INTRODUCTION

Schmitz and Moritz (1998: 183) published alignments and analyses of two molecular datasets which, they claimed, “provide strong evidence that sociality has independently evolved twice in the Vespidae.”

The data consisted of sequences from the 16S mt-rDNA and 28S rDNA loci for the following sample of wasps of the family Vespidae: three Vespinae (Vespa crabro, Provespa nocturna, and Vespula germanica), three Polistinae (Belonogaster petiolata and two species of Polistes), three Stenogastrinae (Liostenogaster vechti, Eustenogaster fraterna, and Parischnogaster mellyi), four Eumeninae (two species of Ancistrocerus and a different species of the genus Eumenes for each sequence dataset). Vespinae and Polistinae are all highly social species, while species of Stenogastrinae are primitively social and those of Eumeninae are mostly solitary (for a review of behavior, see Ross and Matthews, 1991). The datasets also included two bees of the family Apidae (both of them species of Apis) and a different parasitoid outgroup for each sequence dataset (one species of Pteromalidae and one of Braconidae).

For the 16S dataset, there were 169 informative characters out of 314 aligned base pairs (Schmitz and Moritz [1998: 187] excluded 62 base pairs from their analyses because of “poor alignments”); for the 28S dataset, 125 out of 331 aligned base pairs were informative.

Schmitz and Moritz's (1998) analyses included the usual (misguided) distance and likelihood procedures commonly used in molecular systematics, as well as invalid statistics. I will not repeat criticisms of such methods, which have been abundantly detailed in the cladistic literature (Farris, 1981, 1983, 1985, 1999; Farris in Werdelin, 1989; Carpenter, 1990, 1992a, 1992b, 1996; Kluge and Wolf, 1993; Farris et al., 1996; Siddall, 1998; Siddall and Whiting, 1999). However, their parsimony cladograms indeed supported a closer relationship of the eumenine sample to the polistines + vespines than the stenogastrines, thus diphyly of social wasps, as figures 1 and 2 show, and this was also supported by analysis of the two alignments combined. This result is in conflict with the cladistic analyses of vespid subfamilies by Carpenter (1981, 1988) and Carpenter and Rasnitsyn (1990), which supported Stenogastrinae as the sister group of Polistinae + Vespinae, with Eumeninae in turn the sister group to that clade, and thus monophyly of social wasps. This latter arrangement is also in line with traditional taxonomic treatment of the group (Richards, 1962). The disparity in results is obviously significant for the evolutionary study of social behavior in wasps: Whether social behavior evolved once or twice independently is crucial for the interpretation of studies of behavioral features in primitively social versus highly social wasps.

Schmitz and Moritz (1998: 183) cited Carpenter (1981) to the effect that the morphological characters grouping Stenogastrinae with Polistinae + Vespinae “may be prone to homoplasy”, and indeed I have pointed out that the morphological evidence is not abundant and that the behavioral evidence most convincingly supports the grouping (see Carpenter, 1988). The behavioral characters relate mostly to social behavior, but of this Schmitz and Moritz (1998: 184) stated:

The claim of greater plasticity rendering behavioral characters less informative is actually without foundation (Wenzel, 1992), and behavioral characters do not show elevated levels of homoplasy relative to morphological characters (De Queiroz and Wimberger, 1993). The critical test of whether the social behavior characters in these wasps are informative or not would be to include them in a simultaneous analysis together with the morphological and molecular characters (Nixon and Carpenter, 1996). A related fallacy in the quoted statement is the equation of homoplasy and lack of phylogenetic informativeness. But homoplasy in and of itself does not prevent a character from being informative—a homoplastic character may still be informative in a particular clade, a fact which is part of the foundation of numerical cladistics (the “Wagner method” of Farris [1970], which permits evolutionary reversibility) and phylogenetic analysis itself (see Farris, 1983). As Wenzel (1997: 31) put it, “Eliminating characters because they are expected to show high homoplasy is an unacceptable ad hoc protection of an hypothesis from a legitimate test.”

Figures 1 and 2 also show the family Vespidae as paraphyletic, in terms of the bee genus Apis. Vespids and apids are placed in different superfamilies and are not at all closely related (see Brothers and Carpenter, 1993; Brothers, 1999). The separation of these two families is supported by abundant morphological data. Schmitz and Moritz (1998: 189) termed their arrangement “unusual”, and stated, “To clarify the exact position of the Stenogastrinae among the aculeate hymenoptera, a more extensive study, including a range of additional vespid and nonvespid members of the Vespoidea, is required.” Nevertheless, Schmitz and Moritz (1998: 190) concluded that their data provide “strong evidence for the sister group relationship of Eumeninae to Polistinae + Vespinae.”

SIMULTANEOUS ANALYSIS

Whether the evidence of Schmitz and Moritiz is “strong” should really be assessed through simultaneous analysis of the molecular data combined with morphological and behavioral characters (see Nixon and Carpenter, 1996, for review). I now present such an analysis. I adduced the relevant morphological and behavioral data from Carpenter (1981, on vespid subfamilies; 1987, on vespine genera; 1988, on stenogastrine genera; 1989b, on social behavior; 1991, on polistine genera), Brothers and Carpenter (1993, for aculeate family relationships), Ronquist et al. (1999, for superfamily relationships), and Hunt (1999, on vespid subfamilies), for a total of 125 additional variables. These variables are listed in appendix 1 and in a matrix scored for the sequenced taxa at the end of that appendix.

The matrix was produced by: (1) taking each cited character matrix, reducing each to the pertinent family, subfamily, or generic vectors using the program Winclada (Nixon, 2002) through deletion of irrelevant terminals, (2) flagging the uninformative characters by use of the “mop uninformative chars” command of Winclada, and then deleting those characters; and (3) combining the pertinent scores as summary scores with the sequenced exemplars by merging the matrices (see Nixon and Carpenter [1996] on mechanics and terminology). None of the 16S data were excluded.

Analysis with the program Nona (Goloboff, 1999) results in a single cladogram for the 16S alignment combined with morphology and behavior (fig. 3) and two cladograms for the 28S alignment combined with morphology and behavior (fig. 4 is the consensus tree); the minimum-length tree was found by most searches. The results of the simultaneous analysis do not indicate a reclassification of Aculeata, nor reinterpretation of vespid phylogenetic relationships. Vespidae are supported as a monophyletic family, and social wasps are supported as monophyletic; that is, Stenogastrinae are the sister group of Polistinae + Vespinae. Simultaneous analysis of both alignments combined with the morphology and behavior leads to the same result (fig. 5). This is in accord with the results published by Carpenter (1981, 1988), and it is thus seen that the data of Schmitz and Moritz (1998) are scarcely “strong”. When other characters are considered, the sequence alignments are overruled.

In support of their conclusions, Schmitz and Moritz (1998: 189) observed that rearranging their tree for both alignments to support monophyly of Vespidae required 31 additional steps. Rearranging the cladogram of figure 5 to support paraphyly of Vespidae, with Stenogastrinae excluded, requires 49 additional steps. Schmitz and Moritz also cited bootstrap values as confirming their tree, 100% for the Eumeninae + (Polistinae + Vespinae) and placement of Apidae as sister group to this clade in the tree for both alignments (the values are, respectively, 98% and 58% if 16S sites are not excluded). Use of bootstrap values for assessing confidence is misplaced (see Farris in Werdelin, 1989; Carpenter, 1992a, 1996; Kluge and Wolf, 1993), but in the present context it is perhaps worth pointing out that the monophyly of Vespidae is likewise supported by 100% bootstrap values (1000 replicates) for each combination of alignment with morphological and behavioral characters (figs. 3–5). Bremer support values, on the other hand, show a discrepancy: as calculated with Nona, for the combination of the two alignments, paraphyly of Vespidae is supported by just 1 step, while for the combination of molecular with morphological and behavioral characters, monophyly of Vespidae is supported by 36 steps.

REALIGNMENT

The lack of stability of the results supported by the sequence alignments could be due to a number of factors, but one obvious possibility is the alignments themselves. Schmitz and Moritz (1998: 186) produced their alignments “by using the CLUSTAL V program … which were improved by comparison of the secondary structure of the rRNAs.” Unfortunately, criticism of any alignment raises a difficult issue: in general, there is no optimality criterion for alignment. Thus, the common practice of inputting sequences into the Clustal program, obtaining an output alignment, and then changing that alignment by the user is acceptable, despite the fact that the changes are made on an arbitrary basis. Of course, authors may state that they are taking into account secondary structure or applying a weighting scheme for multiple substitution, and so on, but replicability of such procedures is, to state the obvious, not straightforward. To this is added the common practice of discarding data on the grounds that they are “difficult to align”. Even if a program is employed, with defined cost parameters that the user does not override ad libitum, those parameters are purely arbitrary. This has been cogently discussed by Wheeler (1995), who proposed to deal with the problem by means of sensitivity analysis, wherein the cost parameters are varied and multiple reanalyses undertaken in order to assess how sensitive the output alignment is to particular parameters. He also proposed to select among parameter schemes by means of congruence, with the costs chosen based on minimization of incongruence among datasets. Thus he appealed to an optimality criterion external to the alignment procedure to select the alignment.

Application of an alignment optimality criterion would obviate the necessity for sensitivity analysis, and in the realignment that follows I propose to use parsimony, specifically in the sense of preferring an alignment that implies fewer steps than another. As with the use of parsimony in phylogenetic analysis (Farris, 1983), parsimonious alignment would maximize similarities accounted for by ancestry (inheritance) by minimizing separate origins of features (= nucleic acid residues). Obviously, for this criterion, gaps must be factored so as to avoid trivial invariant alignments, as could be obtained by treating gaps as “missing” (i.e., no cost) and merely inserting them so that the alignment implied no steps. Treatment of gaps as a “fifth state” would obviate that problem, and then a bound could be set on changes. For example, if the most parsimonious alignment is the one that has the fewest changes, set as lower bound the length of the longest sequence (= maximum number of steps on a bush). Then the difference between maximum steps minus the longest sequence could be minimized.

However, Schmitz and Moritz (1998) treated gaps as missing, and so for this realignment will I. Rather than beginning with the raw sequences, I began with their published alignment, which can be “improved” if the gaps can be rearranged so that the same number of aligned variables is maintained, but fewer steps are implied. This can be readily accomplished. Schmitz and Moritz's alignment for 16S of 314 variables implies a maximum number of steps of 802 and a minimum of 267, with the difference being 535; their 28S alignment of 331 variables implies a maximum of 600 steps and a minimum of 231, with a difference of 369. These numbers were improved using Winclada, using the Alignment Insert/Delete function. The procedure is illustrated in figures 6–9. Figure 6 shows a screen display of a portion of the 28S alignment published by Schmitz and Moritz. In Figure 7, the display of character statistics for this alignment has been toggled on, which shows at the left top, in order: the total information content (= difference between maximum and minimum steps), the maximum number of steps, the minimum number of steps, the character weights (all = 1), the character activities (all active), and character additivites (all nonadditive). The same numbers are shown for each character above that character's column. I merely inserted or deleted gaps, checking the step totals each time and saving changes when these numbers went down. This is shown for one of the characters in figure 8: gaps have been inserted in front of the cytosines in character 45, moving them to character 51. This results in a reduction in maximum number of steps to 596, an improvement of 4, with decreases as well in other step counts (note that the values for weights and additivities now appear higher; this is because Winclada adds gaps at the end of the matrix when the Alignment Insert/Delete function is toggled on; at the end of the procedure the trailing gaps are stripped). Figure 9 shows the subsequent step, where gaps were inserted in the place of the adenine in character 51, moving it to character 50, and deleted in front of the thymines in character 50, moving them to character 49. This reduces the maximum number of steps to 594.

Improvement of the alignments following this procedure was rapid in each case, up to a reduction of around 100 maximum steps, after which it became more difficult, and so was halted. This indicates a potential limitation of the method, similar to the well-known problem with local optima during tree searching: beginning the realignment at particular regions may limit the ability to attain a global optimum. As the procedure is manual, there is no guarantee that a global optimum has in fact been attained (but note that if there is a sufficient number of terminals that a global optimum cannot be guaranteed with multiple alignment programs). However, it is clear that the realignments produced are a substantial improvement over those published by Schmitz and Moritz, and they are sufficient to illustrate a critical point. The realignments are given in appendix 2. For the 16S data, the resulting alignment of 314 variables implies a maximum number of steps of 713 and a minimum of 256, with the difference being 457. Analysis of the alignment results in a single cladogram, with topology identical to that of figure 3 and length of 463, consistency index of 0.55, and retention index of 0.54. For the 28S data, the resulting alignment of 331 variables implies a maximum of 499 steps and a minimum of 209, with a difference of 290. Analysis of the alignment results in a single cladogram, with topology identical to that of figure 4, except that relationships among Vespinae are resolved as Vespula + (Provespa + Vespa) with a length of 260, consistency index of 0.80, and retention index of 0.82. Combination of the two realignments then results in a single cladogram, identical to that of figure 5, with a length of 723, consistency index of 0.64, and retention index of 0.65.

The sequence alignments of Schmitz and Moritz (1998) are thus not robust, and improved alignments support Vespidae as a monophyletic family and social wasps as monophyletic, that is, Stenogastrinae are the sister group of Polistinae + Vespinae.

CONCLUSIONS

Note that the results shown in figures 3–5 are not completely in accord with other studies even now; specifically, generic relationships depicted within Stenogastrinae and Vespinae do not match those in Carpenter (1987, 1988) and Carpenter and Starr (2000). That deserves reexamination, but rather than undertake that with these data, I note that Schmitz and Moritz (2000) have more recently published a paper alluding to sequence data for some additional vespid taxa, including two species of the subfamily Masarinae. Through a more judicious selection of outgroups (e.g., using just Apis), some of their analyses no longer resulted in vespid paraphyly (a result that however resurfaced when the parasitic wasps were included as outgroups), but still showed Eumeninae as more closely related to Polistinae + Vespinae than Stenogastrinae (which in yet another highly novel arrangement on one of their trees are less closely related than the Masarinae, which are sister group to Eumeninae). I do not anticipate that these results would withstand simultaneous analysis any better, or realignment. However, I cannot investigate this at present. Despite Schmitz and Moritz's citation of Genbank accession numbers for these sequences, not all of them are in fact in that database, including both Masarinae.