Available online at www.sciencedirect.com SCIENCE //^) D I R E C T < > < < m 00 09 O O ^- en en fN n >? >? ? < > <: < S > < < i n ri >< >? < < ?ri in o o ?o ? n ri >" >? < < T Ov o >0 oo o ? oo o ?. ? o fn fn m rn ri rg n (N < < < < \o r^ oo o n n n n ^ CJN C7^ C7^ O O O O m m m m ri n ri n >?><>?>- < < < < ^- rf in m ?n r^ t^ >n 0^ ^ ^ C7^ O O & O en en en en n n n n >?>>?>? < < < < r~ 00 o\ Q ?n tn ?n ^ O O O O en m en en ri rj n (N >">?>?>< < << < g t/5 VI ON r- OO * 00 _^ s 1^ =? il 00 ^ .s ? J(3 ?S o ?o ? 3 . So Cu u o ?* 2 S ? -2 ? .Sf* ~ OU n c u S ^ ?? ?? ?, c '-' 3 Crt Il il ! s. " -?i s l3 S ?S -S- IJ en ri >- < OV o ON O S I r^ r^ r*\ ri n ri >? >- >- < < < t wn o = = ? r^ f*^ r*^ r'? o o o o f^ r-i r^ f^ ?N r-1 ?N ?S << << ?n r- r~- Qs C7\ 0\ O O Q <^ m f*S fN ri fN (N > > > > << << en ??* O o << << t~- -f -t s, vo ?n m ?li X X < S < > < ri ri >" < X >> < ON rl >? < rl >> < ? rl >> < ? 8 o ^ r ? o O < < > < < n t/? o O u -a ca ? T3 .h "O ? o ?>? CJ 8 Cos S s; M 3 X O- . . o u asi- ese^ ?2 -3 -2 ^ p: K h. (- ti fu fu ? ^ ^. EQ ? ? ? M ? ? rt w =3 ,-o 3 3 3 3 3 3 3 3 3 _-g ?T3 =3 .-o -o 3 3 ?o 3 ^ i ?C 'C *c 'C 'C 'C 'C 'C 'C u ? ? U u ? u u u j= ^ si J= .c ja JC j= ^ f (- y- f- H ?- H r^ 3 ?. ?. This paper provides the first phylogcnetic hypothesis for theridiid intrageneric relationships based on molec- ular data. The results allow us to test current hypotheses of relationships based on morphology, and provide a framework to analyze the great array of ecological and behavioral traits displayed by the family. 2. Materials and methods 2.1. Taxonomic sampling Theridiid terminals included 40 species representing 33 of the most species-rich and ecologically diverse genera. Important genera, in tenns of species diversity and morphological distinctiveness that could not be sampled in the present study include: Caniiella Thaler & Steinbcrger, 1988 (8 species); Coleosoma O.P.-Cam- bridgc, 1882 (10 species); Coscinida Simon, 1895 (13 species); Ecliinotlieridioii Ixvi, 1963 (9 species); Hadro- tarsus Thorell, 1881 (5 species); Molioiie Thorell, 1892 (3 spp.); Moneta O.P.-Cambridge, 1870 (18 species); Pai- discura Archer, 1950 (4 species); Takayus Yoshida, 2001 (13 species); Tekellina Levi, 1957 (6 species.); Theonoe Simon, 1881 (6 species). Only the genus Trigonobothrys Simon, 1889 have been included from the recently res- urrected hadrotarsine genera (e.g., Yoshida, 2002) which, although diverse, are poorly de?ned and have dubious species composition. Representatives of nine additional araneoid families were included to test theridiid monophyly. In all the analyses, exemplars from the family Araneidae were used as the primary outgroup under the assumption of their sister-group relationship to the remaining arane- oids (Griswold et al., 1998). More than one species of the genera Argyrodes, Tlieridioii, and Anelosiiniis were included in the analysis to test some contrasting views on their taxonomic limits. The list of the specimens sampled in the present study is shown in Table 1. 2.2. Characters Live specimens were collected in the field and fixed in 95% ethanol, except when fresh material was not avail- able, in which case specimens from museum collections (preserved in 75% ethanol) were used for extractions, with success mostly dependent on the time since pres- ervation. Only one or two legs were used for extraction, except for specimens preserved in 75% EtOH, for which as many as four legs plus the carapace were used. The remainder of the specimen was kept as a voucher (de- posited at the National Museum of Natural History, Smithsonian Institution, in Washington, DC and the Essig Museum of Entomology, University of California 230 M. A. Arnedo el eil. I Molecular Phylogenelics and Evolution 31 (2004J 225-245 at Berkeley). Total genomic DNA was extracted fol- lowing the phenol/chlorofomi protocol of Palumbi et al. (1991) or using Qiagcn DNeasy Tissue Kits. The ap- proximate concentration and purity of the DNA ob- tained was evaluated through spectophotometry and the quality was verified using electrophoresis in agarose/ TBE (1.8%) gel. Partial fragments of the mitochondrial genes cytochrome c oxidase subunit I (COI) and 16S rRNA (16S) and the nuclear genes 18S rRNA (IBS), 28S rRNA (28S) and Histone H3 (H3) were amplified using the following primer pairs: [COI] Cl-J-1751 and Cl-N- 2191 (designed by R. Harrison's lab, Simon et al., 1994), [16S] LR-N-13398 (Simon et al., 1994) and LR-J-12864 (CTCCGGTTTGAACTCAGATCA, Hsiao, pers. comm.), [18S] 5F or 18Sa2.0 and 9R (Giribet et al., 1999), [28S] 28SA and 28SB (Whiting et al., 1997), and [H3] H3aF and H3aR (Colgan et al., 1998). The thermal cyclers Perkin-Elmcr 9700, Perkin-Elmer 9600, and Bio-Rad ?Cycle were used indiscriminately to perform either 25 (mitochondrial genes) or 40 (nuclear genes) iterations of the following cycle: 30 s at 95 ?C, 45 s at 42- 58 ?C (depending on the primers, see below), and 45 s at 72 ?C, beginning with an additional single cycle of 2 min at 95 ?C and ending with another one of 10 min at 72 ?C. Positive amplification for COI and 16S primers was achieved at annealing temperatures ranging from 42 to 45 ?C. For the 28S and H3 a single annealing tempera- ture of 48 ?C yielded positive amplifications in most cases. For the 18S primer a "touchdown" strategy was applied, beginning at 58 ?C and lowering proportionally the temperature in each cycle for 20 cycles down to 45 ?C and keeping that annealing temperature for an addi- tional 20 cycles. The PCR reaction mix contained primers (0.48 \xM each), dNTPs (0.2 mM each), and 0.6 U Perkin-Elmer AmpliTaq DNA polymerase (for a 50^1 reaction) with the supplied buffer and, in some cases, adding an extra amount of MgCb (0.5-1 .OmM). PCR results were visualized by means of an agarose/ TBE (1.8%) gel. PCR products were cleaned using Geneclean II (Bio 101) or Qiagen QIAquick PCR Pu- rification Kits following the manufacturer's specifica- tions. DNA was sequcnccd directly in both directions through the cycle sequencing method using dye termi- nators (S?nger et al., 1977) and the ABI PRISM BigDye Terminator Cycle Sequencing Ready Reaction with AmpliTaq DNA Polymerase FS kit. Sequenced prod- ucts were cleaned using Princeton Separations CentriSep columns and run out on an ABI 377 automated se- quencer. Sequence errors and ambiguities were edited using the Sequencher 3.1.1 software package (Gene Codes). Sequences were subsequently exported to the program GDE 2.2 (Genetic Data Environment) (Smith et al., 1994) running on a Sun Enterprise 5000 Server, and manual alignments built, for management purposes, taking into account secondary structure information from secondary structure models available in the liter- ature for 16S (Arnedo et al., 2001), 28S (Ajuh et al., 1991), and 18S (Hendriks et al., 1988). Alignment of the protein-coding genes was trivial since no length varia- tion was observed in the sequences. 2.3. Analysis 2.3.1. Alignment Insertions and deletions (hereafter called either indels or gaps) arc common events in the evolution of non- protcin-coding DNA sequences, as inferred from dif- ferent length fragments resulting from amplification of homologous DNA regions across different taxa. Indel events present two main challenges in phylogenetic analysis of DNA sequence data: positional homology (i.e., alignment) and indel treatment (Giribet and Wheeler, 1999). Unlike nucleotide bases, indels are not observable characters but gaps inserted to accommodate homologous DNA sequences of unequal length to define the putative homologous characters amenable to phy- logenetic analysis. Although homologous landmarks (e.g., secondary structure in structural genes) can facil- itate manual sequence alignment, they almost never re- solve all the ambiguity and subjectivity in the position assignment. As a result of these problems, a common approach is to avoid or discard regions that have ex- perienced such events (Lee, 2001). However, gaps can contain important phylogenetic information that can have dramatic effects on tree topology and clade support (Simmons et al., 2001). Alternative methods for incor- porating indels include using automatic algorithms to evaluate objective optimality function. In particular, programs using the dynamic programming algorithm of Needlcman and Wunsch (1970) provide methods of aligning sequences thai are repcatablc. Automatic alignment algorithms require explicit parameter costs (e.g.. gap opening, extension costs, and transition/ transversion ratio) and thus provide a comparative framework to investigate the effects of changes in these parameters. Regardless of the actual method or pa- rameter cost scheme employed, the final outcome is one or several alignments that are then subject to phyloge- netic analysis. The most common way to incorporate gaps into the analysis once an alignment is obtained is by considering them either missing data or as a state in addition to the four nucleotides (i.e., gaps as 5th state). The reasons cited for treating gaps as missing data include the lack of a proper treatment to deal with events that are the re- sults of processes different from those acting on base substitution (Swoffbrd et al., 1996) or their lack of reli- able phylogenetic information (Simmons et al., 2001). Recently, Simmons and Ochoterena (2000) have criti- cized the gaps as 5th state approach because gaps do not constitute alternative forms of bases but are essentially a different form of change. In addition, scoring gaps as a M.A. Arnedo el al. I Molecular Phylogenetics and Evolution 31 (2004) 225-245 231 5th state can result in treating contiguous gap positions as multiple independent characters although they are most parsimoniously considered as a single indel event. These authors recommended that indels be scored as additional absence/presence characters, according to a set of rules based on gap overlaps and sharing of the 5' and/or the 3' termini (Simmons and Ochoterena, 2000). This two-step procedure (alignment followed by analysis) for analyzing DNA sequences of different length considers alignment and tree search as logically independent steps (Simmons and Ochoterena, 2000). Alternatively, the alignment can be considered as an integral part of the phylogenetic analysis. This approach considers indels as transformations not observations and thus claims it is logically inconsistent to consider them as characters. 2.3.2. Optimization alignment The optimization alignment method, also referred as the direct optimization method (Wheeler, 1996), cir- cumvents this inconsistency, and the whole alignment issue, by incorporating indels as one of the possible transformations during the optimization process linking ancestral and descent nucleotide sequences. Unlike the standard two-step procedure where a "static" alignment (Wheeler, 2001) is constructed and submitted to phylo- genetic analysis, optimization alignment produces an alignment that is tree-dependent and thus the homology statements are dynamic instead of fixed or static. In the current study, optimization alignment analyses were performed with the computer program POY v. 2.0 (Gladstein and Wheeler, 1997). Due to computational demands most of the analyses were run in a twenty-eight I GHz processor Beowulf cluster running PVM (Parallel Virtual Machine) (ca. 28 Gflops) based at the Depart- ment of Organismic & Evolutionary Biology at Harvard University. The heuristic search strategy implemented consisted of 500 random iterations (1000 for the simul- taneous analyses of all the combined gene fragments under equal parameters) of the following unit: 32 inde- pendent trees were built by random addition of taxa with a subsequent round of SPR branch swapping and the best tree submitted to successive rounds of SPR and TBR branch swapping, followed by successive rounds of tree fusion (10 fusing pairs, each pair including a mini- mum of 5 taxa, and saving a maximum of 100 fused trees) and tree drifting (10 SPR followed by 10 TBR branch-swapping rearrangements keeping trees equal or better than the originals under a criterion based on character fit and tree length with 30 topological changes accepted per drift round and subsequently subjected to full SPR and TBR branch swapping accepting only minimal trees) (Goloboff, 1999). In order to speed up the computation, POY uses certain shortcuts to calculate tree length that can occasionally result in the miscalcu- lation of the exact length. To circumvent this problem. all cladograms found within 0.5% of the minimum tree length were examined, and an extra TBR branch- swapping round was applied to all cladograms found within 1% of the minimum tree length. Bremer support (BS) (Bremer, 1988, 1994) was used as a measure of clade support by implementing searches to disagree with constraints that corresponded to the clades obtained in the analysis under equal parameters. Forty-six (as many as clades) constrained searches were run locally on a PC Pentium 4 at 1.7 GHz. A less exhaustive search strategy (10 replicates of 5 iterations of random addition of taxa, otherwise the same as the general searches) was used to speed up computation, which could result in an over- estimation of the actual Bremer values. The contribution to the combined tree of each data partition was mea- sured by means of the partitioned Bremer support (PBS) (Baker and DeSalle, 1997). PBS values were obtained by inferring the length of the trees obtained in the con- strained searches for each individual data set. Optimi- zation alignment analyses were performed under different parameter cost combinations. Gap costs of 1, 2, 4, and 8 (the version of POY available at the time the analyses presented in this study were performed did not implement gap extension costs, but this option has been incorporated in more recent versions of the program) were combined with transition/transversion ratios of 1/1, 1/2, and 1/4. The robustness of clades to changes in parameter costs was assessed by the ratio of parameter combinations where an equal cost clade was observed (at least in one of the trees obtained under that partic- ular parameter combination) to the total number of combinations assayed. 2.3.3. Static alignments Static alignments, i.e., alignments with fixed homol- ogy statements, were also constructed to explore the sensitivity of the results to alternative phylogenetic treatments of gaps and alternative phylogenetic infer- ence frameworks. Static alignments for the ribosomal genes were constructed following the method of Hedin and Maddison (2001). Multiple automatic alignments corresponding to different combinations of gap opening and gap extension costs (8/2, 8/4, 20/2, 24/4, and 24/6, transition weight fixed to 0.5) were built with ClustalX (Thompson et al., 1997). Alignments for each particular set of parameters were built using a single guide tree as currently implemented in Clustal and as it is the com- mon practice in the literature. However, it is well known that automatic alignments are order-dependent and a different addition series of taxa may result in different alignments. Moreover, many equally optimal align- ments may exist for the same set of sequences (Giribet et al., 2002). The use of a single guide tree precludes the exploration of such alternative aHgnments. We chose gap opening/extension cost ratios that favored both relatively gappy (e.g., 8/2) and compressed (e.g., 24/6) 232 M. A. Arnedo el al. I Molecular Phylogene?ics and Evolution 3J (2004) 225-245 alignments (Hedin and Maddison, 2001). A particular gap opening/extension cost alignment was chosen based on topological congruence to the ehsion matrix (Wheeler et al., 1995) obtained by appending all the alignments constructed for a given gene fragment. To- pological congruence was measured by the number of nodes in common between the consensus tree of each individual matrix and the elision matrix and by calcu- lating the average symmetric-dilTercnce distances (Swofford, 2001) between all the trees from the indi- vidual matrix and the elision matrix. All the former analyses were run considering gaps as a 5th state. A combined matrix was then constructed by adding the best alignment selected for each ribosomal gene plus the COI and H3 fragments. The combined static matrix was then analyzed by considering gaps as either missing data or as 5th character-state. The same combined data ma- trix was receded using the gap coding method of Sim- mons and Ochoterena (2000) to explore the effect of coding gaps (indels) as presence/absence characters ra- ther than as a 5th state. The computing program Gap- Coder (Young and Healy, 2002) facilitates the automatic receding of an alignment using the simple indel coding version of Simmons and Ochotorena's method. Parsimony analyses of static alignments were con- ducted using PAUP* (SwofTord, 2001) and NONA (Goloboff, 1993) computer programs, and manipula- tions of the data matrices and trees were performed with MacClade (Maddison and Maddison, 2000) and WinClada (Nixon, 2002). Unless otherwise stated, all the static matrices were analyzed using a heuristic search with 100 random additions, keeping a maximum of 10 trees per iteration and an overall maximum of 1000 trees. Branch support was assessed by means of Bremer support and bootstrap proportions (Felsenstein, 1985). In PAUP*, Bremer support was implemented through the TreeRot program (Sorenson, 1999). Bootstrap pro- portions were obtained from 100 replicates of a heuristic search with 15 iterations of random addition of taxa holding 20 trees per iteration. Analyses in a parametric statistical framework were performed using Bayesian inference as implemented in the computer program MrBayes (Huelsenbeck and Ronquist, 2001). Bayesian inference was favored over standard Maximum Likeli- hood because it is less demanding computationally and because the posterior probability of the trees provide a natural and more intuitive measure of node support (hereafter referred to as posterior probability value) (Leache and Reeder, 2002; Lewis and Swofford, 2001). It should be pointed out that posterior probability values have recently been criticized as being excessively liberal when using concatenated gene sequences, and has been suggested that bootstrap probabilities are more suitable for assessing the reliability of phylogenetic trees than posterior probabilities (Suzuki et al., 2002). However, other studies have reached opposite conclusions, also based on simulations, that have led their authors to claim that Bayesian support values represent much better estimates of phylogenetic accuracy than do non- parametric bootstrap support values (Wilcox ct al., 2002). The computer program Modeltest (Posada and Crandall, 1998) was used to assess the model of evolu- tion that best fit the combined data. The parameters corresponding to the model selected were treated as unknown variables with equal a priori probability and estimated as part of the analysis based on Bayesian in- ference. Four MCMC (Markov Chain Monte Carlo) chains (one cold and three heated) were run simulta- neously on random starting trees for 1.5 x 10* genera- tions with results sampled every 100 (printed every 1000) generations. The -In of the trees was plotted against generations to determine the number of generations re- quired to achieve stability of the results and the trees obtained in generations below the stability value were discarded as "bum-in." To explore sensitivity of the results to changes in pa- rameter costs, ahgnment construction, and gap treat- ment was explored by taking as a reference the topology obtained from optimization alignment analysis under equal parameter costs (gap = transversion = transition) as a reference because we prefer this method in theory (see Section 4). We assessed similarity between analyses in two ways. Topology similarity is the ratio of the number of shared clades between the trees found in each analysis to the total number of clades obtained in the reference analysis. Second, we calculated the increase in length of the topologies found by each analysis under the assumptions of the equal costs optimization alignment method (i.e., diagnosed length of a particular tree under equal cost optimization alignment minus the minimum length of the equal cost optimization alignment analysis). The different values obtained were rescaled by dividing them by the difference between the maximum and the minimum possible lengths of the equal cost optimization alignment analysis. This index, which we call Rescaled Length Increase (RLI), is reminiscent of the Retention Index (Farris, 1989) and measures the overall decrease in fit of characters to a non-optimal topology under a particular set of assumptions. We also constructed an additional index to measure the decrease in fit of each character partition when an- alyzed simultaneously with other partitions. We call this measure the partition-based retention index (PRI), and calculate it by subtracting the minimum possible length of a data partition from the actual length of the parti- tion in the topology obtained from the analyses of the combined data set. It is even closer to Farris' Retention Index. We scaled the raw value by dividing the differ- ences between the maximum and the minimum possible lengths of that partition. In the static alignment, the M.A. Arnedo el al. I Molecular Phylogenetics and Evolution ?I (2004) 225-245 233 maximum and minimum values are the sum of the maximum and minimum steps for each character, while for the optimization alignment they are estimated from actual searches. 3. Results The gene fragments scqucnccd yielded the following lengths (primers excluded): COI 472bp, H3 328bp, 18S 779-829 bp, 28s 297-320 bp, and 16S 428-467 bp. Se- quences have been deposited in GenBank and their ac- cession numbers are listed in Table 1. Table 2 summarizes some of the results of the opti- mization alignment analyses and Table 3 gives clade support measures. Analysis of the combined data matrix under equal parameter costs resulted in 1 tree of 7283 steps, consistency index (CI) (all values reported with uninformative characters removed) = 0.29, retention in- dex (RI) = 0.42, depicted in Fig. 3. The implied align- ment yielded 2825 positions, 916 of which were parsimony informative. This tree supports theridiid monophyly, including the Hadrotarsinae genera, as do all tlie analyses with gap costs of one or two, but the outgroup structure contradicts previous hypotheses based on morphology. For the most part, outgroup re- lationships are very sensitive to changes in parameter costs, with the only exception of the sister-group relationship of cyatholipids (Alaranea merina) and thcridiosomatids {Theridiosoma geiivnosuni), which is supported in all optimization alignment analyses. A clade formed by synotaxids (Synotaxus sp.) plus nestic- ids (Nesliciis sp.) is found in 75% of the analyses but only receives a Bremer support value under equal costs of 3. If the outgroup topology is constrained to the current morphological hypothesis 4 trees result that are 58 steps longer (0.0254 increase in informative varia- tion). Outgroups and basal theridiid clades typically have well-developed coluli (black circles in Fig. 3); re- duced coluli cluster in more distal lineages (grey circles in Fig. 3), although reduction is not monophyletic. Support for the nodes concerned (42,9) is low: only 33.3% of the cost schemes explored (gap costs below 4) and Bremer support values of 7 and 11, respectively. Conversely, genera lacking a colulus and colular setae (node 12, hereafter referred as the "lost colular setae (LCS) clade," white circles in Fig. 3) are monophyletic in all analyses regardless of parameter values, and re- ceived a Bremer support of 23. Spintharinae sensu Forster et al. (1990) is not supported under any cir- cumstance. The dramatically redefined Spintharinae, including Steiivnops but excluding formerly assigned species in at least eight genera, contains taxa with re- duced coluli and is supported under most parameter combinations and by a Bremer value of 16. Genera without coluli or setae formerly assigned to Spintharinae Table 2 Statistics of the trees obtained in the different analyses performed Matrix Gap cost tv/ts Length Trees Diagnosed length RLl % Shared clades Optimization 1 7283 1 7283 0 100 1 11,464 1 7371 0.0340 47.8 I 19.492 1 7355 0.0278 54.3 2 7912 1 7318 0.0135 67.4 2 12,631 1 7357 0.0285 56.5 2 21.892 1 7363 0.0309 56.5 4 8925 5 7401.8 0.0458 47.8 4 14.573 4 7450.25 0.0645 39.1 4 25.715 3 7494 0.0814 36.9 8 10,661 4 7552.5 0.1040 21.7 8 18.081 1 7644 0.1393 26.1 8 32.595 1 7695 0.1590 21.7 Static 7 7066 3 7326.6 0.0168 60.9 1 7975 2 7396.5 0.0438 54.3 A^ 7772 1 7357 0.0285 54.3 - GTR - - 7348 0.0251 60.9 Optimization: Analyses performed with optimization alignment. Siaiic: analyses performed on the combined fragments aligned with Clustal with the following alignment parameters (in all cases transition cost set to 0.5): 18S = 8:2 (gap opening cost:gap extension cost), 28S = 8:2. 16S = 8:4 (see text for justification of the selection of these values). Gap cost: For the optimization alignment analyses refers to the gap cost used in the analyses, and for the static analyses refers to the gap treatment. A/P; Gaps recoded as absence/presence characters, tv/ts: Transversions/transition ratio. GTR: Values of the different base transformation derived from the General Time Reversible model. Trees: Numbers of trees obtained. Diagnosed length: Length of the topologies obtained from each analysis under optimization alignments with equal cost parameters. RLl: Rescaled Length Increase (see text for details). % Shared clades: Percentage of the clades supported in the optimization alignment under equal costs present in at least one of the trees obtained in each of the remaining analyses. 234 M.A. Arnedo el al. I Molecular Pliylogenetics and Evolution 31 (2004) 225-245 Table 3 Measures of support of the clades supported in the optimization alignment under equal costs Optimization alignment Static alignment Miss. 5th A/P Node NH SA BS PBS Bl COI H3 I8S 28S 16S BP BS BP BS BP BS PP 1 2 8.3 7 -6 10 2 -4 5 2 3 16.7 12 11 -3 -4 1 7 _ ? ? ~- ? ? ? 3 4 16.7 10 10 -6 3 3 0 0 2 _ ? ? 4 5 100.0 23 -4 6 18 -5 8 55 3 0 3 23 5 3 25.0 10 -7 2 9 -1 7 _ _ ? ? ? ? ? 6 4 33.3 6 -7 I 6 0 6 ? ? ? ? ? ? 7 5 75.0 3 3 3 10 -6 -7 _ _ _ ? 8 2 50.0 14 -2 10 8 -7 5 ? ? ? 73 9 3 33.3 II 12 -14 3 -6 16 _ 71 10 4 8.3 14 -3 0 13 5 -1 ? ? _ ? ? ? ? 11 5 50.0 17 -5 2 15 3 2 95 12 96 16 99 15 100 12 6 75.0 23 5 -7 13 2 10 95 11 57 1 70 10 100 13 7 66.7 14 -6 5 -1 3 13 0 I 0 1 0 2 83 14 8 33.3 18 0 -11 0 -1 30 0 I 54 2 ? _ 100 15 9 50.0 12 0 10 8 -9 3 0 1 52 I 0 2 72 16 10 41.7 14 6 -6 -2 6 10 85 13 89 II 87 13 100 17 10 41.7 II -7 2 5 4 7 71 5 79 5 76 4 99 18 9 33.3 13 0 -2 1 0 14 0 1 74 2 71 5 98 19 8 16.7 7 6 -1 1 0 1 ? ^ _ - 20 9 41.7 4 -1 0 0 2 3 _ ? ? ? ? 69 21 9 66.7 19 -5.5 3 6.5 -0.5 15.5 ? _ _ ? ? ? 92 22 10 83.3 7 -7 6 7 -7 8 0 1 0 2 ? _ 55 23 10 100.0 13 5 2 9 0 -3 94 6 99 9 98 9 100 24 7 83.3 13 -4 9 0 -2 10 _ ? _ _ ? ? 68 25 8 83.3 10 14 -6 -2 7 -3 93 7 100 12 95 14 99 26 6 50.0 20 5 -9 24 3 -3 80 9 51 I 69 8 100 27 5 8.3 12 5 8 10 -4 -7 ? ? 28 6 41.7 13 4 -1 -7 1 16 68 13 53 0 77 9 99 29 4 16.7 8 4.5 -5.5 7 2.5 -0.5 0 0 0 2 0 4 30 5 83.3 18 7 7 8 0 -4 58 6 57 2 51 5 100 31 6 100.0 57 20 -4 27 3 11 100 39 100 41 100 42 100 32 7 8.3 9 -10 15 9 _2 -3 ? ? 0 1 0 I _ 33 8 8.3 1 -3 3 3 -1 -1 0 1 0 I 34 7 58.3 27 -12 9 15 -1 16 87 7 90 8 81 9 100 35 5 25.0 18 -3 -3 19 -4.5 9.5 72 6 73 8 74 10 100 36 6 75.0 16 9 -4 -1 2 10 ? ? ? ? ? ? 100 37 7 58.3 8 -3 -2 13 -6 6 ? ? ? __ ? .? _ 38 8 33.3 10 -2 6 7 -1 0 66 4 66 7 77 11 39 8 8.3 7 -1 2 -2 5 3 53 5 57 5 66 2 ? 40 6 25.0 15 -0.5 3 13.5 5 -6 55 1 55 0 63 0 100 41 7 75.0 17 2 10 _2 -3 10 _ ? ? ? ? ? 100 42 3 33.3 7 -4 8 6 -7 4 0 2 ? ? ? ? 98 43 4 16.7 8 0 -2 -1 1 10 0 0 ? ? _ ? ? 44 5 100.0 40 -I -7 24 1 23 100 34 100 49 100 50 100 45 6 66.7 7 0 -8 -3 -1 19 0 0 82 3 78 5 ? 46 5 83.3 6 629 -8 16.5 6 46.5 3.5 301.5 5 -14.5 -0.5 279 0 2 88 8 66 4 Kode: Node number in Fig. 3. NH; Node height. SA: Sensitivity analysis support expressed as percentage of the analyses under different parameter costs (total = 12) that supported the particular clade. PBS: Partial Bremer supports of the optimization alignment under equal costs atialysis. Miss.: Gaps as missing data. 5th: Gaps as 5th state. A/P: Gaps recoded as absence/presence characters. Bl: Bayesian inference analysis. BP: Bootstrap proportions. BS: Bremer support. PP: Posterior probability. remain in the LCS clade. Hadrotarsinae is monophyletic with a Bremer support of 15, although contradicted by most analyses under differential parameter costs. The same is true for the Hadrotarsinae-Spintharinae clade (node 35, Fig. 3). Neither Theridion nor Anelosimus is monophyletic. Conversely, argyrodinc monophyly { = Argyrodes sensu Exhne and Levi, 1962, see Yoshida, 2001a) and its sister-group relationship to Enoplognathu is supported by all analyses. However, Argyrodes apart from Rhomphaea and Arianmes (sec Yoshida, 2001a) is M.A. Arnedo el al. I Molecular Phylogenetics and Evolution il (2004) 225-245 235 ARANEIDAE Argiope arg?ntala LINYPHIEDAE Linyphia triangularis PIMOIDAE Pimoa sp. 1 NESTICIDAE Nesticus sp. SYNOTAXIDAE Synotaxus sp. MYSMENIDAE Mysmena sp. TETRAGNATHIDAE Tetragnatha mandibulata 42 THERIDUDAE ? CYATHOLIPIDAE Alaranea merina - THERIDIOSOMATIDAE Theridiosotna gemmosum ' Robertus neglectiis 46 I # Pholcomma hirsutum * Anelosimus (Selkirkiella) sp. 30 29 Latrodectus mactans Crustulina sticta Steatoda bipunctata ' Enoptognatha caricis Latrodectines 34 31 32 33 Argyrodes (Neospintltarus) trigonum Rhomphaea metalissima Argyrodes (Argyrodes) argentatum W Ariamnes attenuata ? Argyrodes (Faiditus) chickeringi Argyrodines 40 35 41 36 ' Dipoena cf. hortoni ' Trigonobothrys mustelinus ' Euryopis funebris ' Thwaitesia sp. 39 37 38 Episinus angulatus Stemmops cf. servas Chrosiothes cf. jocosas Spintharas flavidas Hadrotarsines Spintharines 27 10 28 26 11 Phoroncidia sp. ? Cerocida strigosa # Styposis sells w Anelosimus (Anelositnas) eximias I . . . % Anelosimus (Kochiura) aaticus I AnelOSmuS S.S. O Theridula opulenta 24 12 25 I O Chrysso sp. "O Helvibis cf. longicauda 19 13 ^0 I O Ameridion sp. 21 "O Theridion grallator 23 I O Theridion frondeam 22 Theridion longipedatum ~0 Ragathodes sexpunctatus "O Thy moi tes unimaculatus 18 14 ~0 Keijia mneon 15 ?O Neottiura bimaculata 17 I O Nesticodes rufipes O Theridion varians 16 LCS j O Achaearanea tepidariorum O Tidarren sisyplwides Fig. 3. Single cladogram obtained from the optimization alignment of all gene fragments combined, with uniform parameter costs (gap = transi- tions = transversions = 1). Figures above branches refer to clade numbers. Tree statistics are included in Table 3. and different measures of clade support are shown in Table 2. Circles at the tips of the branches refer to the degree of developments of the colulus [according to Levi and Levi ( 1962) with additional modifications based on scanning electron microscope images, Agnarsson in prep.] in the corresponding theridiid taxon: Black circle denotes well-developed colulus, grey circle denotes colulus reduced or substituted by two setae, white circle denotes no trace of colulus or colular setae. LCS. Lost colular setae clade. 236 M.A. Arnedo ei ai I Molecular Pliylogenetics and Evolution 31 (2004) 225-245 not monophyletic. The Latrodectinae (Latrodectus, Steatoda, and Crustulina) is the only other major cladc supported by all optimization alignment analyses. The sister relationship of LCS clade (node 12) to Anelosiimis sensu strictu (i.e., all Anelosinnis species apart from the ones formerly included in Selkirkiella) form the 'lost colulus (LC) clade" (node 11) and received Bremer sup- port of 17 and occurred in half of the different parameter costs analyses. Trees resulting from the search with outgroups constrained to the topology based on current morphological knowledge, largely agree with equal cost results for the ingroup. The only difference is that Phoroncidia + Cerocidia + Styposis (clade 27 in Fig. 3) joins at node 9. When the independent gene fragments are analyzed separately, the 18S and 16S genes performed best in terms of percentage of shared clades (188 = 23.9%, 168 = 22.2%, 288 = 21.7%, H3= 14.6%, and COI = 11.1%) when compared to the simultaneous analyses, while the protein-coding genes performed the worst. However, according to the PRI the protein-coding fragments and 188, are the ones that best fit the com- bined tree for most analyses (Table 4). The 188 and 168 are also the genes that contribute the most to the total Bremer support of the simultaneous analyses, but the ribosomal 288 contribution is negative (Table 3). Par- titioned Bremer support of the different gene fragments do not show any clear relationship with time of diver- gence as measured by node depth, suggesting that they contribute information in all time windows. As parameter cost increase results diverge increas- ingly from those under different costs (Table 2). How- ever, this trend is neither proportional nor monotonie. For a constant gap cost, both RLI and percent shared clades suggest that doubling the transition/transversion ratio (is/tv) from 1 to 2 has more effect than from 2 to 4 (where similarity to the reference analysis actually im- proved under gap cost 1). For constant ts/tv ratio, trends vary: at ts/tv = 1, doubling gap costs loses about a third of shared clades each time, but at 2 and 4 doubling gap cost increases both measures but quadrupling it decreases both. In any case, results are quite sensitive to parameter choice: roughly 50%) of the reference clades are lost if any value is changed. CluslalX alignments with gap opening cost 8 and extension gap penalty of 2 were selected as the ones that best represented the elision matrix for both the 188 and the 288 gene fragments under the optimality crite- ria adopted (see 8ection 2). For the 168 alignment op- timal parameter costs depended on the criteria used. Table 4 Statistics of the partial analysis of the different gene fragments Matrix L T /.combined M. Max/. Min/. PRl Opiimizaliou COI 1736 19 1797 61 2285 402 0.0324 H3 1005 12 1082 77 1509 225 0.0600 18S 1382 1 1428 A6 ?>?}'>(, 1382 0.0545 28S 747 49 818 71 1226 747 0.1482 16S 2103 1 21S8 55 2708 2103 0.0909 Static, miss. COI 1736 227 1808 72 2285 402 0.0382 H3 1003 >1000 1077.6 74.6 1509 225 0.0581 18S 1335 >1000 1392.3 57.3 2213 609 0.0357 28S 667 18 730 63 1094 300 0.0793 16S 1949 >1000 2059 110 2779 599 0.0505 Static. Sill COI 1736 227 1822.5 86.5 2285 402 0.0459 IB 1003 >1000 1079 76 1509 225 0.0592 18S 1501 242 1616.5 115.5 2482 714 0.0653 28S 807 >1000 892 85 1247 381 0.0982 16S 2458 >1000 2565 107 3701 826 0.0372 Static. AIP COI 1736 19 18S1 115 2285 402 0.0611 H3 1005 12 1119 114 1509 225 0.0888 ISS 1501 728 1757 256 2484 704 0.1438 28S 770 54 908 138 1217 368 0.1625 16S 2304 % 2488 184 3372 m 0.0726 Matrix: gene fragment. Optimization: optimization alignments. Static, miss.: Clustal-based alignments, gaps as missing data. Static. 5ih: Clustal- based alignments, gaps as 5th state. Static, AIP: Clustal-based alignments, gaps recoded as absent/present characters. L: Tree length, T: Number of trees. L combined: Number of steps of the tree obtained in the combined analysis of all the gene fragments, for a particular gene fragment. AL: Difference in length of the combined tree and the best tree for each particular gene fragment. MaxL: Maximum possible length of a particular gene fragment. MinL: Minimum possible length of a particular gene fragment. PRI: Partition-based retention index (see text for details). M.A. Arnedo et aL I Molecular Ptiylogeiietics and Evolution 31 (2004) 225-245 237 Using percent shared clades, gap opening 8 and gap extension 4 were optimal. Using average symmetric- difference distance criterion selected gap opening 8 and gap extension 2 were optimal. Because the 8/4 alignment shared more clades than the 8/2 (34-23) and was only marginally worse under average symmetric-difference distance (15-14), we chose this alignment to merge the static data in the combined data matrix. The number of characters for each of the preferred alignments was 871, 343, and 549 for the 18S, 28S. and 16S, respectively. The static alignment of the five gene fragments combined yielded 2562 poshions. Under gaps as missing data (975 - ARANEIDAE Argiope arg?ntala 1 "THERIDIIDAE" 55 9 55 39 100 MYSMENIDAE Mysmena sp. NESTICID AE Nesticus sp. PIMOIDAE Piinoa sp. LINYPHIIDAE Linyphia triangularis TETRAGNATHIDAE Tetragnatha mandibulata CYATHOLIPIDAE Alaranea merina THERIDIOSOMATIDAE Theridiosoma genunosuin SYNOTAXIDAE Synotaxus sp. Phoroncidia sp. Enoplognallia cari?is Argyrodes (Faidilus) chickeringi Ariamnes atlenuala Argyrodes (Argyrodes) argentalum 1 I Rhompliaea metalissima 87 I Argyrodes (Neospintliarus) trigonum Roberlus negteclus Anelosimus (Setkirkiella) sp. Pholcomma hirsutum iMlrodeclus maclans Cruslulina slicla Steatoda hipunctata Spintharus ?a\ 'idus Chrosiothes cf. jocosus 5 I Episinus angutatus Slemmops cf. servus Thwaitesia sp. . I Eiiryopis funebris , Trigonobolhrys musielinus I Dipoena cf. horloni Argyrodines 34 100 Latrodectines 72 33 1 55 Spintharines + Hadrotarsines 13 68 12 95 80 11 95 Cerocida strigosa Slyposis sells Anelosimus (Anelosimus) exiinius Anelosimus (Kochiura) aulicus Theridula opulenta Anelosimus s.S. 69 93 .Chrysso sp. -Helvibis cf. longicauda . Theridion grallator 94 1 13 85 5 -Theridion tongipedalum - Theridion frondeum -Ameridion sp. . Rugalhodes sexpunclalus 1 Thymoiles uiiimaculatus -Neottiura bimacutata -Keijiamneon j Neslicodes rufipes 1 Theridion varions I Tidarren sisyphoides "I Achaearanea lepidariorum LCS 71 ' Fig. 4. Strict consensus of the 3 trees resulting of the parsimony analysis of the static combined alignment with gaps as missing data. Numbers above branches are Bremer support and below branches bootstrap proportions. Additional statistics and support showed in Tables 2 and 3. LCS, Lost colular setae clade. 238 M.A. Arnedo el al. I Molecular Pliylogenetics and Evolution 31 (2004) 225-245 informative positions) parsimony analysis resulted in three trees of length 7066, CI = 0.26, RI = 0.36 (Fig. 4). Under 5th state gap coding (1092 informative positions) the same matrix yielded 2 trees of length 7975, CI = 0.27, RI = 0.37 (Fig. 5). The Simmons and Ochotorena's simple indel coding method added 95, 71, and 238 indel absence/presence characters to 18S, 28S, and 16S gene fragments, respectively, for a total combined alignment of 2967 positions (1153 informative). Analysis of the recoded matrix resulted in 1 tree of 7772 steps, CI = 0.26, RI = 0.37 (Fig. 6). Surprisingly, "gaps as missing data" analysis is much more similar to the op- timization alignment analysis under equal costs than to other static gap-coding methods (Table 2). The PRI- measured performance of the independent gene frag- ments if analyzed separately varies drastically depending on the gap treatment (Table 4). When gaps were coded as missing data, the 18S most resembles the combined ARANEIDAE Argiope arg?ntala 2 I CYATHOLIPIDAE Alaranea merina I NESTICIDAE Nesticus sp. 2 I SYNOTAXIDAE Synotaxus sp. I LINYPHIIDAE Linyphia triangutaris PIMOIDAE Pimoa sp. 6 I MYSMENIDAEAfyimenflsp. TETRAGNATHIDAE Tetragnatha mandibulata Phoroncidia sp. THERIDIOSOMATIDAE Tlieridiosoma gemmosum ? Cerocida strigosa Styposis sells ? Enoplognatha caricis 56 55 2 "THERIDIIDAE" 57 8 41 100 90 1 73 49 100 82 ? Argyrodes (Neospintharus) trigonum - Rhomphaea metalissima ? Argyrodes {Argyrodes) argentattim 1 Ariamnes attenuata -Argyrodes (Faiditus) chickeringi Argyrodines 81 66 1 Dipoena cf. hortoni Trigonohothiys mu.^telinus Spintharus ?avidus Chrosiothes cf. jocosas Eiiiyopis funebris Thwailesia sp. 5 1 Episinus angulatus - Steinmops cf. servus 1 57 " - Latrodectus mac tans I Crustulina sticta 1 Steatoda bipunctata Spintharines + Hadrotaisines Latrodectines Robertus neglectus 8 I Anelosimus (Selkirkiella) sp. go I Pholcomma hirsutum 1 I Anelosimus (Anelosimus) e.ximius ? Anelosimus (Kochiura) aulicus ? Theridula opulenta 16 96 51 I Anelosimus s.S. 1 57 12 I Chrysso sp. YQQI Helvibis cf. longicauda Theridion grallator 99 1 - Theridion longipedatum . Theridion frondeum - Ameridion sp. . Rugathodes se.xpunctatus . Thyinoites unimaculalus 54 74 1 11 52 89 79 . Neottiuru bimaculata . Keijia inneon I Nesticodes rufipes I Theridion varians I Tidarren sisyphoides I Achaearanea tepidarionun LCS Fig. 5. Strict consensus of the 2 trees resulting of the parsimony analysis of the static combined alignment with gaps as Sth state. Numbers above branches are Bremer support and below branches bootstrap proportions. Additional statistics and support showed in Tables 2 and 3. LCS, Lost colular setae clade. M.A. Arnedo et al. I Molecular Phylogenetics and Evolution il (2004) 225-245 li') ? AR ANEID AEAr^fiop? arg?ntala ? NESTICIDAE Neslicus sp. CYATHOLIPIDAE Alaranea merina THERmiOSOMATIDAE neridiosonui gemmosum SYNOTAXIDAE Synotaxus sp. LINYPHIIDAE Unyphia triangularis Phoroncidia sp. 3 51 "THERIDIIDAE" 51 PIMOIDAE Pimoa sp. MYSMENIDAE Mysmena sp. TETRAGNATHIDAK Tetragnatha mandihulata Cerocida strigosa Styposis sells Latrodeclus mactans I Crustulina sticta I Latrodectines Steatoda bipunctata I Enoplognatha caricis 100 1 10 74 79 11 77 1 Argyrodes (Neospintharus) Irigonum Rhomphaea metalissima Argyrodes (Argyrodes) argentatum Ariamnes attenuala Argyrodes (Faiditus) chickeringi Dipoena cf. hortoni Trigonobolhrys mustelinus Spintharus ?avidus Chrosiothes cf. jocosas Tliwailesia .sp. Argyrodines 1 Euryopis funebris Episinus angulatus Stemmops cf. servas 66 Robertas neglectus 4 I Anelosimus (Selkirkiella) sp. Spintharines + Hadrotarsines 66 15 99 69 10 70 Pholcomma hirsulam Anelosimus (Anelosimus) eximias Anelosimus (Kochiura) ?ulicas Theridula opulenta I Anelosimus s.s. 14 10 81 95 ? Chrysso sp. ? Helvibis cf. longicauda ? Tlieridion grallator 98 1 LCS Theridion longipedatum The rid ?on frondeum Ameridion sp. Rugathodes sexpunctatus Thymoites unimaculatus Neottiura bimaculata Keijia mneon Nesticodes rufipes Tlieridion varions Tidarren sisyphoides Achaearcmea tepidariorum Fig. 6. Strict consensus of the single tree resulting of the parsimony analysis of the static combined alignment with gaps recoded as absence/presence characters (see text for details). Numbers above branches are Bremer support and below branches bootstrap proportions. Additional statistics and support showed in Tables 2 and 3. LCS, Lost colular setae clade. results. However, 16S performs better under 5th state coding and under absence/presence COI has the lowest PRI. In all cases, 28S performs the worst. All parsimony static analyses of the combined data set agree with the optimization alignment under equal costs in supporting the monohlyly of Latrodectinae, the Hadrotarsinae + Spintharinae + Stemmops clade (although Hadrotarsinae and Spintharinae are not monophyletic in some gap treatments explored), the Argyrodinae and Enoplognatha + Aigyrodinae, Anelosimus sensu stricto, 2'tO M. A. Arnedo et al I Molecular Pliylogenetics and Evolution 31 (2004) 225-245 and the sister relation between the latter and LCS clade. The Hadrolarsinae + Spinlharinae + Stemmops clade is sister to the Eiioplogiiarha + Argyrod'mac clade only in some trees with gaps as missing data. Unlike optimi- zation alignment analyses, parsimony analyses on the static alignments refuted theridiid monophyly, and for gaps as a 5th state or absent/present, Hadrotarsinae. Clades under static analyses that disagree with the equal cost optimization alignment tended to be poorly sup- ported (Table 3). Both the hierarchical likelihood ratio test and the Akaike information criterion preferred the GTR +1 + F (Yang, 1994) as the model for the static combined data. The Bayesian inference results for the static combined matrix is shown in Fig. 7. The only well-supported outgroup clade is Piinoa sp.?Liiiyphia triangularis. 100 50 ARANEIDAE Argiope arg?ntala NESTICIDAE Nesticus sp. PIMOIDAE Pimoa sp. LINYPHIIDAE Linyphia triangularis TETRAGNATHIDAE Tetragtiatha mandibulata 50 48 50 23 ?c 100 98 73 THERIDIIDAE MYSMENIDAE Mysmena sp. CYATHOLIPIDAE/i/aranea/Mmnfl THERIDIOSOMATIDAE Theridiosoma gemmosum SYNOTAXIDAE Synotaxus .sp. - Anelosimus {Selkirkiella) sp. Pholcomma hirsutum Rohertus neglectus 100 Slealoda hipunctata 96 I Cruslulina sticta 99 Latrodectines 71 100 46 100 Latrodeclus mactans Cerocida strigosa Styposis selis Enoplognatha caricis Argyrodes (Faiditus) chickeringi lOOi Argyrodes {Neospintharus) trigonum Rhomphaea metalissima L^ 100 100 48 100 I 100 100 98 50 changes Ariamnes attenuata Argyrodes (Argyrodes) argentatum Dipoena cf. hortoni I Trigonobothrys musleiinus I Euryopis funebris [ ~ Thwaitesia sp. Episinus angulatus Chrosiothes cf. jocosas 72 I Spintharusflavidus ' Stemmops cf. servus Phoroncidia sp. Anelosimus (Anelosimus) eximius Anelosimus (Kocliiura) aulicus Theridula opulenta 99 I Chrysso sp. Argyrodines Hadrotarsines Spintharines I Anelosimus s.S. Helvibis cf. longicauda Ameridion sp. Theridion grallator Theridion longipedatum Theridion frondeum Rugaihodes sexpunctatus Thymoites unimaculatus Neottiura biinaculata Keijia mneon Tidarren sisyphoides Achaearanea tepidariorum Nesticodes ni?pes Tlwridion varions LCS Kig. 7. The 50% majority rule consensus of the trees obtained from the Bayesian analysis (GTR +1 + F model) of the static combined alignment ( 500 first generations of the MCMC bum in). Numbers above branches are posterior probability values. Additional statistics and support showed in Tables 2 and 3. LCS, Lost colular setae clade. M.A. Arnedo et al. I Molecular Pltylogenetics and Evolution il (2004) 225-245 241 concordant with current morphological analyses. Syn- otaxus sp. is shown as the sister to all theridiids, contra morphology that places nesticids as sister to theridiids and synotaxids as sister to cyatholipids (here A. merino). The tree supports most of the ingroup clades obtained by the equal costs optimization alignment. Theridiid monophyly receives low posterior probability as do the interrela- tionship of most subfamilies, but subfamily posterior probabilities arc high. Support for Argyrodinae + Enop- lognatlia, hadrotarsines + spintharines + Stettvnops, Ane- losiimts sensu stricto, the LC clade and the LCS clade is high. At lower levels, the internal arrangements within these clades differ and Cerocida + Styposis jumps from Phoroncidia to basal theridiids. Most of the discrepan- cies receive low support, as measured by the posterior probability values. 4. Discussion 4.1. Alignments Unsurprisingly, different assumptions of alignment construction, gap treatment, or phylogenetic inference method yielded conflicting phylogenetic hypotheses (Morrison and Ellis, 1997; Wheeler, 1995). In this study, we preferred a particular inference method a priori for two reasons. First, more detailed and explicit hypotheses are more easily falsified. Second, picking a reference tree simplifies the sensitivity analysis by reducing the number of comparisons to be performed. Preference for a par- ticular method with specific assumptions should precede the actual analysis to avoid the pitfalls, and circularity, of preferring those results that better suit our precon- ceptions. We favor optimization alignment for episte- mological reasons. It is superior to static alignment because it more accurately treats indel events as trans- formations rather than as character states. Although alignment construction and tree search are arguably independent by analogy to the primary homology con- cept in morphological data (Simmons and Ochoterena, 2000), in practice static alignments usually come from automatic alignment algorithms, with or without sub- sequent manual modifications, all of which are in turn based on a guide tree (Frost et al., 2001). We preferred equal costs during optimization alignment because, ab- sent an objective criterion to choose across differential costs (Faith and Trueman, 2001), equal costs seem simpler, add less to background knowledge, and thus maximize explanatory power (Kluge, 1997a,b). Although the results of a particular analysis may be favored a priori, it is still worth exploring the effect of assumptions on these results. Molecular systematists commonly use taxonomic congruence across different methods of data analysis as a measure of clade robustness. Our varied analyses generally shared more than 50% of the clades, except optimization alignment with gap costs of 4-8 (Table 2). Static alignments that dismiss gap information (gap as missing and Bayesian inference. Table 2) are most sim- ilar to the equal cost optimization alignment results. This surprising result may relate to the way optimization alignment treats gaps. Optimization alignment mini- mizes the numbers of indel events necessary to explain the data for given gap and base transformation costs. Our analyses corroborate this point, although at first sight it may seem the opposite. Optimization alignment implied more characters than did static Clustal (2825 and 2562, respectively). However, shorter alignments do not necessarily result in most parsimonious explanations of the data. In this particular situation, optimization and static alignments are difficult to compare because parameter costs vary (gap cost 1, ts/tv 1 in the former, gap opening 8, gap extension 2 or 4, ts/tv 0.5 in Clustal). The number of informative characters is a less tricky comparison. Implied alignment yields fewer number of informative characters (916) than the static alignment, regardless of the gap treatment (975 as missing data, 1092 as 5th state). Fewer informative indels (gaps in a static alignment) probably affect the final output less, making the results more similar to the static alignment analyses when gaps are not considered. Of course, the generality of these comments must await further re- search on alternative alignment methods and in addi- tional taxa and genes. 4.2. Morphological implications and evolutionary patterns Different analyses chiefly disagree on the basal part of the tree, outgroup relationships and theridiid mono- phyly. The Theridiidae is an extremely diverse family, and instances of non-monophyly as a result of a few odd taxa incorrectly placed in the family might be expected. However, is it likely that Phoroncidia, a genus with many classic theridiid features, is not a theridiid? Phoroncidia, shares many theridiid synapomorphies, e.g., absence of a basal, ectal paracymbium, distal cymbial hook present and involved in cymbial-lock mechanism, and grossly flattened aggregated gland (AG) spigots. On the other hand, Phoroncidia lacks some typical theridiid features: male palpal tibial rim regularly and strongly hirsute and facing palpal bulb, abdominal stridulatory picks, and theridiid-type tarsal comb. It is definitely aberrant, and although morphol- ogy supports its placement within theridiids, its exact phylogenetic position remains inconclusive. Outgroup topology in this study was highly sensitive to parameter change and mostly disputes morphological evidence. No analyses support theridiid-nesticid monophyly. This sister-group relationship is based on morphology and behavior (Coddington, 1986, 1989; Forster et al., 1990; Griswold et al., 1998; Heimer and 242 M.A. Arnedo el al. I Molecular Pliylogenetics and Evolution 31 (2004J 225-245 Nentwig, 1982). The list of synapomorphies includes: exactly two coluiar setae present, reduced posterior lateral spinnerets (PLS) piriform spigot field, lobed PLS aggregate (AG) glands, cobweb, and sticky silk placed on gumfoot lines. Nesticid somatic and genital mor- phology differs considerably from theridiids. but, liny- phioids and cyatholipoids are equally divergent. The placement of nesticids in this study is unlikely to endure. Similarly, no analyses supported 'araneoid sheet-web weavers' or 'spineless femur clade' monophyly, both well supported morphologically (Griswold et al., 1998). This incongruence may be caused by several factors, includ- ing sparse outgroup taxa and use of genes that perform better at lower (species/genus) taxonomic levels. These results confirms the monophyly of Hadro- tarsinae and Argyrodinae, but dispute Argyrodes sensu Yoshida (2001a), Spintharinae sensu Forster et al. (1990) and Theridiinae sensu Yoshida (2001b). The latter two include essentially the same taxa, but our re- sults suggest that Spiniharus is not closely related to the lost coluiar clade. Spintharinae comprises mainly taxa with reduced, highly specialized, webs. None of the genera represented by more than one species is monophyletic in all analyses. Both Theridion, and Argyrodes (sensu Yoshida, 2001a) are polyphyletic regardless of the gap treatment or method of inference used. The Theridion result confirms a problem that has long been suspected (e.g.. Forster et al., 1990; Levi and Levi, 1962); Theridion is a 'waste basket' group that urgently needs revisionary work. Yoshida's (2001a) at- tempted to improve the classification of argyrodines by recognizing two very distinctive clades: Rhotnphaea and Ari?nnes. This, however, rendered the remaining Ar- gyrodes paraphyletic, which could be remedied by rec- ognizing Faiditus and Neospiiitharus as well. The five groups are highly distinct, and difier strikingly in morphology and behavior. The monophyly of Anelosi- nnis e.xiwius and A. aulicus is only contradicted by optimization alignment analyses with high gap costs (>4), while the Chilean 'Anelosimus' ( = Selkirkielld) does not cluster with the other Anelosimus under any condition. Our study suggests novel relationships among the argyrodine genera that affect interpretations of the ori- gin and evolution of kleptoparasitism and araneophagy (Fig. 8). The species Argyrodes (Faiditus) diickeriiigi nests deep within argyrodines here, suggesting that its rather generalized prey catching strategy is derived. The specialized araneophagic genera Ariamnes and Rhonip- haea are not sister here; Whitehouse et al. (2002) also suggested that their unique hunting strategy was prob- ably convergent. They also favored the hypothesis that kleptoparasitism evolved once at the base of Argyrodi- nae. Given the distribution of kleptoparasitism in Fig. 8, three optimizations are possible: gain at the argyrodine node with loss in R/tomphaea and Ariamnes, three con- I I 05 ?ft :B i J? $