Study site and sampling

Sarawak Forest Department provided permission to collect ant samples. Ants were collected from 98 trees representing 10 Macaranga species from sections Pachystemon and Pruinosae at LHNP, Sarawak, Malaysia (4°2'N, 113°50' E, 150–200 m a.s.l.) from 1999 to 2008. Ten workers from each colony were preserved in 99.5% ethanol and constitute one sample. Voucher specimens were deposited at the Faculty of Science, Shinshu University, Matsumoto, Japan. For the mtDNA phylogenetic analysis, we used two species from GenBank as outgroups: (1) a phytoecious Decacrema species from Sulawesi that inhabits stem domatia in Neonauclea; and (2) a Crematogaster species (C. cf sp. SKY10) not in the subgenus Decacrema that inhabits M. winkleri, a myrmecophytic species not closely related to sections Pachystemon and Pruinosae. The two ant taxa are suitable as outgroups for the Decacrema inhabitants of Macaranga, because the Decacrema group inhabiting Neonauclea are the sister group of those inhabiting Macaranga (Quek et al. 2004) and because C. cf sp. SKY10 inhabiting M. winkleri belongs to the subgenus Crematogaster which is taxonomically distant taxa from Decacrema (Quek et al. 2004). Collection locations of the specimens and their GenBank accession numbers are listed in S1 Table.

SSR genotyping

From each ethanol-preserved ant colony sample, DNA was extracted from the whole body of single individual with a DNeasy Blood & Tissue Kit (Qiagen, Hilden, Germany) following the manufacturer’s protocols. Five microsatellite loci (Ca5, Ca12, Ca15, Ca18, and Ca19) developed by Feldhaar et al. [36] were amplified in 98 samples. A multiplex polymerase chain reaction (PCR) analysis was performed with a Type-it PCR master Kit (Qiagen, Hilden, Germany). The PCR temperature profile was 95°C for 10 min, then 25 cycles of 95°C for 30 s, 52°C for 30 s, and 72°C for 1 min, and final extension at 72°C for 60 min. The amplified product was run on an ABI 3130 Genetic Analyzer (ABI, Weiterstadt, Germany), sized relative to Genescan LIZ-500, and genotyped with the GeneMapper* version 4.0 program (ABI, Weiterstadt, Germany).

STRUCTURE clustering

To detect the nuclear genetic structure from SSR genotype data, we performed a Bayesian model-based clustering analysis with STRUCTURE version 2.3.3 software [37,38]. We ran the admixture model of STRUCTURE for 20 iterations with values of K (where K is the true number of clusters) from 1 to 10. Each run consisted of a Markov Chain Monte Carlo function performed for 100,000 generations after 100,000 burn-in generations. To detect the most likely value of K in the data, we estimated the log likelihood of the data, lnP(D), for each value of K across all 20 runs of STRUCTURE and examined an ad hoc quantity, ΔK, based on the second order rate of change of the likelihood function with respect to K [39].

MtDNA phylogenetic analysis

The mitochondrial COI gene was amplified by PCR with Takara ExTaq* (Takara Bio, Shiga, Japan) using the primers CI-13 (5'-ATA ATT TTT TTT ATA GTT ATA CC-3') and CI-14 (5'-GT TTC TTT TTT TCC TCTT TC-3') [14]. The PCR temperature profile was 35 cycles of 94°C for 30 s, 42°C for 30 s, and 72°C for 90 s. After amplification, the PCR product was purified with ExoSap-IT* (USB, Cleveland, Ohio, USA). Cycle sequencing reactions for both strands were performed with a BigDye* Terminator version 1.1 Cycle Sequencing Kit (ABI, Weiterstadt, Germany) on an ABI 3130 Genetic Analyzer.

COI sequences were edited and aligned with the SeqScape* version 2.5 program (ABI, Weiterstadt, Germany). Base-frequency homogeneity was tested by a χ2 test in the Kakusan4 program [40]. The χ2 test did not reject the hypothesis of homogeneity of nucleotide frequencies in each pair of taxa (P > 0.50). The degree of substitution saturation in the third codon position of the COI sequences was assessed by plotting the transition and transversion rates against genetic distance for each data set with the DAMBE version 5 program by the method of Xia and Xie [41] (S1 Fig.). In the saturation plot analysis, we used the simple JC69 substitution model [42] because DAMBE does not support the J2 substitution model, although in the model selections described in the next paragraph, we used the Jobb (J) 2 substitution model [43]. Substitution saturation in the third codon position was not detected (P < 0.001). Additionally, we did not find evidence of any mitochondrial pseudogenes, that is, nuclear mitochondrial transfers (numts), in the mitochondrial COI sequences, which can lead to an erroneous phylogeny [44]. The COI sequences did not have any indels (small insertions or deletions) or stop codons. As a result, we used all codon positions for the phylogenetic analyses.

We selected the best-fit substitution model for each codon position by using Bayesian information criterion 5 (BIC5) in the Kakusan4 software package [40]. As a result, we selected the following models: J2ef + G for the first codon position; J1 + G for the second codon position; and J2 for the third codon position. We performed a maximum likelihood (ML) analysis with TREEFINDER version October 2008 software [43] and the models selected by Kakusan4. Clade support was assessed by 1000 bootstrap replications in TREEFINDER. In addition, Bayesian posterior probability and maximum parsimony (MP) bootstrap support were obtained with MrBayes version 3.1.2 software [45] and PAUP* 4b10 software [46], respectively. The models selected by BIC5 in Kakusan4 were used in the Bayesian analysis: General Time-Reversible (GTR) [47] for the first codon position; Hasegawa-Kishino-Yano (HKY) 85 [48] for the second codon position; and GTR + G for the third codon position. The Bayesian analysis was run for 1,000,000 generations, with sampling every 100 generations. We assessed the log-likelihood of each sampling point against generation time to identify when the Markov chains reached a stationary distribution, and then discarded the initial 2001 trees as burn-in. The parsimony bootstrap support was assessed with 1000 bootstrap replicates by using heuristic searches with tree bisection and reconnection and 100 random addition replicates for each.

Host preferences

The Macaranga myrmecophytes were divided into three groups according to their stem texture and taxonomic group. Section Pachystemon, which has naturally hollow stems and includes 21 myrmecophytic species, was divided into waxy Pachystemon, which has waxy crystals on its stem surface, and smooth Pachystemon, which has smooth stems that do not secrete wax [21,49]. Section Pruinosae, the third group, included five myrmecophytic species with waxy plugged stems that are hollowed out by their ants [21,49].

The preference of each ant genotype for a plant group or species was determined by a one-way χ2 test of the extent of departure of the observed proportion of ants of the genotype using a plant species or species group (the frequency with which ants of the genotype used a plant group or species) from the expected proportion (the plant group or species frequency at LHNP), and 5000 Monte Carlo simulations were also performed to assess the reliability of the χ2 test result. The number of samples varied among the plant groups (waxy Pachystemon, n = 40; smooth Pachystemon, n = 46; Pruinosae, n = 12), plant species (from M. sp. A, n = 1 to M. bancana and M. beccariana, n = 15 each), and ant genotypes (from MS2, n = 6, to MS5, n = 29). These biases reflect the natural frequencies of the plant groups and species and the ant genotypes at LHNP, because we carried out random sampling without regard to the distributions of the plants and ants.

SSR genotyping analysis

At least three microsatellite loci were amplified for each ant sample. Non-amplified loci were scored as absent (0) data. STRUCTURE likelihood results were compared with lnP(D) and ΔK. The highest mean lnP(D) was observed for K = 5 (lnP(D) = -1944.40 ± 26.73 [mean ± SD], ΔK = 1.71), whereas the highest K was observed for K = 3 (lnP(D) = -2021.77 ± 2.53 [mean ± SD], ΔK = 73.96). In addition to these two estimates of admixture proportions (K = 3 and 5), we also conducted the genotyping analysis with the value of K = 6 which is consistent with the number of mtDNA clades (see next paragraph). In spite of relatively low value of mean lnP(D) under K = 6 (lnP(D) = -1964.65.74 ± 43.66 [mean ± SD]), the SSR genotyping clearly divided the ants into 6 groups (Fig. 2). This low value of lnP(D) may be owing to the lack of sampling size of MS4 and MS6 (n = 10 and n = 8, respectively). These three estimates were compatible and the K = 6 clustering contained elements of both the clustering of K = 3 and K = 5 (Fig. 2). Therefore, we decided to adopt K = 6 as the most suitable clustering. In addition, some samples appeared to be admixtures of two genetic clusters, whereas each value of K was associated with a definite genetic cluster (see Figs. 2 & 3).

Download:

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

Fig 2. Admixture proportions based on STRUCTURE clustering of five microsatellite loci.

Analysis of 98 samples (bars) of Decacrema ants inhabiting Macaranga in Lambir Hills National Park yielded six genetic clusters (MS1–MS6), which are color coded here for K = 3, 5, and 6. Triangles indicate samples that appear to be admixtures of two genetic clusters, suggesting hybridization.

https://doi.org/10.1371/journal.pone.0116602.g002

Download:

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

Fig 3. Comparison between the mtDNA phylogeny (A) and nrDNA SSR genotyping (B).

The maximum likelihood (ML) phylogeny of 98 samples of Decacrema ants inhabiting Macaranga in Lambir Hills National Park was inferred from a 569-bp DNA sequence of the mitochondrial gene cytochrome oxidase I. The numbers above the branches are the ML bootstrap support/Bayesian posterior probability ratio, and those below the branches indicate the maximum parsimony (MP) bootstrap support. An asterisk (*) replaces one node bootstrap support value because the node was not recovered in the MP bootstrap analysis. The same individuals that were used for the mtDNA phylogeny were genotyped for the five microsatellite loci. The admixture level of each genotype in individual samples is indicated by the color intensity as shown at the bottom of (B).

https://doi.org/10.1371/journal.pone.0116602.g003

MtDNA phylogenetic analyses

We inferred the molecular phylogeny of the Decacrema ants inhabiting Macaranga from a 569-bp sequence of mitochondrial COI by ML, Bayesian, and MP analyses. The phylogenetic analyses revealed five well-supported primary mtDNA clades that are identical to clades E, D, F, G1, and H of Quek et al. [17] (Fig. 3). Additionally, we identified two highly supported subclades (H1 and H2) within clade H (Fig. 3). Each method yielded a similar topology, but the monophyly of F, G1, and H was not supported by MP bootstrapping, and the monophyly of D, F, G1, and H was poorly supported by ML, MP bootstrapping, and Bayesian posterior probabilities (Fig 3).

Comparison between SSR genotypes and mtDNA clades

We mapped nrDNA SSR genotypes onto the mtDNA phylogeny of Decacrema ants inhabiting Macaranga in LHNP (Fig. 3). The degree of congruence between the SSR and mtDNA groupings was 96.9% (95/98). The six groups obtained by STRUCTURE clustering basically corresponded to the six clades obtained by the mtDNA phylogenetic analyses: MS1 corresponded to H1, MS2 to H2, MS3 to G1, MS4 to F, MS5 to D, and MS6 to E (Fig. 3). The three incongruent samples (itkS38, yk0613, and yk0820) exceptionally showed genotypic signals (>25%) derived from two different Decacrema clusters (Fig. 3): itkS38 was composed mainly of MS1 (25%) and MS4 (56%); yk0613 was composed mainly of MS1 (52%) and MS3 (38%); and yk0820 was composed mainly of MS3 (37%) and MS5 (41%) (Figs. 2 & 3).

Specificity

The preference tests of SSR genotypes toward Macaranga groups or species indicated that all genotypes showed significant preferences toward or against a particular Macaranga group or species (Table 1): MS1 preferred smooth Pachystemon and M. trachyphylla but avoided waxy Pachystemon; MS2 preferred waxy Pachystemon and M. lamellata but avoided smooth Pachystemon; MS3 preferred Pruinosae and M. rufescens but avoided smooth Pachystemon; MS4 preferred smooth Pachystemon and M. bancana but avoided waxy Pachystemon; MS5 preferred waxy Pachystemon but avoided smooth Pachystemon and MS6 preferred smooth Pachystemon and M. umbrosa but avoided waxy Pachystemon.

Download:

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

Table 1. Results of χ2 tests of the specificity of ant genotypes toward Macaranga groups/species.

https://doi.org/10.1371/journal.pone.0116602.t001

Congruence between SSR genotypes and mtDNA clades

The overall congruence between SSR and mtDNA groupings (Fig. 3) suggests that the mtDNA phylogeny reflects species boundaries and also indicates that the SSR genotyping may certainly delimit the morphospecies of Fiala et al. [13], which were identified on the basis of queen morphology and life history characters [12]. In addition, the genetic resolution of the SSR genotyping and that of the mtDNA phylogeny were almost the same: both methods divided the ants into six groups (Figs. 2 & 3). These results suggest that the COI phylogeny is suitable for inferring evolutionary history, phylogeography, and population genetics of Decacrema ants inhabiting Macaranga, and they are in contrast to the findings of Feldhaar et al. [12], who reported that only SSR genotyping delimited the morphospecies of the ants; the mtDNA phylogeny did not.

Is there a one-to-one correspondence between SSR genotype and Decacrema species? Itino et al. [14] documented congruence between the mtDNA phylogeny and species based on worker morphology. We tentatively assigned each SSR genotype to the species identified by Itino et al. [14] based on the species assignment of the mtDNA clades in our phylogeny. On this basis, we assigned MS1, MS2, and MS4 to C. borneensis; MS3 to C. sp. 4; and MS5 and MS6 to C. decamera. This assignment indicates the absence of a one-to-one correspondence between SSR genotype and species. Thus, it would be desirable to compare the SSR genotypes with the morphospecies of Fiala et al. [13]. At present, however, such a comparison is impossible because the method used to classify the morphospecies has not been published (See [11,12]).

The SSR genotyping is generally used to detect intraspecific genetic variability among individuals or populations with the use of its hypervariable polymorphism. The genotyping conducted here is an exceptional case because it was used to detect genetic variability among species within a subgenus. However, we judged this analysis to be suitable because Feldhaar et al. (2005) designed these SSR markers to delimit the Decacrema taxa and STRUCTURE clearly divided the ants into five or six clusterings (Fig. 2). If more Decacrema taxa or more samples from other site were added to the analysis, what would happen? If more Decacrema taxa from the same site (Lambir) were included, the same result as presented in the manuscript would be obtained: the presented taxa in the manuscript would be divided as it is and the added taxa would form a new grouping. However, if more samples from other sites were added, the same result would not be necessarily obtained. Previous mtDNA phylogenetic analysis revealed that the genetic variation among geographic populations was so high that each ant mtDNA clade was divided into geographic sub-clades (Quek et al. 2007). If more sensitive SSR maker than mtDNA was used, exact clustering may not be reconstructed because geographical genetic variation of SSR may reach saturation and may exceed the variation among species. The SSR genotyping conducted here was successful probably because it excluded geographical variability by using samples collected from a single site. Similar to this, Feldhaar et al. (2010) also conducted the SSR genotyping of the ants by only using samples collected from a limited region, the state of Sabah.

Hybridization

Although we detected overall congruence between the SSR and mtDNA groupings, we found admixtures of two genetic clusters in three of the 98 samples (Figs. 2 & 3); we attributed these admixture patterns to hybridization between Decacrema genotypes. Feldhaar et al. [12,50] also reported hybridization among Decacrema ants, about 2% of individuals of the captiosa group (this ant group is probably is identical to C. borneensis of Itino et al. [14]) hybridized. The existence of hybrids can introduce errors into the mtDNA phylogeny, because, as Feldhaar et al. [12] reported, the hybrids may be fertile because hybrid queens can produce workers. However, the high congruence between the SSR genotypes and the mtDNA phylogeny suggests little introgression among genotypes (Figs. 2 & 3). Therefore, hybridization probably may not cause any problematic deviations in the phylogenetic classification based on mtDNA.

To understand how hybridization occurs, it is essential to determine when and where the ants perform their nuptial flights. However, little is known about the environmental cues that lead to nuptial flights in tropical forests [51]. In Decacrema ants inhabiting Macaranga, it has been inferred that nuptial flights might occur at night because foundress queens were found to colonize saplings at night [35]. But Fiala and Maschwitz [35] reported never seeing a nuptial flight, though they performed numerous checks during both day and night hours. Thus, in a future study, the swarming and mating behaviors of Decacrema ants need to be investigated.

Host preference

Tests of Decacrema genotype preferences toward Macaranga groups or species in LHNP indicated that all genotypes (MS1, MS2, MS3, MS4, MS5, and MS6) showed significant preferences and avoidances toward particular Macaranga groups or species (Table 1). Quek et al. [17] determined host preferences by using about 400 ant samples collected throughout the Asian tropics and identified significant host preferences in the four corresponding mtDNA clades (H, G1, F, D, and E, respectively). Comparison of host preferences test results between this study and that of Quek et al. [17] showed that each genotype and its corresponding mtDNA clade showed a significant preference toward the same plant group or species (e.g., MS1 and H preferred the smooth Pachystemon). Several coadaptations between Decacrema and Macaranga could maintain this high specificity: host selection by foundress queens [52–57]; a trade-off between the host's chemical defense and the biotic defense provided by the ants [29,32,58,59]; and ant adaptations to specialized stem textures of the host [60–62]. The maintenance of high specificity over evolutionary time may promote codiversification in the Decacrema-Macaranga mutualism [16].

MS1 and MS2 (clades H1 and H2), which together correspond to clade H, showed significant preferences toward different Macaranga species: MS1 preferred M. trachyphylla (smooth Pachystemon) and MS2 preferred M. lamellata (waxy Pachystemon) (Table 1). These results suggest that clade H, although identified as a single evolutionarily significant unit, is actually composed of two genotypes, each exhibiting high specificity toward a particular plant species. This result increases the evaluated host specificity between Macaranga and Decacrema. This finding is evidence for the short-term ecological and genetic differentiation of Decacrema ants following a host shift (eg. [63,64]), in contrast to the previously suggested long-term codiversification in the Decacrema-Macaranga mutualism [65].