Taxa included in the analysis are 169 species of salticids and representatives of four dionychan families as outgroups (Table 1, Suppl. material 1). Based on previous phylogenetic work (Maddison and Hedin 2003, Maddison et al. 2008, Bodner and Maddison 2012, Zhang and Maddison 2013, in press), about 70 species of salticids from the major clade Salticoida were selected because they would represent most known major lineages, and because several genes are available for each (Table 1, Suppl. material 1). In addition, a few salticoids were added because their placement was unclear: Agorius, Diplocanthopoda, Echeclus, Gedea, Nungia, Phaulostylus, and Synagelides.

Our sample targeted especially the non-salticoid salticids, those that lie outside the major clade of familiar salticids (Maddison and Needham 2006). We included most available data from non-salticoid salticids, both new data and data previously published by Su et al. (2007) and others (Maddison and Hedin 2003, Maddison and Needham 2006, Maddison et al. 2007, Bodner and Maddison 2012, Ruiz and Maddison 2012, Zhang and Maddison 2013, Maddison and Piascik, in press). Included for the first time in a molecular phylogeny are the cocalodines (Wanless 1982, Maddison 2009), which are Australasian non-salticoid salticids. Also analyzed for the first time are the lyssomanine genera Chinoscopus and Pandisus, the lapsiine Lapsias, and the spartaeines Brettus, Meleon, Sparbambus, and Taraxella.

Some previously-published data from non-salticoid salticids was either excluded or represented under a different species name here. Excluded are sequences of Hispo cf. frenata, because its limited data made it unstable in the analyses (see Maddison and Piascik 2014), “Portia labiata” from Su et al. (2007), because its identification is in doubt and no voucher specimen is available, and the actin 5C sequence of Tomomingi sp. voucher d243, which we discovered to have been a contaminant from the euophryine Ilargus. The species labeled as Phaeacius yixin by Su et al. (2007) is included here as “Phaeacius sp. [Hainan]”, because the specimen was a juvenile female and thus identified with doubt; by its DNA we suspect it is Phaeacius lancearius. The specimen labeled as Mintonia ramipalpis by Su et al. (2007) is actually a female Mintonia silvicola. This misidentification arose because of an error in male-female matching by Wanless (1984), whose female “Mintonia ramipalpis” is actually the female of Mintonia silvicola. The correct match of male and female Mintonia silvicola is evident by intimate co-collecting in a recent expedition to Sarawak (Maddison and Piascik, unpublished) and in DNA sequence comparison. We have therefore blended data from Su et al.’s female with that from our males to represent Mintonia silvicola.

Some of the species studied appear to be undescribed, or are doubtfully the same as described species. Following the usual convention, the names of some of our specimens includes “cf.” to indicate that they may be the same as the mentioned species, “aff.” to indicate that they are close to, but distinctly different from, the mentioned species. Figures 1–13 give illustrations of some of the undescribed species, in order to facilitate future association of our data with a species name. The species we refer to as “cf. Phaeacius [Sarawak]” (Figs 1, 2) is known from a single female and juvenile from Lambir Hills, Sarawak. It resembles Phaeacius but the legs are shorter, and the epigynum is distinctively different. Phaeacius sp. [Sarawak] (Figs 3, 4) is a fairly typical Phaeacius whose epigynum resembles that of Phaeacius leytensis Wijesinghe, 1991, but with the atria elongated posteriorly. Onomastus sp. [Guangxi] is shown in Fig. 5. Sonoita aff. lightfooti (Fig. 6) has longer grooves for the openings of the epigynum than Sonoita lightfooti, and is distinctive in gene sequences as well. Gelotia sp. [Guangxi] (Fig. 7) has a palp resembling Gelotia syringopalpis, but the tibial apophyses are much shorter. Echeclus sp. [Selangor] (Figs 8, 9) was identified as an Echeclus by the distinctive form of the palp tibia, and the embolus hidden behind a ledge of the tegulum, through which several dark sclerites can be seen (Prószyński 1987). It might equally well have been identified, by the same features, as a Curubis species (Zabka 1988). Indeed, the two genera are likely synonyms. “Echeclus” is used as that is the older name. Taraxella sp. [Johor] (Figs 10, 11) and Taraxella sp. [Pahang] (Figs 12, 13) are typical species of Taraxella. The specimen MRB024 identified as Cotinusa sp. is the same as that named “unidentified thiodinine” by Bodner and Maddison (2012). The Hypaeus specimen (d130) was formerly identified as Acragas sp. (Bodner and Maddison 2012). The specimen d105 labeled as “Nannenus lyriger” by Maddison et al. (2008) is not Nannenus lyriger, but another apparently undescribed species of Nannenus. The data for Cheliceroides longipalpis comes from two specimens, d222 which is clearly Cheliceroides longipalpis, and d415 which may be a different but very closely related species. Notes on the undescribed hisponines are given by Maddison and Piascik (2014), whose data we use.

Specimens whose voucher ID’s (Table 1, Suppl. material 1) are of the form S###, d###, MRB###, or JXZ###, SWK12-####, or ECU11-####, where # is a digit, are deposited in the Spencer Entomological Collection of the Beaty Biodiversity Museum, University of British Columbia. The remaining vouchers are in the Lee Kong Chian Natural History Museum (formerly Raffles Museum for Biodiversity Research or RMBR), National University of Singapore.

In addition to analyses done on all 176 sampled taxa (“Complete”), subsets of taxa were analyzed alone. A first subset (“Salticoida”) of 78 taxa highlighted the Salticoida, with just 7 non-salticoid outgroup taxa (4 hisponines, 1 spartaeine, 1 cocalodine, 1 lapsiine), in order to obtain an alignment that was less perturbed by highly divergent non-salticoids. A second subset highlighted the non-salticoids (“Non-salticoid”, 120 taxa), to obtain an alignment primarily for non-salticoid salticids, and also to be able to explore their relationships in more detail.

Eight genes were used for this analysis. Two are nuclear ribosomal genes, 28s and 18s (Maddison and Hedin 2003, Maddison et al. 2008). Four are nuclear protein coding genes: actin 5C (Vink et al. 2008, Bodner and Maddison 2012), wingless (Blackledge et al. 2009), myosin heavy chain (“myosin HC”, Blackledge and Hayashi, unpublished), and histone 3 (Su et al. 2007). Two mitochondrial regions were also used, CO1 and another region including 16s and NADH1 ("16sND1", Hedin and Maddison 2001, Maddison and Hedin 2003). Following Bodner and Maddison (2012), the intron region of actin 5C was deleted from the analyses as it is highly variable and difficult to align.

The sequencing protocols for wingless and myosin HC are described below. For other genes, sequences marked “^S” in Table 1 and Suppl. material 1 were obtained by the protocols of Su et al. (2007), all others by the protocols of Bodner and Maddison (2012) and Zhang and Maddison (2013).

For most wingless sequences, the forward and reverse primers used were respectively Spwgf1 and Spwgr1 (Blackledge et al. 2009). PCR amplification included a 2 min 94 °C denaturation and 35 cycles of 30 s at 94 °C, a 30 s annealing step at 52–57 °C, 30 s at 72 °C and one 3 min extension step at 72 °C. For some specimens this did not succeed in amplifying wingless, and for those we used a nested protocol starting with outer primers wg550F and wgABRz (Wild and Maddison 2008). The resulting product was then amplified using two internal primers, forward Wnt8MBf1 5’-TGTGCTACTCARACKTGYTGG-3’ and reverse Wnt8MBr3 5’-ACAAWGTTCTGCA ACTCATRCG-3’. For both the external and internal reactions amplification was done with 2 min 94 °C denaturation and 37 cycles of 20 s at 94 °C, a 20 s annealing step at 52 °C (wg550F/wgABRz) or 56 °C (wnt8MBf1/wnt8MBr3), and 2 min at 72 °C, and no final extension. The nested protocol obtained sequences for Bavia aff. aericeps (voucher d389), Hasarius adansoni (d295), Philodromus sp. (GR011), Simaetha sp. (d027), and Yllenus arenarius (JXZ173). In other specimens, the nested protocol often resulted in amplification of a different member of the wingless family (e.g. WNT-8), but these were readily detected (and excluded) by BLASTing them to other genes in the NCBI database (http://www.ncbi.nlm.nih.gov).

The region of myosin HC sequenced corresponds mostly to an intron. Primers used are (forward) Myhc1f 5'-ACAACAATTCTTCAACCATCAC-3' and (reverse) Myhc5r 5'-CTTCCTCAAGGATGGACA-3' (Blackledge and Hayashi, unpublished). PCR amplification included a 2 min 95 °C denaturation and 35 cycles of 20–45 s at 95 °C, a 45 s annealing step at 52 °C, 1 min at 72 °C and one 10 min extension step at 72 °C. The boundary between the exon and intron was determined by aligning the salticid implicit amino acid translations against the known transcript for myosin HC in Cyrtophora citricola (Genbank accession AAM97635.1; Ruiz-Trillo et al. 2002).

Two small single-nucleotide errors in the sequences were corrected after the analyses but before submission to Genbank. These are near the ends of CO1 of MRB199 (Gelotia sp. [Guangxi]) and MRB231 (Eburneana sp. [Gabon]). Given that CO1 had little resolution, these are unlikely to have affected the results.

Automatic multiple sequence alignment was performed by MAFFT (Katoh et al. 2002, 2005), run via the align package of Mesquite (prerelease of version 3, Maddison and Maddison 2014), aided by Mesquite for manual corrections and for alignment by amino acid. Coding regions were easily aligned by hand according to amino acid translations. This was done starting with an initial automated nucleotide alignment, followed by hand correction in Mesquite using the Color Nucleotide By Amino Acid function to reveal amino acid translation. Non-coding regions (28s, 18s, noncoding region of 16sND1, myosin HC intron) were aligned by MAFFT using the L-INS-i option (--localpair --maxiterate 1000). Mesquite was used to color the matrix via the option ‘‘Highlight Apparently Slightly Misaligned Regions’’ so as to identify regions that needed correction. These were almost always near the ends of sequences.

Alignment was done separately on the Complete, Non-salticoid and Salticoida datasets. Following the MAFFT alignment, the Salticoida dataset required 5 small realignments by hand in 18s. The first 60 positions in the initial alignment of 16s were also realigned locally, and in addition 8 minor shifts by one or two positions were made by hand. The Non-salticoid dataset required three simple hand fixes in 28s. The first 24 positions of 16s in the initial alignment were realigned by MAFFT in isolation because of several obvious misalignments. The Complete dataset appeared poorly aligned in 28s from sites 375 to 489 in the initial alignment, which were therefore realigned by MAFFT in isolation. The first 60 positions in the initial alignment of 16s were also realigned locally, and in addition 8 minor shifts by one or two positions were made by hand. Five small shifts were performed by hand for 18s. Many analyses were done with different variants of the alignments as this study was progressing, and the phylogenetic trees remained substantially consistent.

Phylogenetic analyses using maximum likelihood were run using RAxML version 7.2.8alpha (Stamatakis 2006a, 2006b). The protein coding genes and 16sND1 were each divided into partitions. Protein coding regions were divided into one partition for 1st and 2nd codon positions, and another partition for third codon positions. Introns and non-coding regions were treated as separate partitions. For the fused 8 gene analyses, there were 7 partitions total: (1) 1st + 2nd codon positions in nuclear genes, (2) 3rd codon position nuclear, (3) nuclear intron, (4) nuclear ribosomal, (5) 1st + 2nd codon positions mitochondrial, (6) 3rd codon position mitochondrial, (7) noncoding mitochondrial. Each partition was permitted to have its own model parameters.

Analyses were done for each gene region separately with the Complete taxon set. In addition, analyses fusing all 8 genes were done for the Non-salticoid and Salticoida taxon sets. For all of these, RAxML runs assuming the GTRCAT model were used with 100 search replicates, to seek maximum likelihood trees. In addition, likelihood bootstrap analysis was performed with 500-1500 bootstrap replicates (as indicated in the figures), each involving a single search replicate. Phylogenetic analyses using GARLI version 1.0.699 (Zwickl 2006) under the model GTR+gamma+I were also done but are not reported; they resulted in substantially similar trees.

Our results help resolve or add strength to relationships at the deepest level of salticid phylogeny. Wanless (1980) recognized three major subdivisions of lyssomanines: (1) the New World genera Lyssomanes and Chinoscopus, (2) the Asian Onomastus, and (3) the remaining Old World genera including Asemonea. He suggested these three groups are so distinct that they may not belong together. The molecular data agree: the three groups’ divisions are so deep that their relationships have not yet been recovered, and it is possible, even likely, that they do not form a monophyletic group. Different analyses give different results of the relationships of these three, with some showing the New World genera as sister to the spartaeine-lapsiine-cocalodine clade (as recovered by Su et al. 2007), other results showing Onomastus in that role, and others showing the three lyssomanine groups together.

Spartaeines, lapsiines and cocalodines form a clade (node 1, Fig. 14). Although Rodrigo and Jackson (1992) concluded that spartaeines, Holcolaetis and the Cocalodes group form a clade (they were unaware of lapsiines), our analysis provides the first support for such an arrangement – their analysis included only a single taxon outside the group, and therefore it could not speak to the monophyly of the group. Our new result is intuitively appealing, as it groups together all of the extant medium-sized generalized non-salticoids/non-hisponines that are typically brown or gray. However, these presumably are or could be plesiomorphic traits; there had been no obvious reason to expect the spartaeines, lapsiines and cocalodines should have fallen together. There is no known morphological synapomorphy of this clade.

Within this spartaeine-lapsiine-cocalodine clade, the subclade historically best known by morphology is Wanless’s (1984) narrow version of the Spartaeinae, delimited by the presence of a tegular furrow (Wanless 1984). The Spartaeinae sensu stricto is primarily Afro-Eurasian, with a few Australasian species. Outside of this clade, there are no clear morphological synapomorphies defining subclades, and yet there is a striking geographical pattern: all of the Neotropical species belong to a clade, thus forming the lapsiines, while all of the Australasian species belong to a clade, thus forming the cocalodines. It is unsatisfying that we lack morphological synapomorphies for the lapsiines or cocalodines. The data suggest that the lapsiines and cocalodines are sister groups, with spartaeines more distant (Fig. 14).

Our results continue to support the relationship of hisponines with the Salticoida (node 2, Fig. 14; Figs 15–17; Maddison and Needham 2006, Bodner and Maddison 2012).

The placement of Eupoa remains unclear. As noted under Results, the 28s and 18s genes of Eupoa may be unreliable phylogenetically, although Maddison et al. 2007 found those genes to place Eupoa among non-salticoid salticids. In our results Eupoa likewise has no clear placement, except for being outside the clade of Salticoida + Hisponinae. This result appears in the Non-salticoid and Complete datasets, and with the separate analyses of wingless, CO1, and 16sND1.

Our results strongly support the monophyly of the Spartaeinae sensu Su et al. (2007), placing Holcolaetis and Sonoita together with the Spartaeinae in the narrow sense. This is concordant with Wanless’s (1985) hypothesis that Holcolaetis and Sonoita formed a clade with the spartaeines to the exclusion of Cocalodes. The analyses of Su et al. (2007) did not sample Sparbambus, Taraxella, Brettus or Meleon, but otherwise their results were largely concordant with ours, which are: (1) Phaeacius (with Sparbambus) diverge deep, (2) Yaginumanus is sister to Spartaeus, (3) Gelotia, Neobrettus, Brettus and Meleon are monophyletic, (4) Paracyrba and Cyrba are sisters, (5) Portia is sister to Cyrba and Paracyrba. There is strong support for Gelotia through Cyrba as a monophyletic group, and for their relationship with Cocalus. By our data the exact placements of Taraxella and Mintonia are unclear.

A few spartaeine taxa in our analyses were problematical in appearing unstable, having different placements by different analyses. One of these is Spartaeus spinimanus, for which we have only 16sND1 and CO1 data, both gene regions that appear to evolve too quickly for reliable phylogenetic placement at this level (Bodner and Maddison 2012, Zhang and Maddison 2013). The other is “Spartaeus” uplandicus, whose 28s sequence appears strongly divergent from others. This sequence is from Maddison and Hedin (2003, as “unidentified spartaeine”, vouchers 185 and 186), and it groups “Spartaeus” uplandicus with one species of Holcolaetis, against the placements by morphology, CO1 and 16sND1. There is a chance that this gene was mis-sequenced in “Spartaeus” uplandicus. Because of the instability generated, we excluded Spartaeus spinimanus and “Spartaeus” uplandicus from our analyses giving bootstrap results.

Because of the concordance of our phylogenetic results with those of Su et al. (2007), our phylogeny continues to support their conclusions on the stepwise evolution of a complex predatory strategy in spartaeines.

The Salticoida’s basal divergence places the primarily-Neotropical Amycoida as sister group to an unnamed clade (node 3, Fig. 14) that contains most of salticid diversity. This surprising result, first discovered by Maddison and Hedin (2003), had very strong support in the analyses of Bodner and Maddison (2012). We here add support from two new genes, wingless and myosin HC, both of which independently resolve both the Amycoida and its sister group as monophyletic.

There have been hints of a clade uniting the Marpissoida, Astioida and baviines (Bodner and Maddison 2012). In our analyses the clade does not receive bootstrap support above 50% in the Complete or Salticoida analyses. The maximum likelihood trees either show the three as monophyletic or not, depending on taxon inclusion and details of the analysis (e.g., Figs 15 and 18). At present we must conclude the relationship between these three and the Saltafresia is unresolved.

The astioids as delimited by Maddison et al. (2008) continue to be resolved as a clade, with new support from myosin HC and wingless (Figs 18, 22, 23). Although the body form of Nungia resembles that of baviines and the marpissoid Metacyrba, our data clearly place it as an astioid.

Bodner and Maddison (2012) proposed a clade, the Saltafresia, containing salticoids other than amycoids, astioids, baviines and marpissoids. They found this clade reasonably well supported – 0.78 likelihood bootstrap and 1.0 posterior probability – but no single gene supported it on its own. Our data here continue to support it when all genes are combined. Two genes support it separately, with the exception of single taxa: 28s (but Tisaniba is included) and wingless (but Phintella is excluded).

Previous work had established Habrocestum and Chinattus as close relatives of Hasarius (Maddison et al. 2008). We here add several more genera to the group, all Asian. These are Gedea, Echeclus and Diplocanthopoda. The relationships among these genera are not clearly resolved except for a well-supported relationship between Hasarius and Echeclus (Figs 14, 19).

The relationship between Salticus and the Philaeus group proposed by Maddison et al. (2008) receives additional support from wingless, along with previously-demonstrated support from 28s and actin. With high posterior probabilities (Bodner and Maddison 2008) and reasonable likelihood bootstrap values (Figs 15, 19), and supported by different genes independently (Figs 20, 22, 24), this relationship can now be considered sufficiently secure that we here formally place the genera of the Philaeus group into a subfamily — the Salticinae. In addition to genera previously analyzed (Salticus, Philaeus, Carrhotus, Tusitala, Mogrus, and Pignus) the subfamily also includes Phaulostylus, which is related to Tusitala (Fig. 14).

A set of four major groups (plexippoids, aelurilloids, leptorchestines and the Salticinae) form a clade in our analyses (node 5, Fig. 14). This group is resolved in the All Genes analyses with high bootstrap values, and it appears, almost, in the independent analyses of each of three genes (18s, wingless, myosin HC). We say “almost” because three of the genes have one or two taxa missing from or added to the group (Fig. 14). While we believe the evidence is good that these form a clade, there is a possibility that the Euophryinae might also fall nested within it. For instance, in the analyses of Bodner and Maddison (2012) the euophryines were placed as sister to the plexippoids. In our analyses the Euophryinae is placed as sister to the Plexippoida + Aelurilloida + Leptorchesteae + Salticinae.

This major clade is almost entirely Afro-Eurasian, with the plexippoid Habronattus being the only exception with more than a handful of species (others are Pellenes, Sibianor, Evarcha, Phlegra, Paramarpissa and Salticus, each with fewer than 15 described New World species).

The 14 euophryine taxa in the analyses are resolved strongly as a monophyletic group. This is a stronger test of monophyly than that of Zhang and Maddison (2013), because it includes additional genes and more non-euophryine taxa. The All Genes analyses, along with wingless and myosin HC individually, suggest that the euophryines are the sister group to node 5 (Fig. 14).

Morphologically, the antlike agoriines Agorius and Synagelides are puzzling, with strangely contorted legs and unusual genitalia (Szüts 2003, Logunov and Hereward 2006, Prószyński 2009). While they appear to be salticoids, morphology has given little guidance as to their placement. As noted already, their 28s and 18s genes appear anomalous, and give no clear indication as to their relationships. In the All Genes analysis their placement is ambiguous, though they appear to be salticoids. In an attempt to determine their placement, an additional analysis was done, using a dataset that included Agorius constrictus and a chimera of Synagelides cf. lushanensis and Synagelides cf. palpalis (to have a single Synagelides taxon with three genes). The aberrant nuclear ribosomal genes of agoriines were excluded from the analysis. The other taxa included were the 70 taxa having at least 4 genes other than CO1 and histone 3. A RAxML likelihood analyses placed Agorius and Synagelides within the sister group of the Amycoida (node 3, Figure 14) with high support (bootstrap percentage 88), but exactly where was highly unstable. Among the 100 likelihood non-bootstrap search replicates were 7 different placements: sister to leptorchestines, sister to baviines, sister to node 5 in Figure 14, sister to the Saltafresia, sister to astioids + marpissoids + baviines, sister to node 3, or sister to node 3 without the baviines. While a relationship with the leptorchestines is appealing, as it would allow their antlike body forms to be homologous, the best we can say at present is that agoriines likely belong within the sister group of amycoids (node 3).

Most of the genera for which we have multiple species – e.g., Asemonea, Portia, Mintonia, Phaeacius, Cyrba – are inferred to be monophyletic in our analyses, corroborating existing concepts based on morphology. The clearest exception is Tabuina, in which Tabuina rufa and the similar Tabuina aff. rufa fall apart from the type species Tabuina varirata, which had been anticipated as a possibility by Maddison (2009). Lyssomanes, Galianora, and Gelotia are reconstructed as paraphyletic, but in each case the bootstrap values are low.

The placement of cf. Phaeacius [Sarawak] as sister to Phaeacius, with strong molecular divergence from the other species, would justify establishing a new genus for it.

Previous work (Maddison and Hedin 2003, Bodner and Maddison 2012) has suggested that 28s and actin 5C are phylogenetically informative to a reasonable degree for deeper salticid phylogeny, insofar as their results are concordant with summed genes analyses, morphological resemblances, and biogeographical patterns. 16sND1 is useful at the shallower levels (Hedin and Maddison 2001) but has difficulties recovering deeper relationships, while CO1 struggles through both shallow and deep levels (Maddison and Hedin 2003, Bodner and Maddison 2012).

One surprise in our analyses was the informative behaviour of CO1 in deeper relationships among the non-salticoid salticids. Although CO1 is almost nonsensical in its inferred relationships within the Salticoida, it succeeds in recovering the Spartaeinae, the Spartaeineae sensu Wanless, the lapsiines, and the Salticoida as monophyletic.

Two new genes added, wingless, myosin HC, both show clear concordance with the 28s and previous all genes analyses. Wingless supports many of the previously recognized clades, including the Salticoida, Amycoida, the sister clade to Amycoida, Plexippoida, Marpissoida (in part), Astioida (in part), Spartaeinae sensu Wanless, and lapsiines. We find it encouraging that a haphazardly chosen protein-coding gene, independent from 28s, supports previous molecular results in Salticidae. There are still, however, many aspects of salticid relationships yet to be resolved, such as the deepest relationships in the family, including the relationships among the three subgroups of lyssomanines, the placement of Eupoa and the agoriines, and the relationships among astioids, marpissoids, baviines and the Saltafresia. With the coming era of genomic data, we expect large quantities of new data will be available for exploring these relationships.