Technical Comments

Response to Comment on "Hexapod Origins: Monophyletic or Paraphyletic?"

Science  12 Sep 2003:
Vol. 301, Issue 5639, pp. 1482
DOI: 10.1126/science.1087632

Assessing the relationships among arthropod taxa is an intensely debated issue in metazoan phylogeny, with various studies testing different character sets, phylogenetic methods, and strategies for analyzing molecular data (13). Delsuc et al. (4) criticize our recent hypothesis that Hexapoda is not monophyletic (3) on methodological grounds, including use of a suboptimal substitution matrix, lack of correction for among-site rate variation (ASRV), and biased taxon selection. They further introduce a method (5) that recodes nucleotide sequences into only two categories— purines (R) and pyrimidines (Y)—and use a maximum-likelihood approach to reanalyze our data recoded in this fashion. They claim that this method corrects for artefactual clustering of taxa that results from compositional bias of nucleotide content; this view is bolstered by the correct placement of the honeybee and louse, which were anomalously placed in our tree and others (1, 6, 7), within Insecta. Their reanalysis places collembolans at the base of Hexapoda, although with moderate to low support, and thus questions the main conclusion of our study.

Despite the correct placement of the honeybee and louse, it is not clear that the Delsuc et al. method performs better generally. In fact, one might expect that reducing nucleotide sequence data set to only two states (R and Y) might exacerbate saturation at many sites, and be more susceptible to problems of nonstationarity of substitutions than an amino acid data set (8). In the latter, the model of substitution reflects the probability of a replacement to be fixed, and the use of 20-state characters decreases the possibility of overlooking convergence. The inference of phylogenies based on amino acid sequences is common practice and is generally accepted as among the most reliable of methods (810). Although the matrix of amino acid change used in the first of our two analyses was based on vertebrate sequences, it has been extensively used to study relationships among invertebrates with no reports of significant flaws (1, 6).

In our previous study (3), we used an alignment of amino acid sequences and two likelihood-based methods of analysis: A classical likelihood reconstruction using a fixed-parameter model of amino acid substitution, and a Bayesian analysis based on a general time reversible (GTR) model of substitution and modeling rate variation across sites using an invariant + Γ distribution. Both analyses converged on the same topology, which supports the placement of Collembola outside of the Insecta + Crustacea assemblage.

The correction for rate variation among sites has been shown to potentially affect estimates of branch lengths and divergence times, but it seems to have only a limited effect on topology (11). To further address this issue, we calculated the likelihood of competing trees under the same matrix, but modeled ASRV using a Γ distribution (12). Table 1 shows that the trees we produce (with Collembola outside Insecta + Crustacea) give higher likelihood scores, although with different degrees of significance, regardless of the use of Γ correction.

Table 1.

Comparison of tree topologies under the amino acid substitution model mtREV24 implemented in PAML (12, 19) and incorporating a Γ-correction for ASRV [pKH = P value of the Kishino-Hasegawa test (17); pSH = P value of the Shimodaira-Hasegawa test (18); pRELL = P value of the RELL bootstrap (17)]. In the 15-taxon data set, the topology derived in Nardi et al. (3) is compared with a topology derived in Delsuc et al. (4), after pruning extra taxa and exchanging Anopheles gambiae with A. quadrimaculatus. In the 25-taxon data set, the topology from Fig. 1 is compared with the topology derived in (4). In the 35-taxon data set, the two topologies derived in (3) and (4) are compared.

Tree Likelihood alpha (of Γ) pKH (View inline) pSH (View inline) pRELL (View inline)
15 taxa Nardi et al. -18488.559 0.42838 1.000 1.000 0.914
Delsuc et al. -18507.904 0.42228 0.087 0.090 0.086
25 taxa Nardi et al. -25482.424 0.41111 1.000 1.000 0.883
Delsuc et al -25492.299 0.41063 0.121 0.119 0.117
35 taxa Nardi et al. -34838.359 0.47124 1.000 1.000 0.920
Delsuc et al -34881.787 0.45337 0.077 0.076 0.080

To investigate the possible effects of taxon exclusion on the analysis, we repeated the analysis described in (3) on the 25-taxon data set of Delsuc et al. (4). To rule out the possibility that the analysis is negatively affected by use of a suboptimal substitution matrix and lack of ASRV correction, we used the Bayesian method outlined in (3) (Fig. 1). Again, Collembola fall outside the Insecta + Crustacea clade, although with only moderate support.

Fig. 1.

Maximum likelihood tree obtained applying the method outlined in (3) as implemented in MrBayes ver. 2.1 (20) (aamodel = gtr; rates = invgamma) to the 25-taxon data set of Delsuc et al. (4). The analysis was run for 570,000 generations and sampled every 100 generations. The first 150,000 generations were excluded from the analysis as the burn-in of the search. Numbers at each node indicate posterior probabilities. Branch lengths are drawn according to estimates obtained with PAML.

Our past (3) and present analyses, the analysis of Delsuc et al. (4), as well as other molecular studies (13) demonstrate that a reliable reconstruction of the phylogeny of Arthropoda—and the assessment of the mono- or paraphyly of Hexapoda, specifically—are still disputable. Results differ when subjecting the same data set to different methods of analysis or when using different subsets of data with the same methods. This leaves the impression that none of the competing hypotheses can yet be rejected with certainty. However, we believe that the theory of hexapod nonmonophyly proposed by several studies (3, 7, 14, 15) must be considered. In this context, the recent discovery of a marine hexapod from the Lower Devonian (16) undermines the traditional association between terrestrialization and the evolution of hexapods, leaving room for alternative hypotheses concerning hexapod origins.

References and Notes

Related Content

Navigate This Article