboplagiata Itata sp. Messua limbata 100 Attidops youngi Platycryptus undatus Peckhamia sp. Marpisaa pikei Zenodurus microphtalmus

Sibianor aemulus Marquesas Habronattus mexicanus 78 Habronattus sp. Habronattus? sp. Hiva Oa 100 Habronattus rufescens Pellenes shoshonensis 53 99 Pellenes cf. longimanus 100 Pellenes cf. apacheus Havaika OK23 99 Havaika OK8 100 Havaika OW111 100 Havaika OW158 90 Havaika EM130 PELLENINES Havaika EM128 99 80 Havaika WM159 pubens 100 Havaika WM88 lineage 54 Havaika K146 Hawaiian Islands 100 Havaika K112

Havaika H137 Havaika 100 Havaika H83 100 Havaika H109 Havaika EM90 Morph D 100 Havaika H110 Havaika H10 H. cruciata 100 Havaika EM81 Havaika MK82 98 100 KAUAI 97 Havaika WM89 Havaika OK24 Havaika OK9 OAHU 100 verecunda 97 89 Havaika OW28 100 Havaika OW29 lineage MAUI NUI Havaika K38 Havaika K85 100 HAWAII Havaika K87 91 Havaika K86

Fig. 3. Strict consensus tree of 16 trees of length 2731 steps resulting for the direct optimization analysis with parsimony of all mitochondrial gene frag- ment combined, using the following parameter values: gap opening 2, extension gap 1, transversion/transition ratio 1. Jackknife support values >50 are shown below branches. also are shown as monophyletic (Molokai and Maui con- of the parameter cost combinations implemented. Hawai- sidered a single unit) with high jackknife values with a sin- ian Havaika and each of the named clades are always gle exception: one haplotype from East Maui joins the monophyletic regardless of the parameter cost combina- haplotypes from the Big Island (i.e. Hawaii Island) in a tion. Conversely, the internal topologies of pubens and vere- well-supported clade. The single male included in the East cunda clades are very sensitive to changes in the cost of the Maui-Big Island clade shows a long, slender embolus grow- explored parameters. ing below the half part of the tegulum, allowing identiWca- The lack of additional partitions in the nuclear data tion of haplotypes from the Big Island as H. cruciata. The (ITS-2) precluded the use of congruence across data sets as Maui female shows some somatic diVerences from Big a criterion for choosing the best parameter combination. Island females, including a sparsely haired clypeus and an Fig. 4 shows the results for the same parameter combina- overall dark brownish color pattern; we will refer to this tion selected for the mitochondrial data. Monophyly of specimen as morphotype D. Finally, remaining Kauai, each Hawaiian Havaika named clades and the sister-group Oahu, and Maui Nui haplotypes are joined in a clade, relationship between (morphotype D + H. cruciata) and the which is characterized by males having a short, stout bulb verecunda clade is recovered but not supported by jackknife embolus that grows from above the half distal part of the values. In several cases, island sequence types did not form tegulum. We will refer to this clade as the verecunda clade. monophyletic groups, for instance, the sequence-type Relationships among island lineages in clades pubens and OW111 from Oahu always clustered with sequence types verecunda are mostly unresolved. from Hawaii, while the sequence-type OK9 from Oahu Pellenine monophyly and monophyly of the mainland clustered with Maui Nui sequence types (only some trees Habronattus and the Marquesan species is recovered in half under cost scheme 441 support Oahu monophyly). In some M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495 481

Habronattus sp. Habronattus? sp. Hiva Oa Marquesas Havaika OW111 Havaika H109 Havaika H83 61 Havaika H137 Havaika EM130 Havaika K146 Havaika OK23 pubens Havaika OK8

lineage Hawaiian Islands Havaika OW158 Havaika WM88 100 Havaika WM159 55 Havaika EM128 Havaika EM90 Morph D Havaika H10 71 H. cruciata 63 Havaika H110 Havaika OK9 KAUAI Havaika MK82 Havaika EM81 51 64 Havaika WM89 verecunda OAHU Havaika K85 Havaika K87 lineage MAUI NUI 90 Havaika K86 Havaika OW29 HAWAII Havaika OK24 Havaika OW28

Fig. 4. Single tree of length 213 steps resulting for the direct optimization analysis with parsimony of the ITS-2 using the following parameter values: gap opening 2, extension gap 1, transversion/transition ratio 1. Jackknife support values >50 are shown below branches. other cases, island sequence-type monophyly seemed to be analysis (Marquesan species and Hawaiian Havaika). parameter value-dependent. The verecunda clade is always Trees obtained with gap opening twice gap extension cost supported as monophyletic except in the parameter combi- and transversions twice the cost of transitions maximized nation 414. Monophyly of pubens and (morphotype D + H. congruence between the mitochondrial and nuclear parti- cruciata) clades is supported in combinations where the gap tions and were selected as reference (Fig. 5). For the extension value was smaller than gap opening. However, Hawaiian taxa, the results strongly resemble those of the none of the topological inconsistencies with the mitochon- mitochondrial data set alone, which is not surprising drial analyses received signiWcant (>50) jackknife support. given the overall higher support values for the branches Simultaneous analyses of the combined data set under in the mitochondrial tree: clades pubens, (morphotype diVerent parameter combination are summarized in D+H. cruciata), vercunda, and (morphotype D + H. Table 2. Only taxa with information for both mitochon- cruciata + vercunda) and the monophyly of each island drial and the nuclear partitions were included in this population are well supported, but the relationships

Fig. 5. Single tree of length 1037 steps resulting for the direct optimization analysis with parsimony of the combined mitochondrial and nuclear partitions using the following parameter values: gap opening 2, gap extension 1, transversion/transition ratio 2. Refer to Table 7 for clade support values. 482 M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495 between island populations in pubens and verecunda are chondrial matrix, including those clades obtained in a not. The position of the specimen HavEM130 in the Bayesian inference analyses with speciWc GTR + I +  pubens clade is sensitive to changes in the parameters of models of evolution for each of the mitochondrial parti- the analyses, with Wve out of 14 combinations (gap open- tions (16S rRNA + tRNAleu, CO1, ND1), averaged ing: gap extension: tv/ts D 212, 222, 414, 424, 444) sup- across three independent iterations with 1,500,000 gener- porting Maui Nui monophyly. Similarly, two ations and discarding 15,000, 17,000, and 22,000 trees as combinations (441, 442) put specimen HavOW111 at the burn in, respectively. base of the Big Island populations. These observations In general, results from static analyses, both in terms suggest that the topological incongruence between the of topology and clade support, are noticeably congruent mitochondrial and nuclear partitions may not be simply with trees obtained through dynamic alignment. Static noise but rather the result of diVerent molecular mecha- parsimony failed to Wnd support for the pellenines, but nisms acting on each partition. the clade is supported in model-based analyses. Mono- phyly of mainland Habronattus and Pellenes, and Mar- 3.2. Static alignments quesan species, is strongly supported in all analyses, while their interrelationships are not. The putative mono- Results of the exploratory analyses of the multiple static phyly of the Marquesan and Hawaiian endemics is nei- alignments obtained under diVerent gap opening/extension ther supported nor rejected by any of the tests costs are summarized in Table 3. Alignments of the 16S implemented, regardless of the inference method and the rRNA + tRNAleu partition generated under costs 20/2 and gap treatment (Table 5). 26/4 showed the smaller topological distances with the eli- Monophyly of Hawaiian Havaika, monophyly of each sion tree, according to the topology metrics calculated. ITS- of the pubens, (morphotype D + H. cruciata) and vere- 2 alignments under costs 24/4 and 24/6 yielded the most cunda lineages and their interrelationship, as well as similar trees to the elision matrix. Since topology metrics monophyly of island populations of each clade are recov- showed no diVerence between the two best costs, one of the ered in every analysis, with high clade support values. two best alignments of each partition was chosen arbi- Conversely, the relationships among the island popula- trarily: 20/2 for the 16S rRNA + tRNAleu partition and 24/ tions of clades pubens and verecunda, respectively, are 4 for the ITS-2. analysis-dependent, and never receive signiWcant support Analyses of the mitochondrial data sets under uni- values, with the only exception being the basal position formly weighted parsimony, treating gaps as a 5th state, of the Kauai populations of clade III relative to the Maui gaps as an absence/presence character and as missing Nui and Oahu populations, supported by the model- data, yielded eight trees of 2849 steps, four trees of 2832 based analyses. steps, and 12 trees of 2762 steps, respectively (results not Analyses of the ITS-2 data sets under uniformly shown). The AIC criterion as implemented in Modeltest weighted parsimony treating gaps as a 5th state, gaps as selected HKY85 with invariants and gamma distribution absence/presence characters and gaps as missing data, (four categories) as the most appropriate evolutionary yielded six trees of 210 steps, one tree of 169 steps, and model to analyze the data. A heuristic search using maxi- one tree of 137 steps, respectively (results not shown). mum likelihood resulted in one tree (¡log L 13,938.06) The TVM with invariants and gamma distribution (four that is shown in Fig. 6. Table 4 summarizes clade support categories) was selected as the best model for the data by for the diVerent analyses performed on the static mito- the AIC criterion, as implemented in Modeltest. A

Table 3 Summary of the results of the parsimony analyses with gaps as absence/presence characters of the static alignments produced by Clustal under diVerent values of gap opening and extension Costs length #trees A/P chars. Partition SD Triplets SD Quartets SD 16S+tRNAleu 8/2 1039 10 72 10 0.070 0.092 8/4 1053 24 58 6 0.030 0.061 20/2 1050 22 39 4 0.052 0.031 24/4 1049 7 38 5 0.061 0.035 24/6 1050 19 39 4 0.052 0.031 ITS-2 8/2 165 1 32 4 0.017 0.051 8/4 165 1 32 5 0.022 0.051 20/2 163 10 28 3 0.018 0.056 24/4 169 1 26 3 0.009 0.025 24/6 169 1 26 3 0.009 0.025 8/2: gap opening 8, gap extension 2 (remaining costs follow same convention); A/P characters: number of new characters added to the matrix after recod- ing the presence of speciWc gaps; SD: symmetric diVerence measurements. M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495 483

Fig. 6. Maximum likelihood single tree of score ¡ln D 13,938.06472 resulting from analysis of all mitochondrial gene fragments combined using the HKY + I +  model with following parameter values: transition/transversion ration D 2.156; nucleotide frequencies: A D 0.39750, C D 0.10550, G D 0.07270, T D 0.42430; among-site rate variation with invariant sites I D 0.4348 and gamma shape parameter  D 0.7497. Refer to Table 4 for clade sup- port values. heuristic search using maximum likelihood resulted in All analyses run on the static alignment of the ITS-2 two trees (¡log L 1362.99) (strict consensus shown in data set support monophyly of pubens and (morphotype Fig. 7). Table 6 summarizes clade support for the diVer- D+H. cruciata) lineages, while the verecunda clade is also ent analyses performed on the static ITS-2 data set, recovered but with marginal support only. The sister-group including clades obtained in a Bayesian inference analy- relationship of the verecunda clade to the clade (morpho- ses under the GTR + I +  model of evolution, with type D + H. cruciata) is supported under parsimony treat- 1,500,000 generations and discarding 25,000 trees as burn ing gaps either as 5th state or absence/presence characters. in. The eVect of the inclusion of gaps in the model-based In some cases, island populations of each lineage were not approximations was studied by running additional recovered as monophyletic, as was also reported for the Bayesian analyses with gaps coded as presence/absence in dynamic optimization analyses. Similarly, the major area of an independent partition and with a particular stochastic conXict with regard to the mitochondrial analyses involves model. Results were almost identical to the analyses per- single haplotypes of pubens from Oahu and Maui that clus- formed with gaps as missing data. ter with the haplotype from Kauai instead of with others 484 M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495

Table 4 trees with gaps as missing data or coded as presence/ Clade support values of all the diVerent analyses performed on static absence. However, signiWcance of the test is lost after Bon- alignment of the mitochondrial data set ferroni corrections for multiple comparisons. The same th Clade # Clade description 5 A/P ? ML BI applies for the Oahu population, although in this case low- 1NSNSNS—96est p values are obtained with gaps as 5th state. Templeton 2NSNS—6299and Winning-site tests agree in rejecting simultaneous 3NSNS——99monophyly of the former groups, regardless of the gap 4—NS——97 W 5NSNS———treatment, although in some cases signi cance is lost after 66860—5897Bonferroni corrections. The Shimodaira–Hasegawa test of 76258NS—98the maximum likelihood trees agrees in rejecting the simul- 8NSNSNS——taneous monophyly of the Maui Nui and Oahu popula- 9727566——tions of pubens and the Oahu populations of verecunda, 10 Pellenines NS NS NS 84 97 11 89 76 86 81 95 although it fails to reject any of them when constrained 12 — — — 51 — independently. The ILD test reveals signiWcant incongru- 13 Pellenes 100 100 100 100 100 ence (p < 0.001) between the mitochondrial and ITS-2 data 14 100 100 100 100 100 sets, regardless of the gap treatment enforced. 15 NS NS NS — — The mitochondrial and nuclear data sets were merged in a 16 Habronattus 78 79 78 83 95 17 Marquesas sp. 100 99 99 100 100 single matrix and incomplete taxa for one of the two parti- 18 Hawaiian Havaika 100 100 100 100 100 tions removed. Analyses of the combined data set under par- 19 pubens 99 98 99 100 100 simony analyses with gaps as 5th state, as absence/presence 20 pubens Maui Nui 72 73 74 81 99 characters and as missing data, resulted in 10 trees of 1052 21 — — — 52 — steps, eight trees of 1004 steps, and 10 trees of 955 steps, 22 100 100 100 100 100 23 NS NS NS — NS respectively. The AIC criterion implemented in Modeltest 24 pubens Big Island 100 100 100 100 100 selected K81uf + I + G as the best model for the combined 25 — — — 65 98 data set. ML analysis resulted in one tree (¡logL 26 NS NS NS NS — 7734.19040). Results using Bayesian inference were obtained 27 pubens Kauai 100 100 100 100 100 by averaging through three independent runs of the MCM- 28 pubens Oahu 100 100 100 100 99 29 99 99 99 94 99 CMC and discarding in each run 35,000 trees as burn in. 30 100 100 100 100 100 Table 7 summarizes clade support across all the analyses 31 verecunda +D+H. cruciata 95 95 96 100 100 performed on the combined nuclear and mitochondrial 32 D + H. cruciata 100 100 100 100 100 matrix. Results nearly mirror those obtained for the mito- 33 H. cruciata Big Island 99 100 100 100 100 chondrial data set alone, except for the lower support for 34 verecunda 99 99 99 100 100 35 verecunda Kauai 100 100 100 100 100 the pubens and (morphotype D + H. cruciata + verecunda) 36 — — — 50 — lineages in the maximum likelihood analysis. The similarity 37 90 88 88 90 — of the result is not surprising given that the conXicting 38 — — — 67 95 clades reported in the partial analyses of the ITS data set 39 verecunda Maui Nui 99 100 100 100 100 were only marginally supported. 40 99 99 99 100 100 41 verecunda Oahu 100 100 100 100 100 42 87 88 87 83 — 3.3. Lineage divergence time 43 100 100 100 100 100 Numbers refer to jackknife support (parsimony analyses), bootstrap pro- Divergence times were estimated on the topology portions (maximum likelihood), and posterior probabilities (Bayesian obtained in the direct optimization analysis of all gene frag- Inference). Clade numbers as shown in Fig. 6. —: clade supported below ments (parameter combination gap opening 2, gap exten- threshold value (50% for jackknife and bootstrap, 95% posterior probabil- sion 1, tv/ts 2). The same topology was obtained in the ity); NS: clade not supported; Wfth: parsimony analyses with gaps as 5th Bayesian analyses of the static matrices and it is not in con- state; A/P: parsimony analysis with gasp recoded as absence/presence X data; ?: parsimony analysis with gaps as missing data; ML: maximum ict with any of the remaining analyses. Only mitochon- likelihood analysis; BI: Bayesian inference analysis. drial genetic divergences were used since there was evidence that some ITS-2 relationships may not reXect historical pat- from the same island. These results are only (very weakly) terns (see Discussion for details). Age estimation was supported under maximum likelihood. Statistical tests on restricted to the Marquesan species and Hawaiian Havaika the topologies obtained by constraining the island sequence representatives. Bianor was included in the analyses as a types of each lineage to monophyly oVer mixed results distant outgroup to asses the root node, but it was pruned (Table 5). The monophyly of the Oahu sequence types of before calculation of divergence times. Maximum likeli- verecunda is never rejected but the monophyly of the other hood branch lengths were obtained with PAUP, using the groups is dependent on the test and the gap treatment. model selected by AIC in Modeltest with the preferred Monophyly of the Maui Nui sequence types of pubens is topology constrained (K81uf + I + G). The split between the only marginally rejected by the Templeton test in some Maui and Big Island populations of morphotype D + H. M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495 485

Table 5 Summary of the results of the statistical tests of alternative topologies Constraint 5th A/P ? ML #tTWS#tTWS#tTWS#tSH ITS-2 None14——1——1——2 pubens Maui Nui 72 0.0833–0.2568 0.25–0.4531 3 0.0455¤–0.1025 0.125–0.2188 234 0.0455¤–0.1025 0.125–0.2188 2 0.162 pubens Oahu 24 0.0008¤–0.0093¤ 0.0010¤–0.0169¤ 3 0.0143¤–0.0339¤ 0.0313¤–0.0703 393 0.0143¤–0.0578 0.0313¤–0.125 1 0.067 verecunda Oahu 32 0.3173–0.763 1.0000 1 0.3173 1.0000 64 – — 2 0.855 All 40 0.0001¤–0.0017¤ <0.0001¤–0.0026¤ 3 0.0067¤–0.0209¤ 0.0078¤–0.0391¤ 242 0.0114¤–0.0335¤ 0.0156¤–0.0703 2 0.041¤ Mitochondrial None 8 4 14 1 MA + HI 8 0.5930–0.9434 0.4334–0.8036 8 0.8185–0.8709 0.8784-0.9087 4 0.8982–0.9014 1.0000 1 0.445 Constraints: None: no constraint enforced; pubens Maui Nui: monophyly of all ITS-2 sequence types of the pubens lineage from Maui Nui; pubens Oahu: monophyly of all ITS-2 sequence types of the pubens lineage from Oahu; verecunda Oahu: monophyly of all ITS-2 sequence types of the verecunda lineage from Oahu; All: all former constraints enforced; MA + HI: monophyly of haplotypes from Marquesas and Hawaiian Havaika. 5th: parsimony analyses with gaps as 5th state; A/P: parsimony analysis with gaps recoded as absence/presence data; ?: parsimony analysis with gaps as missing data; ML: maxi- mum likelihood analysis. #t: number of trees; T: p values of the Templeton test; WS: p values of the Winning-site test; SH: p values of the Shimodaira– Hasegawa test. ¤ SigniWcance at p <0.05. cruciata is the only well-supported node available for cali- gence constrained) and simultaneous constraint of the time bration inside Hawaiian Havaika. The age of this node was of divergence of the two Maui Nui-Big Island splits in the Wxed to 0.5 million years (my), assuming this age of origin tree (set to 0.5 my) resulted in divergence time estimates of the Big Island (Price and Clague, 2002). Other branching that were far older than the inhabited islands. In both cases, patterns in pubens and verecunda clades that also followed cross-validation selected the NPRS method over the PL the timing of island formation were initially not used as cal- method. However, when only the split between the Maui ibration points because they were poorly supported by the Nui and the Big Islands populations of pubens was Wxed data (low bootstrap or jackknife values). The split between (0.5 my), the PL method was preferred (smoothing the two Marquesan species provide an external calibration parameter D 1), and the estimated ages were again compati- point. However, in this case the age of the node was not ble with island ages. Moreover, the estimated time of diver- Wxed but constrained to a minimum age of 2.01 my and a gence between morphotype D and H. cruciata was only maximum of 4 my, which corresponds to the age of the old- slightly higher than the age of the youngest islands (Table est and youngest islands inhabited by these species (Craig 7), although the resulting average rate of sequence evolu- et al., 2001). Age estimation was performed using the penal- tion was higher than in previous analyses (0.03843). ized likelihood method and the truncated Newton algo- rithm, with the smoothing parameter set to 1000. These 3.4. Morphological analyses settings were selected after a preliminary cross-validation analyses that included other values of the smoothing As it currently stands, the taxonomy of Hawaiian Hava- parameter as well as the NPRS and Langley–Fitch meth- ika is of little use for species identiWcation. However, phylo- ods. Problems with local optima in the calculation of diver- genetic analyses provide strong support for the separation gence times were addressed by running three analyses with of Hawaiian species into lineages characterized by face random starting values. Results are shown in Table 7. The ornament patterns. The pubens lineage is characterized by average rate of sequence evolution observed was 0.02649 the presence of yellowish scales on the chelicerae in both per site per million years, which is about twice the rate males and females. The verecunda lineage shows sexual reported for arthropod mitochondrial DNA (Brower, dimorphism with males bearing rows of white bristles on 1994). Most of the age estimates are in accordance with the the chelicerae and females having few bristles on the chelic- geological setting, as far as they suggest time of origins of erae but dense white hairs on the clypeus. A single female lineages that are younger than the age of the islands they specimen from Maui Nui shows sparse clypeal white scales inhabit. However, there is one major exception: The split restricted to the lower margin of the clypeus (Fig. 2), a between the Maui Nui and the Big Island populations of unique pattern not observed in any other female specimen pubens is inferred to be about three times older than the Big and, therefore, it is assigned to a new lineage: morphotype Island. A new set of divergence estimate analyses were run D. A fourth lineage, H. cruciata, includes specimens from with alternative constrained clade ages. Both removal of the Big Island in which face ornamentation patterns are internal age constraints (i.e. only Marquesan species diver- undistinguishable from the verecunda lineage. However, 486 M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495

Fig. 7. Maximum likelihood single tree of score ¡ln D resulting from analysis of ITS-2 using the TVM + I +  model with following parameter values: R matrix D (2.5611, 3.3526, 3.8112, 0.6771, 3.3526); nucleotide frequencies: A D 0.1750, C D 0.2967, G D 0.3049, T D 0.2234; among-site rate variation with invariant sites I D 0.5434 and gamma shape parameter  D 0.6870. Refer to Table 6 for clade support values. bulb characters provide further discrimination in this mid-frontal part, with two longitudinal spots of whitish group. The verecunda clade has a short palpal tibia (palp hairs, and abdomen light cream colored with very variable cymbium/tibia ratio ranging from 1.333 to 1.500) and a darker chevron patterns. Morphotype D stands in sharp stout embolus emerging from the distal part of the tegulum contrast to H. cruciata in terms of body coloration, being (tegulum/embolus origin ratio ranging from 0.500 to 0.576). almost uniformly dark brown with grayish longitudinal Conversely, the only H. cruciata male sampled combines a patterns on the abdomen. short palpal tibia (1.529) with a slender, long embolus origi- Phylogenetic data indicate that island populations in nating from the basal half of the tegulum (0.278). The same each lineage are monophyletic (considering Maui Nui as a type of embolus is also observed in pubens males (tegulum/ single island, see above), although none of the morphologi- embolus origin ratio ranging from 0.281 to 0.450), cal characters considered seem to distinguish between although, in this case, they have a long palpal tibia (palp diVerent island populations. Genetic divergences (Table 8) cymbium/tibia ratio ranging from 0.750 to 1.278). Along between islands in the same clade range from 0.022 to 0.082 with face ornaments and male palp characters, other (within island values: 0.004–0.039) and provide further evi- somatic traits characterize the diVerent lineages. Specimens dence that island populations are well characterized from a in the pubens clade have a dark brown carapace, with two molecular standpoint. Lineages (pubens, verecunda) were lighter longitudinal bands on the dorsal part, and a cream accordingly further divided into island populations. colored abdomen with a variable darker chevron pattern The morphological study of about 200 additional speci- and shiny scales on the dorsal side. Specimens in this clade mens identiWed additional island populations of the afore- are stouter and more robustly built than species in other mentioned lineages. The pubens lineage was sampled in the clades. H. cruciata and the verecunda lineage have a black molecular analysis from West and East Maui but it is also carapace, sometimes bearing a brush of white hairs on the present in the former Maui Nui islands of Molokai and M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495 487

Table 6 ability diVers among variables: Em/Te explains 0.720 from Clade support values of all the diVerent analyses performed on static the 1.287 of the total explainable inertia, followed by Cy/Tb alignment of the ITS-2 data set (0.568) and CL (0.438) last. The two Wrst canonical axes Clades Clade description 5th A/P ? ML BI (eigenvalues 0.780 and 0.293, respectively; sum of all canon- 1 Hawaiian Havaika 100 100 100 100 100 ical eigenvalues 1.287) accounted for 12% of the cumulative 2 pubens 61 87 75 75 100 percentage variance of species data. Permutation tests of 3 pubens Big Island 82 85 60 NS — the Wrst canonical axis and all canonical axes combined 4NSNSNSNS—W V 5NSNSNS—NSrevealed signi cant di erences among groups (lineages/ 6NS——NS—islands) (Wrst axis: F-ratio D 8.449, p D 0.0010; trace: F- 7NS——NS—ratio D 4.951, p D 0.0010). The ability of the measurements 8 5756558396 to discriminate among pubens, verecunda, and morphotype 9 ——537597 D+H. cruciata lineages is clearly seen in a biplot with the 10 NS — — 52 — 11 —— —68— lineages/island groups and the measured variables of the 12 D + H. cruciata 100 100 96 97 100 two Wrst discriminant axes (Fig. 8A). The plots in Fig. 8 also 13 verecunda —— —— — illustrate the similarity of genitalic characters of the indi- 14 verecunda Kauai — — — 71 — viduals from Necker–Nihoa to verecunda, although the Wrst 15 83 85 — 88 — tend to be of bigger size. Conversely, a scatter plot with all 16 — — 52 54 — 17 — 57 — 78 — the individuals analyzed and enveloped by lineage/island 18 verecunda Maui Nui 68586253— (Fig. 8B) shows that the diVerent island populations are 19 —— —61— hardly distinguishable based on the measurements consid- H. cruciata Big Island — — NS — NS ered and that the same hold for the H. cruciata and mor- verecunda +D+H. cruciata 69 58 — — — photype D. Interestingly, the very few cases where Values refer to (i) jackknife support in parsimony analyses; (ii) bootstrap individuals from a particular lineage had canonical scores proportions for maximum likelihood analysis; and (iii) posterior probabil- that overlapped with other lineages always corresponded to ities for Bayesian inference. Clade numbers as shown in Fig. 7. —: clade V supported below threshold value (50% for jackknife and bootstrap, 95% di erent island populations. In the only example of coexis- posterior probability); NS: clade not supported; 5th: parsimony analyses tence of the three lineages, Maui Nui, the diVerent popula- with gaps as 5th state; A/P: parsimony analysis with gasp recoded as tions were clearly very distinct. absence/presence data; ?: parsimony analysis with gaps as missing data; ML: maximum likelihood analysis; BI: Bayesian inference analysis. 3.6. Size variation

Lanai. The same holds for the verecunda lineage, although A plot of the mean and standard deviation of carapace molecular data were also available for the Molokai popula- length in each island population is shown in Fig. 9. Results tion. Morphotype D was also found on Lanai, but not on of the two-way analyses of variance are summarized in Molokai. Males belonging to this morphotype were not Table 10. Both the overall comparison and comparisons sampled in the molecular analyses. Assignment of males to within islands for which male and female specimens were morphotype D was made possible by the characteristic available show that diVerence in size between sexes and the somatic coloration observed in females and received further interaction between lineage and sex were not signiWcant. support by the similarity of the bulb pattern of these speci- However, diVerences between lineages were clearly signiW- mens to H. cruciata, its putative sister taxa (Fig. 2). Finally, cant in most of the comparisons performed. On Kauai, specimens collected in the Northwest Hawaiian Islands of Molokai, and West Maui, where both sexes were not avail- Necker and Nihoa did not Wt in any of the deWned morpho- able for one of the lineages, t tests showed signiWcant diVer- types, although they more closely resemble verecunda and ences in all cases between the two lineages within islands H. cruciata and C in terms of male genitalia and face orna- (Kauai: t D 5.99, df D 67, p < 0.0001; Molokai: t D 3.96, mentation. df D 21, p D 0.0007; West Maui: t D 4.47, df D 9, p D 0.0016). The only situations where diVerences between lineages were 3.5. Discriminant analysis of lineages not signiWcant were those islands with three lineages. On Maui Nui, pubens was signiWcantly larger than verecunda Table 9 summarizes size and male genitalia measure- and morphotype D, but there was no signiWcant diVerence ments obtained from the 240 Hawaiian Havaika specimens between verecunda and morphotype D. The same holds available for study, categorized into 10 groups correspond- when the analysis was restricted to Lanai and East Maui, ing to the lineages discussed above and their island popula- the only Maui Nui islands where the three lineages have tions plus the Necker–Nihoa individuals. Both marginal been found. Interestingly, comparisons between island pop- and accumulative permutation tests (999 permutations) ulations within lineages using one-way ANOVA test (Table revealed signiWcant (p < 0.0167, after Bonferroni correction) 11) reveal that individuals of both pubens and verecunda discrimination power of the three variables (CL D carapace from Oahu are signiWcantly smaller than their counterparts length, Cy/Tb D palp cymbium/tibia ratio, Em/ from Maui Nui. None of the remaining comparisons were Te D embolous/tegulum ratio). However, discrimination signiWcant. 488 M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495

Table 7 Clade support values of analyses performed on the combined mitochondrial and nuclear data sets, and clade age estimates (see text for details) Clades Clade description 212 5th A/P ? ML BI Node 15 D 0.5 my Node 8 D 0.5 my Age Min Max Age Min Max 0 Marquesas — 2.01 4.00 — 2.01 4.00 1 Hawaiian Havaika 100 100 100 100 100 100 3.78 3.68 4.04 2.71 2.63 2.90 2 pubens lineage 100 100 99 100 — 100 2.11 2.07 2.24 1.28 1.24 1.35 3 0.20 0.20 0.22 0.13 0.13 0.14 4 — — — — — — 2.08 1.95 2.12 1.18 1.11 1.23 5 pubens Oahu 100 100 100 100 100 100 0.44 0.42 0.46 0.27 0.26 0.28 6 93 97 97 97 95 100 0.04 0.03 0.04 0.03 0.02 0.03 7 94 99 99 99 100 100 0.14 0.14 0.17 0.09 0.09 0.11 8 — — — — 57 — 1.61 1.57 1.71 0.50 — — 9 pubens Big Island 100 100 100 100 100 100 0.13 0.12 0.15 0.06 0.06 0.07 10 — — — — 82 — 0.06 0.05 0.06 0.03 0.03 0.03 11 pubens Maui Nui — — — — — 99 1.23 1.21 1.33 0.43 0.42 0.44 12 93 92 89 91 99 100 0.96 0.94 1.03 0.36 0.36 0.38 13 100 100 100 100 100 100 0.05 0.04 0.05 0.03 0.02 0.02 14 verecunda +D+H. cruciata 95 99 93 92 — 100 2.48 2.42 2.63 1.79 1.74 1.91 15 D + H. cruciata 98 100 100 100 100 100 0.50 — — 0.62 0.58 0.64 16 H. cruciata 99 100 100 100 100 97 0.15 0.14 0.15 0.15 0.14 0.16 17 verecunda lineage 98 99 98 99 — 100 1.46 1.31 1.42 1.04 0.93 1.01 18 verecunda Maui Nui 99 100 99 100 100 100 0.61 0.57 0.64 0.43 0.40 0.45 19 98 99 100 99 100 100 0.33 0.22 0.25 0.17 0.15 0.17 20 — — — — — NS 1.35 1.31 1.42 0.96 0.93 1.01 21 verecunda Kauai 100 100 100 100 95 100 0.42 0.40 0.45 0.29 0.28 0.31 22 100 100 100 100 96 100 0.42 0.41 0.46 0.30 0.29 0.32 23 verecunda Oahu 100 100 100 100 80 100 0.19 0.18 0.20 0.13 0.12 0.14 24 69 71 54 71 76 — 0.26 0.26 0.30 0.18 0.18 0.21 25 100 100 100 100 100 100 0.01 0.01 0.01 0.01 0.01 0.01 Support values refer to jackknife in direct optimization (212, gap opening 2, gap extension 1, and tv/ts ratio 1) and static parsimony (5th: gaps as 5th state; A/P: gaps as absence/presence; ?: gaps as missing data;), bootstrap in maximum likelihood (ML) and posterior probabilities in Bayesian Inference (BI). Estimated age (age) and inferred conWdence intervals (min: minimum, max: maximum) obtained when Wxing alternatively Node 15 or Node 8 age to be 0.5 my. In all cases, Marquesan species divergence was constrained to a minimum age of 2.01 my and a maximum age 4 my. Clade numbers as shown in Fig. 5. —: clade supported below threshold value (50% for jackknife and bootstrap, 95% posterior probability); NS: clade not supported.

Table 8 Average corrected (K81uf + I + G model) genetic distances between and within (diagonal) Marquesan representatives and island populations of each Hawaiian Havaika lineage Marquesas pubens D Maui H. cruciata Big Island verecunda Kauai Oahu W Oahu K Maui Big Island Kauai Oahu Maui N Marquesas 0.068 0.19 0.196 0.196 0.194 0.199 0.16 0.203 0.185 0.197 0.190 pubens Kauai 0.009* 0.082 0.079 0.075 0.069 0.11 0.128 0.154 0.152 0.145 pubens Oahu W 0.009 0.024 0.073 0.074 0.1 0.124 0.146 0.121 0.126 pubens Oahu K 0.006 0.07 0.067 0.095 0.118 0.148 0.125 0.124 pubens Maui 0.039 0.056 0.101 0.119 0.142 0.128 0.123 pubens Big Island 0.004 0.105 0.12 0.136 0.124 0.116 D Maui — 0.022 0.081 0.072 0.071 H. cruciata Big Island 0.017 0.095 0.094 0.080 verecunda Kauai 0.021 0.067 0.059 verecunda Oahu 0.012 0.048 verecunda Maui N 0.014 Mitochondrial and nuclear data sets combined, except *: only mitochondrial data available.

4. Discussion the current taxonomic position of H. rufescens (recently transferred from Sandalodes Keyserling, 1883, to which 4.1. The Marquesan species Havaika species were originally assigned, to Habronattus). However, H. rufescens lacks any of the synapomorphies of In the present study, the two Marquesan species were Habronattus, including the presence of an elbowed tegular always shown as sister taxa and consistently grouped with apophysis (TA), which makes its current taxonomic posi- the continental Habronattus, in some cases as sister taxa, tion questionable. Indeed, lack of a TA seems to be a diag- and the representatives of Pellenes. These results support nostic character for Havaika and is used to distinguish the M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495 489

Table 9 Measurements of morphological structures (in mm) Lineage Island Sex Carapace length Cy/Tb Em/Te N Mean SD Max Min N Mean SD Mean SD pubens Kauai m 10 3.091 0.485 4.335 2.704 6 1.040 0.099 0.420 0.033 f0———— Oahu m 20 2.922 0.461 3.927 2.244 14 1.071 0.223 0.351 0.034 f 13 2.790 0.310 3.25 2.295 Maui Nui (Molokai, Lanai, Maui) m 9 3.273 0.463 4.160 2.601 10 0.994 0.105 0.338 0.040 f 15 3.191 0.427 3.570 2.805 Big Island m 13 3.025 0.492 3.984 2.193 9 1.025 0.149 0.396 0.042 f 15 3.143 0.413 3.927 2.601 D Maui Nui (Lanai, Maui) m 5 2.558 0.374 2.907 2.132 5 1.355 0.104 0.360 0.035 f 3 2.832 0.411 2.938 2.703 H. cruciata Big Island m 10 2.372 0.303 2.652 1.785 7 1.434 0.222 0.395 0.087 f 10 2.506 0.374 3.060 2.040 verecunda Kauai m 29 2.314 0.368 2.990 1.742 23 1.535 0.229 0.520 0.050 f 30 2.468 0.229 2.964 2.091 Oahu m 13 2.363 0.298 2.856 1.836 11 1.336 0.148 0.485 0.040 f 18 2.669 0.206 2.856 2.448 Maui Nui (Molokai, Lanai, Maui) m 10 2.611 0.276 3.162 2.193 7 1.477 0.168 0.564 0.052 f 14 2.406 0.272 3.016 2.040 ?? Necker, Nihoa m 3 2.960 0.479 3.237 2.407 3 1.180 0.143 0.537 0.032 f 2 2.499 0.036 2.703 2.652 Cy/Tb: cymbium/palp tibia ratio; Em/Te: embolous origin/tegulum ratio; m: males; f: females; N: number of specimens measured; SD: standard deviation; max: maximum values reported; min: minimum value reported. genus from Habronattus (Prószyjski, 2002). The second the observation that the TA has been independently lost in Marquesan species included in this analysis does have a the Marquesas suggests that a simple bulb is not a synapo- TA, although it is not elbowed and is much thinner than in morphy of the endemics of both archipelagos. More data continental Habronattus. A close comparison of the bulbs will be required to reject completely the inclusion of the of the two Marquesan species analyzed (Fig. 2) shows them Marquesan species in the genus Havaika, although the data to be almost identical except for the presence of the TA and at hand seem to point towards this direction. the shape of the embolus tip. The absence of the TA in H. rufescens reveals that the embolus originates not on the lat- 4.2. Species delimitation in Hawaiian Havaika eral margin of the tegulum, as in the Hawaiian species, but closer to its mid-part. A detailed examination of the only Taxonomic delimitation of species within Havaika has available drawing of the bulb of a Marquesan Havaika been hampered by the lack of clear-cut diagnostic charac- (Berland, 1933) shows clearly the embolus originating not ters. Havaika specimens do not present many variable char- on the lateral but towards the basal mid-part of the tegu- acters and the few available tend to be polymorphic. The lum. If this structure reXects phylogenetic aYnity, then H. lability of some characters used in the original descriptions rufescens and the new species are likely to be the same evo- of Havaika species was largely overlooked because of the lutionary lineage as the Marquesan Havaika and can thus reduced number of specimens available for examination be used as representatives for the phylogenetic position of (Simon, 1900). In this study, we circumvent some of the this lineage. Genetic divergence between H. rufescens and past limitations by combining the use of molecular and the undescribed species is about the same range reported morphological information for a large sample of individu- among island populations of lineages in the Hawaiian als in a statistical framework. Our results clearly support Islands (Table 8), which suggests that both Marquesan spe- the existence in Hawaiian Havaika of at least four geneti- cies originated in situ and are consequently endemic to cally distinct lineages that can be diagnosed on the basis of these islands. characters used in traditional taxonomy: the pubens lineage, The relationship between the Marquesan and the the verecunda lineage, H. cruciata, and morphotype D. A Hawaiian species and, therefore, the monophyly of Havaika putative Wfth lineage would include the specimens from as currently deWned deserves further consideration. As Necker and Nihoa, although molecular data are currently stated above, all analyses agree in separating Marquesan unavailable. Additionally, mitochondrial DNA sequence from Hawaiian species. However, clades supporting this data show that the island populations of each lineage form split are poorly supported by the data (e.g. Fig. 6, nodes 12 well-supported, reciprocally monophyletic groups with and 15), and the trees obtained are not signiWcantly better overall genetic divergences between islands well above than topologies resulting from constraining the monophyly within island variation (pubens: between-island of the Marquesan and Hawaiian representatives. However, mean D 0.070, S.D. D 0.008, within-island mean D 0.018, 490 M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495

A Table 10 Two-way ANOVA for lineage and sex eVects on size (carapace length) Cy/Tb across the Hawaiian Archipelago (All) and in each island Island d.f. MS FpComparisons All Lineage 3 7.5924 53.828 <0.0001 ACDB Sex 1 0.2075 1.471 0.2264 pubens Kauai Interaction 3 0.0714 0.506 0.6783 pubens Oahu pubens Maui Nui Error 234 0.1410 pubens Hawaii Oahu H. cruciata Hawaii morph D Maui Nui Lineage 1 8931.6 45.340 <0.0001 AC verecunda Kauai Sex 1 175.6 0.891 0.3489 Em/Te verecunda Oahu CL verecunda Maui Nui Interaction 1 150.0 0.761 0.3864 -4 6 morph Necker-Nihoa Error 60 364.4 -4 6 Waianaes Lineage 1 0.05658 22.222 0.0002 AC B Sex 1 0.00244 0.958 0.3400 Interaction 1 0.00530 2.083 0.1652 Error 19 0.00255 Koalas Lineage 1 2.9824 19.261 <0.0001 AC Sex 1 0.1464 0.945 0.3369 Interaction 1 0.0539 0.348 0.5584 Error 39 0.1548 Maui Nui Lineage 2 3.088657 30.50239 <0.0001 ACD Sex 1 0.000139 0.00138 0.9705 -3 4 Interaction 2 0.162440 1.60420 0.2112 -6 6 Error 50 0.101260 Lanai + EMaui Fig. 8. Canonical correspondence analysis. (A) Plot of the Wrst two dis- Lineage 2 1.47726 14.1543 0.0002 ACD criminant axes with discriminating characteristics (CL: carapace length, Sex 1 0.00869 0.0832 0.7763 Cy/Tb: ratio between the length of the cymbium and the palpal tibia, E/ Interaction 2 0.06202 0.5943 0.5624 Te: ratio between the distance from the base of the tegulum to the base of Error 18 0.10437 the embolus and the total length of the tegulum embolous and tegulum Big Island length ratio) and centroids of classes (lineages and island populations) Lineage 1 4.838069 28.88641 <0.0001 AB shown. (B) Plot of canonical scores of each specimen on Wrst two discrimi- Sex 1 0.183742 1.09706 0.3006 nating axes. Points corresponding to diVerent lineages/islands are shown Interaction 1 0.000749 0.00447 0.9470 with diVerent symbols and an envelope encloses all points belonging to a Error 44 0.187486 particular class. In column ‘Comparisons’ the lineages that did not diVer signiWcantly (SNK test) are underlined. 4

3.5

3 ported clade and both lineages display diagnostic morpho-

2.5 logical characters. pubens 2 verecunda The pubens and verecunda lineages include several nomi- H. cruciata nal species each. However, our data suggest that within 1.5 morph D these lineages geographical distribution is a better indicator 1 Necker-Nihoa of genealogical relationships than the morphological char- Carapace length (mm's) 0.5 acters used in the original descriptions of the species, which 0 are either variable within island populations (color of the Necker-Nihoa Kauai Oahu Maui Nui Hawaii ocular setae, Simon, 1900) or overlap across ranges (posi- Fig. 9. Plot of carapace length mean and standard deviation bars for each tion of the embolus relative to the bulb, Prószyjski, 2002). lineage and island population. In any case, formal description of new species and taxo- nomic amendments are beyond the scope of this study and will be published elsewhere. S.D. D 0.026; verecunda: between-islands meanD 0.058, S.D. D 0.010, within-island mean D 0.016, S.D. D 0.005). 4.3. Origin and biogeography of Hawaiian Havaika Genetic divergence between H. cruciata and morphotype D is the lowest among groups, in the range of within-island Hawaiian Havaika shared their most recent common variation divergences observed in other species groups ancestor between 3.86 my (3.68–4.04 my) and 2.7 my (2.63– (between-H. cruciata and morphotypes D mean D 0.024, 2.90 my) ago depending on the time constraint imple- S.D. D 0.001). Still, H. cruciata haplotypes form a well-sup- mented. These values represent the minimum ages of colo- M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495 491

Table 11 islands. If they did, this might suggest a basal position for One-way ANOVA for island eVect on size (carapace length) of each line- the Necker and Nihoa morphotype, which is most similar age to verecunda lineage. However, given that the molecular Morph d.f. MS Fp SigniWcant comparisons data indicate that verecunda is more distal, the morphology pubens would better support a more recent colonization of the Between islands 3 2594.9 3.71 0.0143 Oahu vs. Maui Nui islets from the younger current high islands. The tree topol- Error 91 699.0 ogy as it currently stands does not allow establishing the verecunda Between islands 3 0.5304 6.25 0.0006 Oahu vs. Maui Nui polarity of these characters and thus they cannot be used to Error 115 0.0849 ascertain the phylogenetic position of the Necker and D Nihoa morphotype. Therefore, resolution of the correct Between islands 1 0.0941 1.16 0.3227 None placement of the Necker and Nihoa species will require new Error 6 0.0811 material from these islands for DNA analyses. The most general biogeographic pattern observed in Hawaiian taxa is what has been referred to as the “progres- nization of the archipelago by Havaika. At that time, Kauai sion rule,” which results from dispersal events (often associ- and perhaps also Oahu were the only existing high islands ated with speciation) strongly skewed in the direction from (Price and Clague, 2002). Although the hotspot that formed older to younger islands (Funk and Wagner, 1995). This the current Hawaiian Archipelago has been active for 80 pattern is found in many groups of Hawaiian arthropods my and islands with similar features have formed in the (Roderick and Gillespie, 1998). In the case of the Hawaiian same location during this period (Carson and Clague, Havaika, progression can neither be accepted nor refuted: 1995), recent studies have shown that only during certain H. cruciata and morphotype D are found on two adjacent periods of time were the extant islands high and close island and do not provide information on the progression enough to each other for existing islands to contribute to pattern. Similarly, although pubens and verecunda clades the biota of new islands that emerged (Price and Clague, include taxa from four and three islands, respectively, the 2002). SpeciWcally, there have been two peak periods during existence (or lack thereof) of a progression is not obvious. which more than one island of elevation above 1000 m However, when one considers the ages of the island popula- existed, the Wrst, between 18 and 8 Mya, and the second 5 tions in each clade (Table 7), then the most plausible expla- Mya corresponding to the current high islands. These two nation for lack of obvious progression becomes apparent. time periods were also associated with periods of maximum In the pubens lineage the time of split of the Kauai popula- closeness (even contact) between islands. Conversely, there tion is 2.11 my (2.07–2.24 my) or 1.28 my (1.24-1.35 my), were two time periods (27–28 and 5–7 my) when volcanoes depending on the time constraint, while the separation of were at their most distantly spaced and relatively low. The Oahu from the Maui Nui + Big Island clade is set to 2.08 fact that one of these periods was immediately prior to the my (1.95–2.12 my) or 1.18 my (1.11–1.23 my). In the vere- formation of the current high islands might explain why cunda lineage, the Maui Nui split is estimated to be approx. most of the current biodiversity in the Hawaiian Islands 1.46 my (1.31–1.42 my) or 1.04 my (0.93–1.01 my) and the seems to have been originated from outside the archipelago divergence of the Kauai and Oahu population about 1.35 since the formation of Kauai (Price and Elliott-Fisk, 2004). my (1.31–1.42 my) or 0.96 my (0.93–1.01 my). Age estimates The time and pattern of formation of the islands is also for each time constraint clearly show that time of diver- consistent with age estimates of the origin of Hawaiian gence of the Kauai, Oahu, and Maui Nui populations of the Havaika. However, specimens collected in the 1950s on the pubens and verecunda lineages largely overlap, which is con- islets of Necker (11.0 my) and Nihoa (7.3 my) have been sistent with a rapid dispersal through the islands by each clearly identiWed during the course of the present study as lineage right after their origination. This is in agreement Havaika. These two dry volcanic rocky islands with maxi- with the geological timeframe given that Kauai, Oahu, and mum areas of 0.18 and 0.57 km2, and elevations of 84 and Maui Nui had already emerged at the estimated time of 273 m, respectively, are the remnants of islands that reached diversiWcation of these clades. Indeed, among plants where elevations above 1000 m immediately after their formation. times of colonization of the Hawaiian Islands have been Now the islets are mostly barren and vegetation is estimated to fall within a similar time frame, the same kind restricted to a few species of grasses and bushes that survive of rapid colonization of multiple islands (in particular, on their southern slopes. The species of Havaika on these Kauai + Oahu, sometimes also Maui) has been found, for islands have adapted to a harsher and drier environment example, in Viola (Violaceae) (Ballard and Sytsma, 2000), than their counterparts in the present high-islands, which Kokia (Malvaceae), and Hesperomannia (Asteraceae) mostly occur in high elevation mesic and wet forests. (Funk and Wagner, 1995). Unfortunately, no recent specimens of Havaika have been An additional proof for the rapid dispersion across the collected from these islets so they could not be included in archipelago of the pubens and verecunda lineages can be found any molecular analysis. Additional data will be required to in the incongruence detected between the mitochondrial and determine whether these species, because of the older age of the ITS-2 data sets. Mitochondrial data supported mono- Necker and Nihoa, gave rise to those on the current high phyly of the island populations of each clade (Figs. 3 and 6), 492 M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495 but there were some instances of non-monophyly in ITS-2 the diVerent lineages of Havaika exploit diVerent resources. sequence types from the same specimens (Figs. 4 and 7). ITS-2 However, size diVerentiation has frequently been linked to is part of the tandemly repeated nuclear ribosomal DNA clus- ecological segregation (Schoener, 1970). In all cases where ter. Concerted evolution tends to homogenize variants within two or more Havaika lineages were found to coexist in a sin- species and populations, while divergences among them are gle island, one of them was always signiWcantly larger than accentuated (Hillis et al., 1991). In spite of this, intragenomic the others (Fig. 9 and Table 10). The lineages diVer by a ratio variation has been documented in several metazoan taxa of 1.25–1.30, which is about the average of the values (Duran et al., 2004; Hugall et al., 1999; Van Oppen et al., reported for other sympatric species (Schluter, 2000b). This 2002; Vogler and Desalle, 1994). Several factors can be relationship was always in the same direction: pubens was the responsible for the presence of multiple copies of ITS in a sin- largest lineage. On the other hand, in Necker and Nihoa, the gle genome, including the existence of rDNA loci at diVerent only islands inhabited by a single lineage specimens were genomic locations or hybridization events. Additionally, rapid observed to be intermediate in size relative to the size of lin- speciation can also generate intragenomic variability if con- eages on other islands, except Oahu where the bigger lineage certed evolution is then slower than the speciation rate (Harris is similar in size to Necker and Nihoa. Three lineages coexist and Crandall, 2000). In the present study, ITS sequences were in Lanai and East Maui: pubens is signiWcantly larger that the generated directly from PCR products and hence intrage- other lineages, but there was no signiWcant size diVerence nomic variation may have been overlooked. However, during between verecunda and morphotype D. The Lanai and East manual editing of the sequences, we observed some cases Maui populations are the largest island populations of compatible with the presence of more than one ITS copy, pubens while mean, maximum, and minimum sizes of mor- including double peaks at single positions or unreadable photype D are larger (but not signiWcantly so) than those sequence fragments because of overlapping peaks. In fact, reported for its sister group, H. cruciata, which inhabits a extensive intragenomic variation in the same marker has been two-species island. However, sizes of the verecunda lineage do found in other Hawaiian spiders of the genera Theridion and not seem to be smaller than they are on other islands. In fact, Argyrodes (Arnedo and Gillespie, in preparation). The mis- verecunda individuals on Oahu are signiWcantly smaller than matches observed in the mitochondrial and ITS-2 topologies those on Maui Nui. This pattern could suggest that the sys- of pubens and verecunda lineages could then be explained by tem is not in equilibrium, i.e. competition is still in the process invoking Wxation or dominance in consensus sequence of of shaping this community, as suggested by the fact that vere- alternative sequence types from ancestral polymorphisms as a cunda colonized these island recently and much later than the result of the rapid colonization of islands. An alternative other lineages (about 1.42–1.31 my—these are maximum esti- explanation to the apparent polyphyly of some island mates, see Table 7 for more recent alternatives). This general sequence types would be the occurrence of hybridization pattern of accentuation of size diVerences when species co- events among islands. However, the fact that conXicting occur resembles that reported for other island organisms for sequence types (EM130, OW111) are related to sequence which it has been demonstrated to be the result of ecological types from islands that are not geographically contiguous ren- character displacement as, for example, Cnemidophorus liz- ders this possibility quite unlikely. ards from the Sea of Cortez islands (Radtkey et al., 1997), Anolis lizards in the Lesser Antilles (Losos, 1990), mustelids 4.4. Inferred evolutionary processes in Britain and Ireland (Dayan and SimberloV, 1994), mon- gooses in the Indian and PaciWc islands, and Darwin’s Wnches The overall pattern of evolution of Havaika corresponds to in the Galapagos (Grant, 1972). Unfortunately, the lack of an early split into diVerent morphologies directly upon coloni- phylogenetic information on the only solitary morphotype zation of the Hawaiian Islands and subsequent dispersal of (Necker and Nihoa) and the weak support recovered for the each lineage along the island chain. The pattern of early mor- internal branching patterns of each lineage of Havaika pre- phological diversiWcation matches what has been reported for cluded testing some of the predictions of the character dis- other organisms. For example, a shift to cursorial behavior in placement hypothesis, namely that (1) size diVerences the web-building spider genus Tetragnatha seems to have between species should be greater in sympatry, (2) species on occurred very early in the evolutionary history of the group in one-species islands should be intermediate in size, and (3) the the Hawaiian Islands (Gillespie et al., 1994). Planthoppers of ancestors of sympatric species occurring on one-species the genus Nesosydne spread from an ancestral monocotyle- islands should be of intermediate size (Schluter, 2000a). don host onto a larger array of diVerent dicotyledons shortly Our data strongly support a single origin of size diVeren- after colonization of the Hawaiian Islands, and maintained tiation early after the arrival of Havaika in the Hawaiian their host-association as they progressed down the island Archipelago. This result suggests that competitive exclusion chain (Roderick, 1997). Additional cases would include the probably explains better the current distribution of lineages Orthotylus mirids (Polhemus, 2002) and Blackburnia carabid that size assortment (Losos, 1990). Under this scenario, beetles (Liebherr and Zimmerman, 1998). only those species already suYciently diVerent in body size Most of the examples listed above involved ecological could invade and coexist on an island. shifts associated with the exploitation of new resources. At The Wnding that individuals of both pubens and vere- present, there is no direct Weld or experimental evidence that cunda on the island of Oahu are signiWcantly smaller than M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495 493 those on Maui Nui presents a conundrum. As mentioned Acknowledgments above, the presence of a third morphotype may explain the larger size of pubens in Maui Nui, but not the larger size of This study would have not been possible without the verecunda in the same islands. Roughgarden (1992) and extensive work on PaciWc Salticidae and the pioneering Roughgarden and Pacala (1989) developed the Taxon- studies on Hawaiian Havaika of James Berry, Joseph Cycle hypothesis as an alternative to character displace- Beatty, and Jerzy Prószyjski. Wayne Maddison provided ment to explain the pattern of body size observed in the insightful comments for the discussion, and together with Anolis lizards in the Lesser Antilles: On an island already Marshall Hedin supplied unpublished (at that time) DNA inhabited by a species of intermediate size, subsequent colo- sequences. Xavier Turón and Eduardo Mateos contributed nization by a larger species might lead both species to valuable expertise on statistical methods. The Nature Con- evolve toward smaller sizes, resulting in the larger species servancy of Hawaii (M. White, P. Bily), the State Depart- approaching the optimal intermediate solitary size, with the ment of Land and Natural Resources, the Hawaii Natural intermediate species being pushed, Wrst to the exploitation Areas Reserve System (B. Gagne), and West Maui Land of marginal resources, and subsequently to extinction. and Pineapple (R. Bartlett) provided permits and logistic However, reevaluation of the Taxon-Cycle in Anolis using support for collecting in their lands. Many specimens used phylogenetic contrasts has failed to support this trend in the study were collected in collaboration with or directly (Miles and Dunham, 1996). A simpler explanation for such provided by the following colleagues: Ingi Agnarsson, patterns is that they are simply artifacts resulting from Todd Blackledge, Jessica Garb, Mandy Heddle, Gustavo small sample sizes. Hormiga, and Nikolaj ScharV. Additional specimens where In any case, to test the validity of the hypothesis in obtained as a loan from the following institutions and indi- Havaika, more molecular information is needed, both to viduals: Jane Beccaloni (BNHM), Frank Howarth identify the phylogenetic position of the Necker and Nihoa (BPBM), Norman Platnick (AMNH), Christine Rollard morphotype and to resolve the internal branching pattern (MNHN), and James Berry. Salvador Carranza provided of the pubens and verecunda lineages. suggestions on the Wrst drafts of the paper, and Gustavo In conclusion, our study shows that the Hawaiian Hava- Hormiga contributed valuable comments to the Wnal ver- ika arrived relatively late in the formation of the current set sion of the paper. Funding for this research was provided of high Hawaiian Islands. At the time they colonized, by a postdoctoral grant of the Spanish Ministry of Educa- Kauai, Oahu, and perhaps also some part of Maui Nui, had tion (M.A.), the National Science Foundation and the already formed. More importantly, large numbers of spi- Schlinger Foundation (R.G.). ders had already diVerentiated and radiated within the archipelago template, in particular Tetragnatha, together References with thomisid (and likely also philodromid) crab spiders, would already have dominated the aerial environment Arnedo, M.A., Oromí, P., Ribera, C., 2001. Radiation of the spider genus (Garb, 1999; Gillespie, 2004). The abundant ecological Dysdera (Araneae, Dysderidae) in the Canary Islands: cladistic assess- ment based on multiple data sets. Cladistics 17, 313–353. space that would have been available to these early coloniz- Arnedo, M.A., Coddington, J., Agnarsson, I., Gillespie, R.G., 2004. From a ers would have been more limited for Havaika. However, it comb to a tree: phylogenetic relationships of the comb-footed spiders does appear that diVerent islands within the Hawaiian (Araneae, Theridiidae) inferred from nuclear and mitochondrial genes. archipelago are inhabited by Havaika species that diVer sig- Mol. Phylogenet. Evol. 31, 225–245. niWcantly in size. Moreover, our analyses support a single Baldwin, B.G., Sanderson, M.J., 1998. Age and rate of diversiWcation of the Hawaiian Silversword Alliance (Compositae). Proc. Natl. Acad. Sci. origin of the two-size classes that is in turn linked to a USA 95, 9402–9406. marked shift in sexual secondary characters. Therefore, we Ballard Jr., H.E., Sytsma, K.J., 2000. Evolution and biogeography of the may envision a scenario whereby Havaika, upon coloniza- woody Hawaiian violets (Viola, Violaceae): Arctic origins, tion of the islands, diversiWed into two lineages character- herbaceous ancestry and bird dispersal. Evol. Int. J. Org. Evol. 54, ized by secondary sexual and male genital characters. Size 1521–1532. W Barker, F.K., Lutzoni, F.M., 2002. The utility of the incongruence length segregation may have occurred early in the diversi cation diVerence test. Syst. Biol. 51, 625–637. of the group perhaps as a result of character displacement Berland, L., 1933. Araignees Des Iles Marquises. Bull. Bernice P Bishop in sympatric lineages. Subsequent colonization of the Mus. 114, 39–70. remaining islands (both older and younger) appears to have Berland, L., 1934. Araignées De Polynesie. Ann. Soc. Entomol. Fr. 103, been achieved by each lineage independently, with size 321–336. V Berland, L., 1942. Polynesian spiders. Occasional Pap. Bernice P Bishop di erences being accentuated on each island, perhaps Mus. 17, 1–24. through competitive exclusion. Berry, J.W., Beatty, J.A., Prószyjski, J., 1998. Salticidae of the PaciWc The interesting evolutionary patterns displayed by Islands. III. Distribution of seven genera with descriptions of nineteen Havaika spiders render the genus an excellent model for new species and two new genera. J. Arachnol. 26, 149–189. studying the role of competition in shaping ecological com- Blackledge, T.A., Gillespie, R.G., 2004. Convergent evolution of behavior in an adaptive radiation of Hawaiian web-building spiders. Proc. Natl. munities. This research paves the way for more detailed Acad. Sci. USA 101, 16228–16233. studies on the evolutionary biology of these previously Brocchieri, L., 2001. Phylogenetic inferences from molecular sequences: overlooked spiders. review and critique. Theor. Popul. Biol. 59, 27–40. 494 M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495

Brower, A.V.Z., 1994. Rapid morphological radiation and convergence Gillespie, R.G., Croom, H.B., Palumbi, S.R., 1994. Multiple origins of a among races of the butterXy Heliconius erato inferred from patterns of spider radiation in Hawaii. Proc. Natl. Acad. Sci. USA 91, 2290–2294. mitochondrial DNA evolution. Proc. Natl. Acad. Sci. USA 91, 6491–6495. Giribet, G., 2001. Exploring the behavior of Poy, a program for direct Brown, F.B.H., 1921. Origin of the Hawaiian Xora. Special Publications. optimization of molecular data. Cladistics 17, S60–S70. Bernice P. Bishop Mus. 7, 131–142. GoloboV, P.A., 1993. Nona. . Buckley, T.R., Arensburger, P., Simon, C., Chambers, G.K., 2002. Com- GoloboV, P.A., Farris, J.S., Nixon, K.C., 2003. Tnt: tree analysis using new bined data, Bayesian phylogenetics, and the origin of the New Zealand technology, version 1.0. Available at: . Caccone, A., Milinkovitch, M.C., Sbordoni, V., Powell, J.R., 1997. Mito- Grant, P.R., 1972. Convergent and divergent character displacement. Biol. chondrial DNA rates and biogeography in European newts (genus J. Linnean Soc. 4, 39–68. Euproctus). Syst. Biol. 46, 126–144. Gregory, H.E., 1928. Types of PaciWc Islands. Proc. Third Pan PaciWc Sci. Campbell, D.H., 1933. The Xora of the Hawaiian Islands. Quart. Rev. Biol. Congr. 2, 1662. 8, 164–184. Gregory, J.W., 1930. The geological history of the PaciWc Ocean. Proc. Carson, H.L., 1989. Natural hybridization between the sympatric Hawai- Geol. Soc. 86, 72–136. ian species Drosophila silvestris and Drosophila heteroneura. Evolution Guillaumin, A., 1928. Les Regions Floristiques Du PaciWque D’apres Leur 43, 190–203. Endemisme Et La Repartition De Quelques Plantes Phanerogames. Carson, H.L., Clague, D.A., 1995. Geology and biogeography of the Proc. Third Pan-PaciWc Sci. Congr. 1, 920–928. Hawaiian Islands. In: Wagner, W.L., Funk, V.A. (Eds.), Hawaiian Bio- Guindon, S., Gascuel, O., 2003. A simple, fast, and accurate algorithm geography. Smithsonian Institution, Washington, pp. 14–29. to estimate large phylogenies by maximum likelihood. Syst. Biol. 52, Carson, H.L., Sato, J.E., 1969. Micro evolution within 3 species of Hawai- 696–704. ian Drosophila. Evolution 23, 493–501. Harris, D.J., Crandall, K.A., 2000. Intragenomic variation within Its1 and Craig, D.A., Currie, D.C., Joy, D.A., 2001. Geographical history of the Its2 of freshwater crayWshes (Decapoda: Cambaridae): implications for central-western PaciWc black Xy subgenus Inseliellum (Diptera: Simu- phylogenetic and microsatellite studies. Mol. Biol. Evol. 17, 284–291. liidae: Simulium) based on a reconstructed phylogeny of the species, Hedin, M.C., 1997. Molecular phylogenetics at the population/species hot-spot archipelagoes and hydrological considerations. J. Biogeogr. interface in cave spiders of the southern appalachians (Araneae: Nes- 28, 1101–1127. ticidae: Nesticus). Mol. Biol. Evol. 14, 309–324. Cunningham, C.W., 1997. Can three incongruence tests predict when data Hedin, M.C., Maddison, W.P., 2001. A combined molecular approach to should be combined? Mol. Biol. Evol. 14, 733–740. phylogeny of the jumping spider subfamily Dendryphantinae (Ara- Dayan, T., SimberloV, D., 1994. Character displacement, sexual dimor- neae: Salticidae). Mol. Phylogenet. Evol. 18, 386–403. phism, and morphological variation among British and Irish mustelids. Hillis, D.M., Moritz, C., Porter, C.A., Baker, R.J., 1991. Evidence for biased Ecology 75, 1063–1073. gene conversion in concerted evolution of ribosomal DNA. Science Duran, S., Giribet, G., Turon, X., 2004. Phylogeographical history of the 251, 308–310. sponge crambe crambe (Porifera, Poecilosclerida): range expansion Hormiga, G., Arnedo, M., Gillespie, R.G., 2003. Speciation on a conveyor and recent invasion of the Macaronesian Islands from the Mediterra- belt: sequential colonization of the Hawaiian Islands by orsonwelles nean Sea. Mol. Ecol. 13, 109–122. spiders (Araneae, Linyphiidae). Syst. Biol. 52, 70–88. Emerson, B.C., Kolm, N., 2005. Species diversity can drive speciation. Huelsenbeck, John, P., Larget, B., SwoVord, D., 2000. A compound Pois- Nature 434, 1015–1016. son process for relaxing the molecular clock. Genetics 154, 1879–1892. Emerson, B.C., Oromi, P., Hewitt, G.M., 2000. Tracking colonization and Hugall, A., Stanton, J., Moritz, C., 1999. Reticulate evolution and the ori- diversiWcation of insect lineages on islands: mitochondrial DNA phylo- gins of ribosomal internal transcribed spacer diversity in apomictic geography of Tarphius canariensis (Coleoptera: Colydiidae) on the meloidogyne. Mol. Biol. Evol. 16, 157–164. Canary Islands. Proc. R. Soc. Biol. Sci. Ser. B 267, 2199–2205. Jordan, S., Simon, C., Polhemus, D., 2003. Molecular systematics and Farris, J.S., Kallersjo, M., Kluge, A.G., Bult, C., 1994. Testing signiWcance adaptive radiation of Hawaii’s endemic damselXy genus megalagrion of incongruence. Cladistics 10, 315–320. (Odonata: Coenagrionidae). Syst. Biol. 52, 89–109. Fleischer, R.C., Mcintosh, C.E., Tarr, C.L., 1998. Evolution on a volcanic Liebherr, J.K., Zimmerman, E.C., 1998. Cladistic analysis, phylogeny and conveyor belt: using phylogeographic reconstructions and K–Ar-based biogeography of the Hawaiian Platynin (Coleoptera: Carabidae). Syst. ages of the Hawaiian Islands to estimate molecular evolutionary rates. Entomol. 23, 137–172. Mol. Ecol. 7, 533–545. Losos, J.B., 1990. A phylogenetic analysis of character displacement in Funk, V.A., Wagner, W.L., 1995. Biogeographic patterns in the Hawaiian Caribbean Anolis Lizards. Evolution 44, 558–569. Islands. In: Wagner, W.L., Funk, V.A. (Eds.), Hawaiian Biogeography: Losos, J.B., 2000. Ecological character displacement and the study of Evolution on a Hot-Spot Archipelago. Smithsonian Institution Press, adaptation. Proc. Natl. Acad. Sci. USA 97, 5693–5695. Washington, pp. 379–419. Maddison, W., Hedin, M., 2003a. Phylogeny Of Habronattus jumping spi- Ganders, F.R., Berbee, M., Pirseyedi, M., 2000. Its base sequence phylog- ders (Araneae: Salticidae), with consideration of genital and courtship eny in Bidens (Asteraceae): evidence for the continental relatives of evolution. Syst. Entomol. 28, 1–22. Hawaiian and Marquesan Bidens. Syst. Bot. 25, 122–133. Maddison, W.P., Hedin, M.C., 2003b. Jumping spider phylogeny (Araneae: Garb, J.E., 1999. An adaptive radiation of Hawaiian Thomisidae: biogeo- Salticidae). Invertebr. Syst. 17, 529–549. graphic and genetic evidence. J. Arachnol. 27, 71–78. Masta, S.E., Maddison, W.P., 2002. Sexual selection driving diversiWcation Garb, J.E., Gillespie, R.G., 2006. Island hopping across the Central PaciWc: in jumping spiders. Proc. Natl. Acad. Sci. USA 99, 4442–4447. mitochondrial DNA detects sequential colonization of the Austral Meyrick, E., 1935a. Pyrales and microlepidoptera of the Marquesas Islands by crab spiders (Araneae: Thomisidae). J. Biogeogr. 33, 201–220. Islands. Bernice P. Bishop Mus. Bull. 114, 333–355. Gillespie, R.G., 1999. Naivete and novel perturbations: conservation of Meyrick, E., 1935b. Pyrales and microlepidoptera of the Society Islands. native spiders on an Oceanic Island system. J. Insect Conserv. 3, 263–272. Bernice P. Bishop Mus. Bull. 113, 109–110. Gillespie, R.G., 2002. Biogeography of spiders on remote Oceanic Islands of Mickevich, M.F., Farris, J.S., 1981. The implications of incongruence in the PaciWc: archipelagoes as stepping stones? J. Biogeogr. 29, 655–662. Menidia. Syst. Zool. 30, 351–370. Gillespie, G., 2004. Community assembly through adaptive radiation in Miles, D.B., Dunham, A.E., 1996. The paradox of the phylogeny: character Hawaiian spiders. Science 303, 356–359. displacement of analyses of body size in Island Anolis. Evolution 50, Gillespie, R.G., 2005. Geographical context of speciation in a radiation of 594–603. Hawaiian Tetragnatha spiders (Araneae, Tetragnathidae). J. Arachnol. Mindell, D.P., 1991. Similarity and congruence as criteria for molecular 33, 313–322. homology. Mol. Biol. Evol. 8, 897–900. M.A. Arnedo, R.G. Gillespie / Molecular Phylogenetics and Evolution 41 (2006) 472–495 495

Nixon, K.C., 2002. Winclada, version 1.00.08. Available at: . Ecol. Evol. 2, 175–178. Nunn, P.D., 1994. Oceanic Islands. Blackwell, Oxford. Simon, E., 1900. Arachnida. In: Fauna Hawaiiensis, or the Zoology of the Page, R.D.M., 1993. Component: Tree Comparison Software For Micro- Sandwich Isles. Royal Society of London Promoting Natural Knowl- soft Windows. The Natural History Museum, London. edge and the British Association for the Advancement of Science, Lon- Pilsbry, H.A., 1900. The genesis of mid-paciWc faunas. Proc. Acad. Nat. Sci. don, pp. 443–519. Philadelphia 52, 568–581. Simon, C., Frati, F., Beckenbach, A., Crespi, B., Liu, H., Flook, P., 1994. Polhemus, D.A., 2002. An initial review of Orthotylus in the Hawaiian Evolution, weighting, and phylogenetic utility of mitochondrial gene Islands, with descriptions of twenty-two new species (Heteroptera: sequences and a compilation of conserved polymerase chain reaction Miridae). J. NY Entomol. Soc. 110, 270–340. primers. Ann. Entomol. Soc. Am. 87, 651–701. Posada, D., Crandall, K.A., 1998. Modeltest: testing the model of DNA SwoVord, D.L., 2001. Paup*: Phylogenetic Analysis Using Parsimony (and substitution. Bioinformatics 14, 817–818. Other Methods). Sinauer Associates, Sunderland, MA. Potvin, C., RoV, D.A., 1993. Distribution-free and robust statistical meth- Templeton, A.R., 1983. Phylogenetic inference from restriction endonucle- ods: viable alternatives to parametric statistics? Ecology 74, 1617. ase cleavage site maps with particular reference to the humans and Prager, E.M., Wilson, A.C., 1988. Ancient origin of lactalbumin from lyso- apes. Evolution 37, 221–244. zyme: analysis of DNA and amino acid sequences. J. Mol. Evol. 27, Thompson, J.D., Gibson, T.J., Plewniak, F., Jeanmougin, F., Higgins, D.G., 326–335. 1997. The Clustal_X windows interface: Xexible strategies for multiple Price, J.P., Clague, D.A., 2002. How old is the Hawaiian biota? Geology sequence alignment aided by quality analysis tools. Nucl. Acids Res. and phylogeny suggest recent divergence. Proc. R. Soc. Biol. Sci. Ser. B 25, 4876–4882. 269, 2429–2435. Thorne, J.L., Kishino, H., 2002. Divergence time and evolutionary rate Price, J.P., Elliott-Fisk, D., 2004. Topographic history of the Maui Nui estimation with multilocus data. Syst. Biol. 51, 689–702. Complex, Hawai’i, and its implications for biogeography. PaciWc Sci. Thorne, J.L., Kishino, H., Felsenstein, J., 1992. Inching toward reality: an 58, 27–45. improved likelihood model of sequence evolution. J. Mol. Evol. 34, 3–16. Prószyjski, J., 2003. Salticidae (Araneae) of the world. . Spawning times, reproductive compatibilities and genetic structur- Prószyjski, J., 2002. Remarks on Salticidae (Aranei) from Hawaii, with ing in the Acropora Aspera group: evidence for natural hybridiza- description of Havaika. Gen. Nov. Arthropoda Selecta. 10, 225–241. tion and semi-permeable species boundaries in corals. Mol. Ecol. Radtkey, R.R., Fallon, S.M., Case, T.J., 1997. Character displacement in 11, 1363–1376. some cnemidophorus lizards revisited: a phylogenetic analysis. Proc. Vandergast, A.G., Gillespie, R.G., Roderick, G.K., 2004. InXuence of vol- Natl. Acad. Sci. USA 94, 9740–9745. canic activity on the population genetic structure of Hawaiian tetrag- Roderick, G.K., 1997. Herbivorous insects and the Hawaiian silversword natha spiders: fragmentation, rapid population growth and the alliance: co-evolution or co-speciation? PaciWc Sci. 51, 440–449. potential for accelerated evolution. Mol. Ecol. 13, 1729–1743. Roderick, G.K., Gillespie, R.G., 1998. Speciation and phylogeography of Vingron, M., Von Haeseler, A., 1997. Towards integration of multiple align- Hawaiian terrestrial arthropods. Mol. Ecol. 7, 519–531. Issn: 0962-1083. ment and phylogenetic tree construction. J. Comput. Biol. 4, 23–34. Ronquist, F., Huelsenbeck, J.P., 2003. Mrbayes 3: Bayesian phylogenetic Vitousek, P.M., Loope, L.L., Stone, C.P., 1998. Introduced species in inference under mixed models. Bioinformatics 19, 1572–1574. Hawaii: biological eVects and opportunities for ecological research. Roughgarden, J., 1992. Comments on the paper by Losos: character dis- Trends Ecol. Evol. 2, 224–227. placement versus taxon loop. Copeia, 1992, 288–295. Vogler, A.P., Desalle, R., 1994. Evolution and phylogenetic information Roughgarden, J., Pacala, S., 1989. Taxon cycle among Anolis lizard popu- content of the Its-1 region in the tiger beetle Cicindela dorsalis. Mol. lations: review of the evidence. In: Otte, D., Endler, J.A. (Eds.), Specia- Biol. Evol. 11, 393–405. tion and Its Consequences. Sinauer, Sunderland, MA, pp. 403–442. Wheeler, W.C., 1995. Sequence alignment, parameter sensitivity, and the Sanderson, M.J., 1997. A nonparametric approach to estimating diver- phylogenetic analysis of molecular data. Syst. Biol. 44, 321–331. gence times in the absence of rate constancy. Mol. Biol. Evol. 14, Wheeler, W.C., 1996. Optimization alignment: the end of multiple 1218–1231. sequence alignment in phylogenetics? Cladistics 12, 1–9. Sanderson, M.J., 2002. Estimating absolute rates of molecular evolution Wheeler, W.C., Hayashi, C.Y., 1998. The phylogeny of the extant chelicer- and divergence times: a penalized likelihood approach. Mol. Biol. Evol. ate orders. Cladistics 14, 173–192. 19, 101–109. Wheeler, W.C., Gatesy, J., Desalle, R., 1995. Elision: a method for accom- Sanderson, M.J., 2003. R8s: inferring absolute rates of molecular evolution modating multiple molecular sequence alignments with alignment- and divergence times in the absence of a molecular clock. Bioinformat- ambiguous sites. Mol. Phylogenet. Evol. 4, 1–9. ics 19, 301–302. Wheeler, W.C., Gladstein, D.S., De Laet, J.E., 1996–2003. Poy. Version 3.0. Schluter, D., 2000a. Ecological character displacement in adaptive radia- ftp.amnh.org/pub/molecular/poy (Current Version 3.0.11). Documen- tion. Am. Nat. 156, S4–S16. tation by Daniel Janies and Ward Wheeler. Commandline documenta- Schluter, D., 2000b. The Ecology of Adaptive Radiation. Oxford Univer- tion by J. De Laet and W.C. Wheeler. Amnh, New York. sity Press, Oxford. White, T.J., Bruns, T., Lee, S., Taylor, J., 1990. AmpliWcation and direct Schoener, T.W., 1970. Size patterns of West Indian anolis lizards. II. Corre- sequencing of fungal ribosomal RNA genes for phylogenetics. In: Innis, lations with the sizes of particular sympatric species-displacement and M., Gelfand, D., Swinsky, J., White, T.J. (Eds.), Pcr Protocols: A Guide to convergence. Am. Nat. 104, 155–174. Methods and Applications. Academic Press, San Diego, CA, pp. 315–322. Shaw, K.L., 1996. Sequential radiations and patterns of speciation in the Wiens, J.J., 1998. Combining data sets with diVerent phylogenetic histories. Hawaiian cricket genus Laupala inferred from DNA sequences. Evolu- Syst. Biol. 47, 568–581. tion 50, 237–255. Wilson, J.T., 1963. A possible origin of the Hawaiian Islands. Can. J. Phys. Shimodaira, H., Hasegawa, M., 1999. Multiple comparisons of log-likeli- 41, 135–138. hoods with applications to phylogenetic inference. Mol. Biol. Evol. 16, Yoder, A.D., Yang, Z., 2000. Estimation of primate speciation dates using 1114–1116. local molecular clocks. Mol. Biol. Evol. 17, 1081–1090. Simmons, M.P., 2004. Independence of alignment and tree search. Mol. Young, N.D., Healy, J., 2002. Gapcoder. . Simmons, M.P., Ochoterena, H., 2000. Gaps as characters in sequence- Zar, J.H., 1984. Biostatistical Analysis, 2nd ed. Prentice-Hall, Englewood based phylogenetic analyses. Syst. Biol. 49, 369–381. CliVs, NJ.