1. Taxonomic sampling

Species identification was based on morphological characteristics according to the current literature and was conducted by SvN and JYR. Morphological observations were performed on 10 to 20 representatives of each species. A total of 81 sycoecine specimens representing all the known genera and 56 species, of which 50% are undescribed, were included in the molecular study (Table 2). Seventy percent of the species were represented by sequences from one specimen only. The type-species of the genera Seres (S. armipes) and Philocaenus (P. barbarus) were also included in our analyses. All necessary permits were obtained for the collection of specimens in nature reserves and national parks (Ministère des Eaux et Forêts et du Reboisement, Libreville, Gabon permit granted by Emile Mamfoumbi Kombila, Directeur de la Faune et de la Chasse; Uganda Wildlife Authority UWA/TDO/33/02; UNCST NS 214; Namibia Ministry of Environment and Tourism permit 1289/2008; Ezemvelo KZN Wildlife permits 2985/1999; 24139/2000; 29820/2002; 4502/2005; 4345/2005; 4346/2005;1958/2007; Kenyan Wildlife Service KWS/RP/5001; Cape Nature permits 288/1999; AAA004-00092-0035; AAA007-00324-0035; Northern Cape Province permits Fauna 131/2010, Fauna 132/2010; Eastern Cape Province permits CRO 101/11CR and CRO 102/11CR; Permits for field work in Borneo were obtained from the Sarawak Forestry Corporation and the Forest Research Institute Malaysia; Permits for specimen collection in China were obtained from the Chinese Natural Science Foundation (30670358, 30571507), KSCX2-YW-Z-003; Permits for the collection of specimen in Sulawesi were obtained from the Research Center for Biology LIPI (No : 3180/SU.3/KS/2007); Specimen collection in Taiwan was funded by the ANR project BioFigs/National Science Council, Taiwan, R. O. C. code: 98WFA0100291). Wasps were collected by sampling figs containing either adults or developing wasp larvae. The figs, containing wasps no more than a few days short of their emergence, were placed in handmade wasp-rearing chambers. Once emerged, adult wasps were killed and preserved in 95% ethanol. While the relationships between the chalcidoid subfamilies remain controversial, closer and more distant relatives were included as outgroups [9,10,35]. Six species belonging to the genera Haltichella (Haltichellinae, Chalcididae), Bruchophagus (Eurytominae, Eurytomidae), Grandiana (Otitesellinae, Pteromalidae), Micranisa (Otitesellinae, Pteromalidae), Walkerella (Otitesellinae, Pteromalidae), and Megastigmus (Megastigminae, Torymidae) were used (Table 2). Each time destructive extraction was used, vouchers were selected among specimens sampled from the same fig tree and the same fig after careful identification. Vouchers are deposited at CBGP, Montferrier-sur-Lez, France and Iziko South African Museum, Cape Town (SAMC). A high definition image library of vouchers was also constructed, using the EntoVision Premium Portable Imaging System, to allow future identification of specific taxa and traceability of our results (see Figure 1 for examples).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Phylogram of relationships among the Sycoecinae obtained from the analysis of the MAFFT alignment (combined dataset, without Gblocks cleaning, 6 partitions: mtDNAcodon1&2, mtDNAcodon3, EF-1α, ITS2, RpL27a, mago nashi).

Uppercase letters refer to clades discussed in the text. The new classification is indicated by colored bars on the right (yellow = oriental species, blue = afrotropical species). Nodes with likelihood bootstrap (BP) values < 70 have been collapsed. BP (> 70) and Bayesian posterior probabilities (> 0.90) are indicated at nodes. Illustrations of female habitus for the main clades are provided on the right. Host fig tree subsections are indicated between parentheses. Black boxes at nodes show cases of probable speciation on a single host Ficus species.

https://doi.org/10.1371/journal.pone.0079291.g001

2. Marker choice, DNA extraction, PCR amplification and sequencing

We selected markers that retain phylogenetic signal within deeper nodes as well as within terminal clades. We combined two mitochondrial protein-coding genes (Cytochrome oxydase I, COI and Cytochrome b, Cytb) and four nuclear genes (the F2 copy of elongation factor-1a, EF-1α; the internal transcribed spacer 2, ITS2; the ribosomal protein RpL27a and the regulatory protein mago nashi).

mtDNA loci and ITS2 have proven useful in resolving insect molecular phylogenies at shallower taxonomic levels. However, they are rapidly evolving, which make them poor markers for deep divergences [26,34]. EF-1α has been successfully used to resolve within-family relationships (e.g. [36–39]). While the exonic portions of RpL27a have proven informative for resolving deep-level relationships within the Hymenoptera [40], its intronic portions have been successfully used to reconstruct the phylogeography of parasitoids of oak galls (Pteromalidae, [41,42]. To our knowledge, mago nashi which encodes for a transcription factor that plays essential roles in Drosophila axis formation [43] has never been included in studies of insect molecular phylogeny.

Genomic fig wasp DNA was isolated using standard phenol–chloroform extraction and Qiagen DNeasy or ZyGEM extraction kits following standard protocols. Primer sequences and amplification protocols followed Cruaud et al. [44] for Cytb and COI, Cruaud et al. [45] for EF-1α , Lopez-Vaamonde et al. [46] for ITS2 and Lohse et al. [41] for RpL27a and mago nashi. PCR products were purified using ExonucleaseI and Phosphatase, and sequenced directly using the BigDyeTerminator V3.1 kit (Applied Biosystems) and an ABI3730XL sequencer at Genoscope, Evry, France. Both strands for each overlapping fragment were assembled using the sequence editing software Geneious v5.5 [47]. All the sequences were deposited in GenBank (Table S1).

3. Phylogenetic analyses

Alignments of COI, Cytb, EF-1α and mago nashi were straightforward due to a lack of length variation. ITS2 and RPL27a were aligned using various procedures, to assess the impact of alignment methods on phylogenetic inferences. ITS2 and RPL27a alignments were reconstructed with ClustalW [48] and MAFFT 6.864 [49]. Default parameters were used in ClustalW and the L-INS-i option was chosen in MAFFT [49]. Alignments were then cleaned from highly divergent blocks using the online version of Gblocks 0.91b [50] using the default settings and a less stringent selection of blocks. For the latter strategy (hereafter named « Gblocks relaxed »), the "minimum number of sequences for a flanking position" and the "minimum number of sequences for a conserved position" were set to half the number of sequences, the "minimum length of a block” was set to 5 and selection of positions with gaps present in less than half of the sequences was allowed.

Alignments of the protein coding genes were translated to amino acids using Mega 4.0.2 [51] to detect frameshift mutations and premature stop codons, which may indicate the presence of pseudogenes. Alignments of individual gene regions obtained with each strategy were first analysed separately to produce single-gene trees and then concatenated, resulting in six combined datasets (ClustalW; ClustalW + Gblocks default parameters; ClustalW + Gblocks relaxed parameters; MAFFT; MAFFT + Gblocks default parameters; MAFFT + Gblocks relaxed parameters). All combined datasets (including charsets) have been deposited on TreeBase (http://purl.org/phylo/treebase/phylows/study/TB2:S14838).

Phylogenetic trees of the six combined datasets were estimated using maximum likelihood (ML) and Bayesian methods. All analyses were conducted on a 150 cores Linux Cluster at CBGP as well as on the CIPRES Science Gateway [52]. Two alternative partitioning strategies were compared using Bayes factors (BF) [53]: Scheme 1: mitochondrial genes, EF-1α, ITS2, RpL27a and mago nashi versus Scheme 2: first + second codon positions of mitochondrial genes, third codon positions of mitochondrial genes, EF-1α, ITS2, RpL27a and mago nashi. Following Kass and Raftery [53], Pagel and Meade [54], and Schulte and de Queiroz [55], Bayes factors were calculated using the following formula: BF = 2*(ln L1 - ln L0) + (P1-P0) * ln (0.01) where ln Li and Pi are respectively, the harmonic mean of the ln likelihoods and the number of free parameters of model i. BF values from 2 to 6 were considered positive evidence, from 6 to 10 as strong evidence, and > 10 as very strong evidence favouring the alternative hypothesis over the null hypothesis. Best fitting model for each partition was identified using the Akaike information criterion [56] as implemented in MrAIC.pl 1.4.3 [57]. We performed ML analyses and associated bootstrapping using the MPI-parallelized RAxML 7.2.8-ALPHA [58]. GTRCAT approximation of models was used for ML boostrapping [58] (1000 replicates). Bootstrap percentage (BP) > 95% was considered as strong support and a BP < 70% as weak.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Phylogram of relationships among the Sycoecinae obtained from the analysis of the ClustalW alignment (combined dataset, without Gblocks cleaning, 5 partitions: mtDNA, EF-1α, ITS2, RpL27a, mago nashi).

Uppercase letters refer to clades discussed in the text. The new classification is indicated by colored bars on the right (yellow = oriental species, blue = afrotropical species). Nodes with likelihood bootstrap values < 70 have been collapsed. BP (> 70) and Bayesian posterior probabilities (> 0.90) are indicated at nodes. Illustrations of female habitus for the main clades are provided on the right. Host fig tree subsections are indicated between parentheses. Black boxes at nodes show cases of probable speciation on a single host Ficus species.

https://doi.org/10.1371/journal.pone.0079291.g002

Bayesian analyses were conducted using a parallel version of MrBayes v. 3.2.1 [59]. We assumed across-partition heterogeneity in model parameters by unlinking parameters across partitions. Parameter values for the model were initiated with default uniform priors and branch lengths were estimated using default exponential priors. To improve mixing of the cold chain and avoid it converging on local optima, we used Metropolis-coupled Markov chain Monte Carlo (MCMC), with each run including a cold chain and three incrementally heated chains. The heating parameter was set to 0.02 in order to allow swap frequencies from 20% to 70% [60]. We ran two independent runs of 20 million generations for the MAFFT-6 partitions and the MAFFT-Gblocks relaxed-6 partitions datasets and sampled values every 2000 generations. We ran two independent runs of 10 million generations for all other datasets and sampled values every 1000 generations. For the initial determination of burn-in, we examined the plot of overall model likelihood against generation number to find the point where the likelihood started to fluctuate around a constant value. The points sampled prior to convergence of the chains were then discarded. Convergence was also evaluated using the effective sample size (ESS) values of each parameters as reported in Tracer v1.5 [61]. The results were based on the pooled samples from the stationary phases of the two independent runs. Posterior probabilities (PP) > 0.95 were considered as strong support.

4. Test of alternative topologies

To compare topologies and assess whether certain alternative relationships among recovered clades could be statistically rejected, we performed AU [62] and SH [63] tests implemented in the program package CONSEL [64]. The program makermt was used to generates K=10 sets of bootstrap replicates. Each set consisted of 100,000 replicates of the row sums (10 times the default number of replicates). Default scales parameters were used (r₁=0.5, r₂=0.6, r₃=0.7, r₄=0.8, r₅=0.9, r₆=1, r₇=1.1, r₈=1.2 r₉=1.3, r₁₀=1.4), meaning that in the k-th set of replicates, N_k = r_kN sites were randomly chosen with replacement to calculate the row sums (with N=total number of sites). RAxML was used to compute the per-site log likelihoods for all trees.

5. Phylogenetic informativeness of markers

We initially compared the evolutionary properties of the markers using a Bayesian framework. For each partition of the complete combined datasets (without Gblocks cleaning), we studied base composition, substitution rates and rate variation among sites (α). We also compared rate variation among partitions, considering the parameter m (rate multiplier).

We profiled the phylogenetic informativeness (PI) of the markers, using the method described by Townsend [27] and implemented in the program PhyDesign [65]). This method uses per-site rate estimates to project the utility of a gene for resolving phylogeny across lineage history [27,66]. For each partition of a combined dataset, the phylogenetic informativeness of all sites can be summed, which provides the net phylogenetic informativeness (i.e. the degree to which the partition is predicted to contribute to resolution of the phylogeny across history). The informativeness can also be divided by the number of sites, resulting in the phylogenetic informativeness per site. We adopted this later approach in the present study as it allows a comparison of the relative power of genes without the influence of gene length, which may be considered as more accurate [67,68]. PI profiles indicate historical epochs during which a partition is most likely to provide phylogenetic signal but do not discount for the misleading effect of homoplasy caused by convergence to the same character state in divergent lineages [27,67,68]. Therefore, one must be careful when interpreting PI profiles. When reading PI profiles from tips to root, PI of the marker increases until reaching its PImax. Once the profile has crested, there are more and more sites that are evolving faster than the optimal rate, which can result in homoplasy (“phylogenetic noise”, [68,69]. Homoplasy plays a greater effect when the internodes are short. Therefore, when trying to resolve relationships, it is preferable to use character sets with PI profiles peaking deeper in time than the epoch of interest [68].

Bayesian and ML analyses of ClustalW and MAFFT alignments produced similar topologies (see results) and informativeness was calculated using the MAFFT datasets (6 partitions, without Gblocks cleaning, default-Gblocks cleaning, relaxed-Gblocks cleaning). Site rates as inferred by MrBayes (report siterates=yes command) were compiled into PhyDesign formatted files (see PhyDesign FAQ) and used as input rates. Those “vector files” are available upon request. The ultrametric tree was obtained from the MAFFT ML tree (6 partitions) using PATHd8 [70] by arbitrarily setting the root to 1.

1. Impact of alignment strategy

The final matrix contained 81 sycoecine specimens representing 56 ingroup species and six outgroups. Outgroup ITS2 sequences and Bruchophagus caucasicus, Haltichella rufipes and Megastigmus aculeatus RpL27a sequences were too divergent to be reliably aligned and were consequently excluded from the analyses. No stop codons or frame shifts were detected in the protein coding regions.

Numbers of aligned base pairs, variable sites and parsimony-informative sites for each gene region used in this study are summarized in Table 3. MAFFT and ClustalW alignments of ITS2 and RPL27a resulted in datasets with similar properties (Δ total sites _{MAFFT/ClustalW} = -2 for ITS2, 28 for RPL27a; Δ parsimony informative sites _{MAFFT/ClustalW} = -8 for ITS2, -4 for RPL27a). Removing highly divergent alignment blocks using Gblocks with default parameters dramatically reduced the number of parsimony-informative sites (about 95% loss for ITS2 and between 66% (ClustalW) and 70% (MAFFT) loss for RPL27a). This loss resulted in far less resolved ITS2 and RPL27a trees, when either the MAFFT + Gblocks default or ClustalW + Gblocks default datasets were analysed (Figures S17, S20 and see Table S3 for a comparison of tree resolutions). The Gblocks-relaxed parameters cleaning only slightly affected the resolution of the ITS2 tree and did not affect the resolution of the RPL27a tree (Table S3, Figures S18, S21).

For all partitions the best-fitting model chosen by MrAIC was GTR+I+Γ. We used a discrete gamma approximation [71] with four categories. A total of four analyses were performed per six combined datasets (2 partitioning schemes (5 versus 6 partitions) each analysed using ML and Bayesian methods, Table 4). For all bayesian analyses, after discarding 25% of the samples as burnin, ESS value of each parameter largely exceeded 200, which showed that convergence was reached. Twenty-four combined trees were obtained (Figures S1-S12) and deposited on TreeBase (http://purl.org/phylo/treebase/phylows/study/TB2:S14838). For all datasets, Bayes factor indicated that the most complex partitioning scheme (6 partitions) was preferred over the least complex one (5 partitions) (Table 4).

All trees had well-resolved backbones and three major clades (A, B, C) were identified (Figures 1-4, S1-S12). As expected the Gblocks-default topologies (Figures 3-4; S3-S4; S9-S10) showed poorer nodal supports at shallower (within C2, C4.5 clades) and intermediate nodes (within clade C). The MAFFT - 6 partitions datasets was used as input for SH and AU tests of alternative topologies (Table S2). Those tests indicated that the Gblocks-default parameters topologies were all significantly different from the others, though our visual comparison of trees revealed that conflicting nodes had poor supports. The only topological conflict between all 24 topologies with a BP > 70 was the position of Diaziella sp. nov. 12 ex Ficus sp. (n° 2989_01) (Figures 3-4, dashed line). Its position was either unresolved (5 trees), as sister to the rest of clade A2 (6 trees, 80 < BP < 92; 0.52 < PP < 1.00) or nested within clade A2 (ClustalW+Gblocks relaxed-6 partitions tree, BP=75, PP=0.89). This uncertainty was probably due to the amount of missing data for this taxon (only 3 markers on 6, Table S1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Cladograms of relationships among the Sycoecinae obtained from the ClustalW alignment of the combined dataset under three different alignment strategies and two partitioning schemes.

The corresponding ML and Bayesian trees are given in Figures 2, S1-S6. Nodes with BP support < 70% have been collapsed and BP supports for main clades are indicated at nodes. Uppercase letters refer to clades discussed in the text (see also Figure 2). The dashed line indicates the only taxon with a conflicting position among trees (see text).

https://doi.org/10.1371/journal.pone.0079291.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Cladograms of relationships among the Sycoecinae obtained from the MAFFT alignment of the combined dataset under three different alignment strategies and two partitioning schemes.

The corresponding ML and Bayesian trees are given in Figures 1, S7-S12. Nodes with BP support < 70% have been collapsed and BP supports for main clades are indicated at nodes. Uppercase letters refer to clades discussed in the text (see also Figure 1). The dashed line indicates the only taxon with a conflicting position among trees (see text).

https://doi.org/10.1371/journal.pone.0079291.g004

2. Evolutionary properties and phylogenetic informativeness (PI) of markers

Model parameter estimates for each partition of the Bayesian analyses of the MAFFT combined dataset - 6 partitions (mean and 95% credibility intervals) are reported in Table 5. Parameter estimates from the analysis of the Clustal W combined dataset (6 partitions) were similar (available upon request). As might be expected, the mtDNA partition showed extreme base compositional bias (69.7% and 96.0% of A/T for mtDNA first + second codon positions and third codon positions respectively). Among the nuclear genes analysed, while RPL27a and mago nashi were A/T-biased (73.6% and 65.4 % respectively), EF-1α and ITS2 show more or less even base composition (50.9% and 55.2% of A/T, respectively). There was a higher rate of A-G and C-T transitions for all partitions (52.6% for ITS2, 64.1% for mtDNA first + second codon positions, 68.8% for RPL27a, 80,3% for EF-1α, 84.9% for mago nashi and 88.8% for mtDNA third codon positions). The mtDNA third codon positions showed a high rate of C-T transitions relative to any other transformations (57.0%). For the nuclear genes, the rates of transitions were close and so were the rates of transversions. The ITS2 rate matrix was the least skewed towards one type of change over another. The gamma shape parameter α was higher for the nuclear genes than for the mtDNA partitions, indicating that nuclear genes show less rate heterogeneity among sites than mitochondrial genes (Table 5). The rate multiplier parameters (m) indicated that rates of substitution were different among partitions. The mtDNA third codon positions evolved about twenty-three times faster than the fastest nuclear gene (ITS2). EF-1α was the most slowly evolving marker.

For the complete dataset (MAFFT without Gblocks cleaning - 6 partitions), mtDNA (first+second and third codon positions) showed sharp peaks of informativeness at shallower nodes (Figure 5). Then, the profiles declined showing potential for homoplasy, which is not surprising when considering mtDNA evolutionary properties (Table 5). This prediction is confirmed by the observation of the real performance of the mtDNA marker (i.e. the single-gene tree) in resolving shallower relationships (Figure S13). Interestingly, PI profile suggested that ITS2 was informative across the whole phylogeny. Again, this prediction was confirmed by the observation of the ITS2 tree (Figure S16), on which main clades were recovered. Overall, shapes of the PI profiles correlated well with the corresponding single-gene tree topologies (Figures S13-S16, S19) and, more precisely, with the distribution of branch lengths across the topologies. The mtDNA topology had longer terminal than internal branches. The deeper in the tree, the shorter the branches, though enough information remained to resolve, at least partially, deeper relationships. Terminal branches of the ITS2 tree were much shorter than those of the mtDNA tree, but longer than those of the RPL27a tree. The deeper branches of the ITS2 tree were longer than those of the RPL27a tree. Neither EF-1α nor mago nashi, the slowest evolving markers contained enough information to resolve relationships within species complexes and sometimes even between species (Figures S14, S15). Our qualitative comparison of single gene tree topologies would lead us to the following ranking in informativeness at shallower nodes : mtDNA > ITS2 > RPL27a > mago nashi ^~ EF-1α, which corresponded to the relative position of the PI profile curves. At deeper nodes, this comparison would lead us to rank the markers as follows: ITS2 (longest branches) > RPL27a > mtDNA ^~ mago nashi ^~ EF-1α, which is again compatible with the PI profiles on Figure 5. mago nashi PI profile curve lays a bit above EF-1α PI profile curve probably because some longer branches do appear in the mago nashi topology (Figure S15).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 5. Per site phylogenetic informativeness profiles of the markers based on the MAFFT dataset (6 partitions).

Uppercase letters refer to clades discussed in the text (see also Figures 1 & 2).

https://doi.org/10.1371/journal.pone.0079291.g005

PI profiles were similar, when a relaxed Gblocks cleaning was performed (Figure S22), and, as expected, the informativeness of ITS2 and RPL27a sharply decreased, when default parameters were used to clean the alignment (Figure S23). A bit surprisingly, the default-Gblocks mtDNA PI score was the highest across the whole phylogeny, though RPL27a showed an equivalent resolution at intermediate nodes (Figures S13, S20).

3. Sycoecinae relationships

As we were interested in relationships at both lower and higher levels, we focussed on trees from the ML and Bayesian analyses of the complete datasets (ie without Gblocks cleaning). We deliberately kept what could be considered the “worst” tree (ie ClustalW + 5 partitions, Figure 2, S1) and the “best” tree (ie MAFFT + 6 partitions, Figures 1, S8) as no one was preferred by the SH/AU tests of alternative topologies. We also wanted to over-emphasize that the few topological differences only occurred between nodes that were either poorly supported in both trees or poorly supported in at least one of the two trees. In the following section, BP and PP values are mentioned as followed: BP ClustalW-tree/BP MAFFT-tree; PP ClustalW-tree/PP MAFFT-tree or are summarized by unique values when no differences between the trees were observed.

The genus Robertsia never clustered with other Sycoecinae species. However, a monophyletic Sycoecinae could not be rejected by both SH and AU tests (Table S2).

All the remaining sycoecine genera formed a strongly supported clade (BP = 99/96, PP = 1.00). The phylogenies supported the monophyly of the genera Crossogaster (BP = 100, PP = 1.00), Diazella (BP = 100, PP = 1.00) and Sycoecus (BP = 100, PP = 1.00) but recovered neither Philocaenus nor Seres as monophyletic. Both AU and SH tests rejected a monophyletic Seres (Table S2).

The Sycoecinae (genus Robertsia excepted) clustered into three major clades: Diaziella species (clade A) appeared sister to a clade grouping all Afrotropical Sycoecinae (BP = 99/98, PP = 1.00/1.00), which clustered into two clades: clade B (Crossogaster spp.) and clade C (Philocaenus, Sycoecus and Seres spp., BP = 100, PP = 1.00).

Within clade A, Diaziella yangi, D. bizarrea and D. sp. ex F. lawesii (clade A1, BP = 100/99, PP = 1.00), were sister to a clade (clade A2, BP = 99/100, PP = 1.00) that grouped all other Diaziella species.

Within clade B, Crossogaster sp. nov. 5 + C. michaloudi (clade B1; BP = 100, PP = 1.00) were sister to the remaining Crossogaster species (clade B2; BP = 100, PP = 1.00). Interestingly, Crossogaster sp. nov. 6 ex F. louisii grouped with the C. odorans clade (BP = 99/100, PP = 1.00).

Within clade C, Seres species hosted by Ficus polita and F. bubu (clade C1, BP = 100, PP = 1.00) were recovered sister to a strongly supported clade (BP = 90/71, PP = 0.99/0.97) including 1) Sycoecus species (clade C2, BP = 100, PP = 1.00), 2) an undescribed Philocaenus species (“clade” C3) and, 3) a strongly supported clade (C4, BP = 95/99, PP = 1.00) clustering Philocaenus species and all the remaining Seres species (type species of the genus Seres included). Within clade C4, four groups of Philocaenus species could be distinguished (C4.1 to C4.4). P. barbarus could be a complex of at least three species. Furthermore, P. liodontus and P. rotundus did not form mutually exclusive clades.

Finally, our analyses recovered several cases, where most closely related species were sampled from the same host tree (black boxes in Figures 1 & 2), which could indicate sympatric speciation (Genera Diaziella, Crossogaster, Sycoecus and Philocaenus) .

1. Impact of alignment and partitioning strategies, accuracy of phylogenetic informativeness profiles

As underlined by Rokas et al. [26], one should be cautious when making generalizations about the taxonomic level at which a given marker might be useful. Genera belonging to different insect families may span a range of evolutionary ages, and hence evolutionary properties and phylogenetic informativeness (PI) of markers should be used as guidelines only [69]. Despite the potential problems while applying these properties across different families, we believe that our results may be useful for the designing of future phylogenetic studies between and within chalcidoid families, especially within the pteromalid complex [9]. It as been shown that MAFFT outperformed ClustalW on several datasets (e.g. [49]), however, this was not the case for our dataset. We also highlighted that complex partitioning schemes, even favoured by Bayes Factors did not lead to different topologies. Then, between what could be considered the “worst” tree (ie ClustalW + 5 partitions, Figures 2, S1) and the “best” tree (ie MAFFT + 6 partitions, Figures 1, S8) only few topological differences were identified and all of them occurred between nodes that were either poorly supported in both trees or poorly supported in at least one of the two trees. We also showed that using Gblocks with default settings on our alignments dramatically decreased the number of parsimony informative sites (Table 3) and tree resolution. This indicates that one must be cautious when using Gblocks to clean alignments. Gene properties should be studied before Gblocks refinement and selection of blocks should be used only when focussing on relationships at high level, especially when an important number of markers are concatenated [72].

Overall, shapes of the phylogenetic informativeness (PI) profiles (Figure 5) correlated well with the distribution of branch lengths on the corresponding single-gene tree topologies (Figures S13-S21). It is noteworthy that ITS2 contained information spanning the whole tree and mtDNA did contain signal to contribute resolving shallower but also deeper nodes of the phylogeny. It was surprising that the mtDNA PI score was still the highest of any gene at the root of our phylogeny when we profiled the informativeness of each partition of the MAFFT-Gblocks default dataset (Figure S23). Klopstein et al. [67] suggested that PI could be biased towards fast rates (due to the whole set-up of the measure). This could explain the relative position of the curves on the MAFFT-Gblocks default PI profiles. Indeed, for this dataset, mean values of m (rate multiplier) scores for the two partitions: mtDNA-third codon (mtDNAnt3) and mtDNA-first + second codon (mtDNAnt1&2) positions were the highest of any partitions (Table 6). Nevertheless, the mtDNA PI score for the MAFFT dataset was not the highest of any gene at the root of our phylogeny (Figure 5), though the rate multiplier of the mtDNAnt3 partition was the highest of any other partitions (Table 5). Moreover, the mean values of m scores for the ITS2 and RPL27a partitions were higher than the mean value of m score of the mtDNAnt1&2 partition, though the mtDNAnt1&2 PI score was higher than ITS2 and RPL27a PI scores at shallower and intermediate nodes (Figure 5).

2. Monophyly of the sycoecine genera and new classification

3. Evolutionary implications of the relationships among the Sycoecinae

As mentioned above, all trees have well-resolved backbones, which indicates stability of higher-level relationships to different alignment strategies (Figures 1-4, S1-S12). Based on this robust phylogenetic hypothesis of relationships among the Sycoecinae, we give preliminary results regarding the evolutionary history of the group.

3.3. Comparison with available Ficus host phylogenies.

Although current taxonomy of the Sycoecinae supports a degree of host specificity between the wasps and their Ficus hosts, our study reveals many examples of break-down in host-specificity. For instance, we frequently observe the association of more than one sycoecine species (often belonging to several genera) with one Ficus species (e.g. Crossogaster and Seres species Figs. 1 & 2). Furthermore, a single sycoecine species can be associated with more than one Ficus species (e.g. S. medius, S. liodontus). Such complex associations imply that events such as host-switches and speciation on host are common incidents in the evolutionary history of these independent lineages.

Ostiolar morphology prevents entry into the fig cavity for wasps that are not specifically adapted [14,83,84]. Externaly ovipositing fig wasps do not have to conform to the morphological adaptations required to enter the fig cavity through a host-specific ostiole. Therefore, internally ovipositing wasps are thought to be highly host specific and less likely to experience host shifts than the externally ovipositing wasps [1,5,46,85]. However, these ideas are still contentious. Recent studies have shown that some externally ovipositing fig wasps may be highly host specific [5,86–88] and could have cospeciated with their host figs [86,87] but see 89,90. Overall, we lack extensive data on non-pollinating fig wasps host specificity and further cophylogenetic studies on representative samplings of both figs and wasps are required. Our results strongly suggest that strict cospeciation has not shaped the evolutionary history of both sycoecines and their host Ficus (Figure 6). It appears that the constraints of internal oviposition may not be enough to prevent host-switching events. However, further studies are needed to uncover the fine evolutionary history of the partners. These should especially focus attention on species that are associated with numerous fig hosts, the majority of which fall within Ficus subsection Chlamydodorae.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 6. Compared phylogenies of the Afrotropical sycoecine fig wasps (this study) and their Ficus hosts

(adapted from Rønsted et al. [94] and Renoult et al. [95]) .

https://doi.org/10.1371/journal.pone.0079291.g006