Coral Reefs

, Volume 33, Issue 1, pp 223–233

Cleaning up the biogeography of Labroides dimidiatus using phylogenetics and morphometrics

Authors

    • School of Biological SciencesThe University of Queensland
  • C. Riginos
    • School of Biological SciencesThe University of Queensland
  • S. P. Blomberg
    • School of Biological SciencesThe University of Queensland
  • T. Huelsken
    • School of Biological SciencesThe University of Queensland
  • J. Drew
    • Department of Ecology, Evolution and Environmental BiologyColumbia University
  • A. S. Grutter
    • School of Biological SciencesThe University of Queensland
Report

DOI: 10.1007/s00338-013-1093-2

Cite this article as:
Sims, C.A., Riginos, C., Blomberg, S.P. et al. Coral Reefs (2014) 33: 223. doi:10.1007/s00338-013-1093-2

Abstract

Cleaner fishes are some of the most conspicuous organisms on coral reefs due to their behaviour and prominent body pattern, consisting of a lateral stripe and blue/yellow colouration. All obligate cleaner fishes share this body stripe pattern, which is an important signal for attracting client fishes. However, variability in the cleaning signal of the cleaner fish Labroides dimidiatus has been documented across its range. Here, we investigate the geographic distribution of cleaner signal polymorphisms in L. dimidiatus and contrast this to phylogeographic variation in mitochondrial (mt) DNA. We used samples from 12 sites for genetic analyses, encompassing much of L. dimidiatus’ range from the Red Sea to Fiji. We obtained morphometric measures of the cleaner signal body stripe width from individuals among six of the sites and qualitatively grouped tail stripe shape. mtDNA control region sequences were used for phylogenetic and population genetic analyses. We found that body stripe width was significantly correlated with tail stripe shape and geographical location, with Indian Ocean populations differing in morphology from western Pacific populations. L. dimidiatus haplotypes formed two reciprocally monophyletic clades, although in contrast to morphology, Japanese cleaner fish fell within the same clade as Indian Ocean cleaner fish and both clade types were sympatric in Papua New Guinea. An additional novel finding of our research was that the inclusion of two closely related cleaner fish species, Labroides pectoralis and Labroides bicolor, in the phylogenetic analysis rendered L. dimidiatus polyphyletic. Overall, the findings suggest the diversity within L. dimidiatus is underestimated.

Keywords

Communication signalsCleaner fishIndo-PacificMorphometricsPhylogeographyPhylogenetics

Introduction

Most species exist as patches of populations across their distribution, and these populations may be differentiated genetically, behaviourally, or morphologically (Beebee and Rowe 2004). Body markings of coral reef fishes, including colours and patterns, have been commonly used to distinguish taxa (Randall et al. 1997), but the advent of genetic technologies has prompted many studies to ask whether these markings do in fact reflect population or species boundaries (Bernardi et al. 2002; Taylor and Hellberg 2003; Messmer et al. 2005; Drew et al. 2008). Body marking polymorphisms commonly occur in fishes and may arise from environmental and ecological selection pressures such as habitat differences, communication or mating behaviour, phenotypic plasticity enabling local adaptation, or stochastic drift from geographic isolation (Warner 1997; Rocha et al. 2005). By coupling genetic data from a putatively neutral locus with the spatial variation in a trait, such as a body marking, it is possible to test whether such traits reflect population boundaries and to infer evolutionary processes that may be acting on the trait (Beebee and Rowe 2004; Runemark et al. 2010).

Geographic concordance between genetic data and body markings has been uncovered in several studies of marine fishes (Bernardi et al. 2002; Rocha 2004; Drew et al. 2008, 2010). For example, Taylor and Hellberg (2003) showed clear genetic differentiation, at the mitochondrial (mt) DNA cytochrome b locus, between the three colour morphs of Elacatinus evelynae. In the study, physical barriers to gene flow were not present, and divergence was attributed to larval behaviour; however, the possible contribution of colour to reproductive isolation could not be tested. Other studies have found discordance between body markings and genetic boundaries (McMillan et al. 1999; Messmer et al. 2005). For instance, DiBattista et al. (2012) found that the colour morphs of Centropyge flavissima that hybridise with Centropyge eibli and Centropyge vrolikii across the Indo-Pacific region are a complex of three deeply divergent groups that correlate to geographic region and not species or colour designation, based on mtDNA cytochrome b and nuclear (n) DNA data. Whether or how body markings delineate population or species boundaries across the geographic range of many coral reef fish taxa remains unclear.

Many animals use body markings as a signal to communicate with each other. Fish cleaning services, which involve the removal of ectoparasites from cooperating client fish by cleaner fish (Feder 1966; Grutter 1997), are communicated to potential clients by body markings (Stummer et al. 2004). All obligate cleaner fishes show convergent evolution of their body marking, which includes a lateral stripe and blue and/or yellow colouration (Cheney et al. 2009). By signalling its status as a cleaner, a cleaner fish gains food (Randall 1958; Grutter 1997). Its identity as a cleaner also appears to provide it with immunity from predation (Côté 2000). Thus, a cleaner’s body marking can be considered a fitness-related trait and stabilising selection should act to maintain the integrity of the signal. However, variations in the body marking of the cleaner wrasse Labroides dimidiatus have been observed (Randall 1958; Kuwamura 1981) but not studied across the species’ range or compared against genetic variation. One possibility is that cleaner marking variants are associated with distinct evolutionary groups, in which case the geographic distribution of the cleaner marking should match genetic processes. Alternatively, variation in the cleaner marking may be decoupled from neutral genetic processes (either due to stabilising selection or disruptive selection), in which case the geographic distribution of the cleaning marking would not be concordant with genetic patterns.

Labroides dimidiatus has a pelagic larval duration of around 20 d (Victor 1986) and a large geographic distribution ranging from South Africa in the Indian Ocean to French Polynesia in the Pacific Ocean. Genetic surveys throughout the Indonesian Archipelago have revealed many examples of deep phylogeographic divergence (Barber et al. 2002; Crandall et al. 2008a, 2008b; DeBoer et al. 2008; Kochzius et al. 2009; Nuryanto and Kochzius 2009). Carpenter et al. (2011) summarise the patterns of population structure between marine populations of different invertebrate and fish taxa within this region due to various barriers. One of the most prominent historical barriers to dispersal, the Sunda shelf, was a major land barrier restricting exchanges between the Indian and Pacific Oceans during low sea level stands throughout the Pleistocene (Voris 2000). For L. dimidiatus, Drew et al. (2008) found near reciprocal monophyly for the mtDNA control region between populations from Fiji and northwest New Guinea (Taluk Cendarawasih), but only these two populations were surveyed. L. dimidiatus may be one of the most ubiquitous and best-documented fishes of coral reefs and has been recognised as a key stone species as it has a large impact on the parasite load, diversity and abundance of other fishes (e.g. Randall 1958; Grutter 1997, 1999; Bshary 2003; Grutter et al. 2003; Waldie et al. 2011); nevertheless, the genetic diversity and population structure across its range remain largely undocumented.

Thus, the aim of this study was to comprehensively examine both morphology and genetic differentiation across most of the range of L. dimidiatus with special attention to: (1) the variation in the body stripe width and tail stripe shape of L. dimidiatus within and among populations and the geographic distribution of the body stripe pattern polymorphism; (2) the population genetic structure of L. dimidiatus across sampled sites; and (3) whether the variation found in the body stripe width was concordant with genetic structure. The contrasts between morphological and genetic data provide insights into the evolutionary processes acting on the cleaner signal.

Materials and methods

Sampling and tissue collection

Sampling was carried out at nine geographical sites: the Red Sea (Thuwal); Seychelles, Maldives, Cocos-Keeling Island and Christmas Island from the Indian Ocean; and Japan (Murote Island), Papua New Guinea (PNG) (Kimbe Bay), Lizard Island and Heron Island from the Pacific Ocean (Fig. 1a). Fin clips or whole specimens were collected for tissue samples and kept in either a dimethyl sulphoxide–ethanol mix or 90–100 % ethanol. For the sites Heron Island, Lizard Island, PNG, Japan, Seychelles and Maldives, photographic images of the left lateral side of each sampled fish (pinned out) were taken on white waterproof paper. Images of fish from Red Sea, Cocos-Keeling and Christmas Island were not taken. Images of specimens and tissue samples were not originally cross-referenced when obtained; therefore, genetic and morphometric data could not be directly linked other than by site. Four individuals of Labroides pectoralis were also collected from Christmas Island.
https://static-content.springer.com/image/art%3A10.1007%2Fs00338-013-1093-2/MediaObjects/338_2013_1093_Fig1_HTML.gif
Fig. 1

a The map shows sites sampled for Labroides dimidiatus across Indian and Pacific Oceans for genetic data (site name colour codes are the same as the tree), those sites with morphometric and genetic data have pie charts. Pie charts represent the proportion of each tail stripe shape found at a site (F fan, HH half-hook, H hook, R round, N not scored). Scale bar is in kilometres. b Photograph of the four tail stripe shapes from left to right: Fan, Half-Hook, Hook and Round. c Phylogenetic relationships of L. dimidiatus based on the mitochondrial DNA control region. The Bayesian tree was based on the general time-reversible evolutionary model. Numbers at nodes are Bayesian posterior probabilities using five million generations. Scale bar represents number of base pair substitutions. Site names are colour coded (light orange circle Heron Island, light green circle Lizard Island, yellow circle Fiji, dark green circle PNG, red circle Japan, dark red circle West Papua, dark orange circle Indonesia, light blue circle Christmas Island, dark blue circle Cocos-Keeling Island, blue circle Maldives, pink circle Seychelles and purple circle Red Sea)

Table 1

Population genetic analyses carried out across the sampled sites

Site

n DNA

n Morph

n Tail

Clade

Haplotype diversity (h)

Polymorphic sites

Nucleotide diversity (Pi  %)

Tajima’s D

Red Sea

4

1

1.00

10

0.021

0.083

Seychelles

9

20

20

1

1.00

21

0.023

−0.818

Maldives

8

12

12

1

1.00

15

0.021

0.222

Cocos-Keeling Island

5

1

1.00

12

0.020

−0.954

Christmas Island

7

1

1.00

21

0.026

−0.750

Indonesia

2

1

West Papua

12

1

1.00

18

0.017

−0.813

Japan

13

16

16

1

0.92

20

0.018

−0.738

PNG1

7

7

7

1

1.00

15

0.028

1.007

PNG2

12

12

12

2

0.83

11

0.007

−1.738*

Lizard Island

18

19

18

2

0.83

21

0.013

−1.539*

Heron Island

32

34

34

2

0.75

23

0.011

−1.723*

Fiji

13

2

1.00

21

0.014

−1.707*

The number of individuals used per site for genetic (n DNA), morphometric (n Morph) and tail stripe shape (n Tail) analyses. Population genetic analyses are for the mitochondrial DNA control region. Papua New Guinea (PNG) was divided into two new sites, PNG1 and PNG2, based on the clades of the phylogenetic tree. The clade number represents the clade group from the phylogenetic tree. Indonesia was not used in population genetic analyses as it contained only two individuals

– No data

P value <0.05

Morphometric analyses

In this study, we defined the cleaner signal of L. dimidiatus as the body stripe width and tail stripe shape. Individuals from six sites (Maldives, Seychelles, Japan, PNG, Lizard Island and Heron Island) were used for morphometric analyses (Table 1). The program ImageJ v1.31 (Abramoff et al. 2004) was used to take body stripe width measurements from digitised images of each fish (Electronic Supplemental Material, ESM Fig. S1). Each individual’s tail stripe shape was then grouped into one of four distinct categories called Fan, Half-Hook, Hook and Round (Fig. 1b). The resulting matrix of morphometric response variables (body stripe width) and two categorical variables (site and tail stripe shape) was then imported into R v2.12.1 (R D Core Team 2011), which was used for all subsequent statistical analyses.

To avoid multicollinearity, the variables were subjected to a dimension reduction principle components analysis (PCA) (Johnson 1998), and the number of principle component (PC) scores included in subsequent analyses were determined using a scree plot (Everitt 2005). Possible groupings of the morphometric data were explored with PCA scatterplots by the two categories variables—site and tail stripe shape. The resulting three PC scores that contained most of the variation in the original data were subsequently used as the new response variables in further analyses.

To investigate whether the body stripe width varied among the levels of the two categorical variables, we used a nonparametric multivariate analysis of variance (MANOVA) based on euclidean distances (Anderson 2001) and assumptions of homogeneity of multivariate dispersion were tested (Anderson 2006). To discriminate among the sites and tail stripe shape categories that were significantly different in the nonparametric MANOVA, we used a canonical discriminate analysis (CDA) (Johnson 1998).

Genetic amplification and sequencing

For samples from the Red Sea, Christmas Island, Japan, Lizard Island, Maldives, Heron Island, PNG and outgroup samples of L. pectoralis, we used a salt extraction protocol to extract genomic DNA (Lee et al. 1995). All DNA samples were suspended in 100 μL purified deionised water and stored at −20 °C. A partial segment of the mtDNA control region was amplified for each sample in 30 μl polymerase chain reactions (PCR). Each reaction tube contained 0.5–1 μl of DNA, 3 μl each of titanium Taq-buffer and DNTPs (10 mM), 0.3 μl of titanium Taq-polymerase and 0.6 μl each of forward and reverse primers (Lee et al. 1995). PCR conditions were as follows: 95 °C for two min, then 35 repeats of 95 °C for 30 s, 55 °C for 45 s and 72 °C for 45 s; and a final extension of 72 °C for 10 min. Samples were purified with exonuclease I (20 U μl−1) and antarctic phosphatase (5 U μl−1) with a PCR cycle of 37 °C for 15 min and 80 °C for 15 min and then sent to Macrogen (South Korea) for capillary sequencing. Unpublished sequences from Seychelles, Cocos-Keeling Island, PNG, Heron Island and four samples of Labroides bicolor collected from Seychelles were obtained from the James Cook University Marine Molecular Laboratory. Additionally, published sequences of L. dimidiatus from Fiji, Indonesia (Krakatau) and West Papua (Teluk Cenderawasih) were included. In total, samples from 12 sites were used for genetic analyses. Multiple sequences of six species obtained from GenBank were used as further outgroups. All samples, their origins and GenBank accession numbers are given in the electronic supplementary material (ESM Table S1).

Phylogenetic and population genetic analyses

The software program CodonCode Aligner v3.5 (CodonCode Corporation 2009) was used to align and edit sequences. Se-al v2.0 (Rambaut 1996) was used to check alignments, remove primer sequences and align additional sequences. MrBayes v3.1 (Ronquist and Huelsenbeck 2003) was used to build Bayesian gene trees among haplotypes. We used jmodeltest v0.1.1 (Posada 2008) to select the general time-reversible model with a proportion of invariable sites and a gamma-shaped variation in rates across sites as the most appropriate model of evolution based on AIC scores (Felsenstein 2004). For the Bayesian analysis, four chains (two chains in two runs) of each five million generations were run with trees sampling every 1,000 generations. Sampling was run until convergence <0.05 was found. A burn-in of 100,000 trees was used for tree calculation. A total of 145 individuals of L. dimidiatus and 55 individuals for outgroups were used in the analyses (Fig 1c and ESM Table S1).

The program Arlequin v3.11 (Excoffier et al. 2005) was used for population genetic analyses and analysis of molecular variance (AMOVA). Since the gene tree found two independent clades of L. dimidiatus, further population genetic analyses were conducted separately on these groups. PNG was the only site that contained individuals from both clades, PNG haplotypes that grouped with clade 1 were renamed as PNG1 and those that grouped with clade 2 were renamed as PNG2. For population genetic analyses 142 individuals were used, as one individual each from Heron Island, Lizard Island and PNG was excluded due to poor sequencing quality (Table 1). In addition, Indonesia (Krakatau) was excluded from the AMOVA analysis as it only contained two individuals; thus, a total of 140 individuals from 11 sites were included. Molecular diversity indices and genetic distances between haplotypes were based on the Tamura and Nei (1993) model. Computation of pairwise genetic differentiations (ΦST) used a significance level of α = 0.05. Tajima’s D test of neutrality was used to assess the neutrality of the genetic marker (Tajima 1989).

Combined morphometric and genetic analyses

We used Mantel tests (Mantel 1967) in R v.2.12.1 (R D Core Team 2011) with the Ecodist package (Goslee and Urban 2007) to assess whether there was spatial concordance between geographical location, genetic divergence and body stripe width structure. The Mantel test is typically used to show an isolation-by-distance pattern and assumes all individuals can disperse equally in all directions in a homogenous environment (Wright 1942). Genetic isolation by distance was tested by comparing pairwise ΦST values against great circle distances for all populations combined (with PNG1 and PNG2 treated as two separate populations) and restricted to comparisons within each clade. Morphological isolation by distance similarly was estimated by taking pairwise euclidean distances and comparing against geographical distance in the Mantel test framework. Student’s two-tailed t test was used to test for a difference in the means of the body stripe width measures by clades in Excel v12.0.4518.1014. However, PNG was excluded since individuals from this site fell in both clades. Since the images of specimens and tissue samples were not originally cross-referenced when obtained, we were unable to directly compare individual morphological distances against genetic distance, and therefore, population means by site were used.

Results

Morphometric analyses

In total, 10 morphometric measures were made from 13 landmarks on 120 individuals of L. dimidiatus (see Table 1 and ESM Fig. S1). Tail stripe shapes showed a gradient from only Round in the Indian Ocean to a variety of tail stripe shapes in the Pacific Ocean. Heron Island, Maldives and Seychelles all had a single tail stripe shape, while Lizard Island, PNG and Japan all had multiple tail stripe shapes. Japan was the only Pacific Ocean site that shared the Round tail stripe shape with the other Indian Ocean sites (Fig. 1a).

Most of the original variables were included in the variance of the first three PC scores, except for the measure Peduncle Top (ESM Table S2). The first three PC scores accounted for 54, 16 and 13 % of the variation, totalling 83 %. For the scatterplot PC1 versus PC2 by site (Fig. 2a), the PC1 axis featured a gradient from thinner body stripes for individuals from Indian Ocean sites to wider body stripes for individuals from Pacific Ocean sites. The PC2 axis represented similar, but also more complex, changes than the PC1. It featured a gradient from wider body stripes at the anterior of the fish and narrower body stripes at the posterior for Maldives, PNG and Lizard Island individuals to thinner body stripes at the anterior of the fish and wider body stripes at the posterior for Seychelles, Heron Island and Japan individuals. The PCA scatterplot by tail stripe shape revealed two main clusters (Fig. 2b). On the PC1 axis individuals with a Round tail stripe shape had thinner body stripes, while those with Fan, Half-Hook and Hook tail stripe shapes had wider body stripes. The PC2 axis did not show any further groupings by tail stripe shape.
https://static-content.springer.com/image/art%3A10.1007%2Fs00338-013-1093-2/MediaObjects/338_2013_1093_Fig2_HTML.gif
Fig. 2

Scatterplot of the principal component (PC) analysis scores by site and tail stripe shape for Labroides dimidiatus. a Scatterplot of the PC1 versus PC2 from the analysis of individuals across sites (light orange circle Heron Island, light green circle Lizard Island, dark green circle PNG, red circle Japan, blue circle Maldives and pink circle Seychelles) shows individuals in Indian Ocean sites have thinner body stripe widths than those in the Pacific Ocean sites. b Scatterplot of the PC1 and PC2 scores from the analysis of individuals across tail stripes (F fan, HH half-hook, H hook and R round) indicates individuals with a Round tail shape have a greater variance of body stripe widths and are significantly different to individuals with the other three tail shapes

The nonparametric MANOVA on the three PC variables found a significant difference in the body stripe width by the categorical variables, site and tail shape, but no interaction between the two categorical variables (Table 2). A test of multivariate homogeneity of group dispersion found site to be nonsignificant (F = 1.5401, df = 5, P > 0.05), while tail stripe shape was significantly different (F = 6.0347, df = 3, P < 0.01). Pairwise comparisons of observed and permuted p-values showed that tail stripe shape was significantly different between the Round tail shape and the other three tail shapes, since individuals with Round tail shapes had greater variance in body stripe widths. The subsequent CDA by site found significant between-site variation across all three canonical variates (Can1: F = 12.72, df = 15, P < 0.01; Can2: F = 9.85, df = 8, P < 0.01; Can3: F = 10.73, df = 3, P < 0.01). The between-class variation explained by the canonical variates was 60.4 % for Can1, 23.6 % for Can2 and 15.9 % for Can3. The CDA plot, however, did show groupings between Japan and Heron Island and PNG and Lizard Island (Fig. 3a). The CDA by tail stripe shape found significant between tail stripe shape variation across only the first canonical variate (Can1: F = 2.81, df = 9, P < 0.01). The canonical variate explained 93.6 % of the variation between tail stripe shapes, and this significant variation was only between Round and the other tail stripe shapes (Hook, Half-Hook and Fan) (Fig. 3b).
Table 2

Nonparametric multivariate analysis of variance analysis (MANOVA), which involved 4,999 permutations and tested whether there was a significant difference in the body stripe width (response variable) among the categorical variables, site and tail stripe shape, and any interaction between the two categorical variables

Test

Df

SS

Mean sqs

F. model

R2

P value

Site

5

0.059

0.012

35.90

0.587

0.00001

Tail stripe shape

3

0.004

0.001

4.47

0.044

0.0005

Site × Tail stripe shape

3

0.001

0.0003

0.88

0.009

0.49

Residuals

110

0.036

0.0003

 

0.360

 

Total

121

0.101

  

1.000

 
https://static-content.springer.com/image/art%3A10.1007%2Fs00338-013-1093-2/MediaObjects/338_2013_1093_Fig3_HTML.gif
Fig. 3

Scatterplot of canonical variates (CV) analysis by site and tail stripe shape for Labroides dimidiatus. a CV1 versus CV2 for site shows most sites (light orange circle Heron Island, light green circle Lizard Island, dark green circle PNG, red circle Japan, blue circle Maldives and pink circle Seychelles) are significantly different to zero on both axes. The groupings between Japan and Heron Island, and PNG and Lizard Island indicate similarities in the body stripe pattern between individuals from these sites, while body stripe patterns of individuals from Seychelles and Maldives are significantly different. b CV1 versus CV2 for tail stripe shapes (F fan, HH half-hook, H hook and R round) shows only the Round tail shape is significantly different to zero, and the body stripe width of these individuals is significantly different to individuals with the other three tail shapes

Phylogenetics and population genetic structure

The mtDNA control region gene tree included 145 individuals of L. dimidiatus and eight outgroup species. Our analysis resulted in a well-resolved phylogeny for L. dimidiatus containing two strongly supported clades (Fig. 1c). The most notable feature is the polyphyly of L. dimidiatus, whereby populations of L. dimidiatus are split between two clades. Clade 1, which is well supported (posterior probability = 1), contains only L. dimidiatus with individuals largely drawn from the Red Sea, Indian Ocean and northwest Pacific. Clade 2, which is also strongly supported (posterior probability = 1), contains three well-supported sub-clades. One of these sub-clades contains individuals consisting only of L. pectoralis (posterior probability = 1) and another contains individuals consisting only of L. bicolor (posterior probability = 1), suggesting that these two species are well supported as reciprocally monophyletic groups. The third sub-clade within clade 2 consists entirely of L. dimidiatus individuals taken from populations in East Australia, PNG and Fiji. Therefore, clade 1 is sister to L. bicolor, L. pectoralis and clade 2 of L. dimidiatus. The net genetic uncorrected pairwise distance between the two clades of L. dimidiatus was 13.1 % sequence divergence. Another striking aspect of the phylogeny is that half the PNG individuals fall into clade 1, while the other half fall into clade 2. Samples for PNG were collected over two time periods and sequenced by two separate laboratories with both clades recovered from both time points and laboratories, providing strong support for this finding. Individuals of L. pectoralis were sampled from Christmas Island and L. bicolor from Seychelles. These individuals are from the Indian Ocean; however, they sit within clade 2 containing individuals of L. dimidiatus from the Pacific Ocean.

Measures of genetic diversity within each site are summarised in Table 1. Haplotype diversity was very high for all sites. Tajima’s D test of neutrality was statistically significant (P < 0.05) for Heron Island, Lizard Island, Fiji and PNG2, but no populations are significant if a Bonferroni correction is applied. The pairwise φST values were high and significant between clades but not within clades. An AMOVA confirmed the high structure between the two clades (AMOVA by clades: φST = 0.87, P < 0.001), while an AMOVA restricted to clade 1 showed some structure (AMOVA by clade 1: φST = 0.25, P < 0.001) and clade 2 had no significant structure (AMOVA by clade 2: φST = 0.007, P = 0.20).

Concordance between the morphometric and phylogeographic structure

Geographical distance was a significant predictor of both genetic (r = 0.351) and morphological (body stripe width) distance (r = 0.885) (Table 3). An isolation-by-distance pattern was discernible within clade 1 (r = 0.54, P = 0.02) but not within clade 2 (r = 0.10, P = > 0.05). No significant correlation was found between the genetic and body stripe width distance matrices (P = 0.18). All distance matrices showed a positive relationship with one another. The Student’s t test did not find a significant difference in the means of body stripe width measures by clade (P = 0.13).
Table 3

Mantel tests for the correlation of body stripe width with genetic distance matrices, and geographical distance with both body stripe width and genetic distance matrices

Test

Populations

Mantel r

P value

Genetic vs Body stripe width

4

0.28

0.18

Genetic vs Geographical

12

0.351

0.012

Body stripe width vs Geographical

5

0.885

0.012

The genetic distance matrix was based on pairwise ΦST values. The body stripe width distance matrix was based on euclidean distances between the rows of the data table. The geographical distance matrix was based on great circle distances. Papua New Guinea (PNG) was omitted from the genetic by body stripe width test and treated as two separate populations, PNG1 and PNG2

Discussion

We found that both morphology and mtDNA clades are geographically delineated but that the geographic divisions between morphological and mtDNA types differ such that disjunctions within the two data types are poorly correlated (Fig 1a, Table 3). Unexpectedly, the inclusion of two closely related cleaner wrasses in the phylogenetic analysis rendered L. dimidiatus polyphyletic (Fig. 1c). Phylogenetic analyses further revealed two deep mtDNA lineages (d = 13.1 %) of L. dimidiatus, which were sympatric at Kimbe Bay, indicating a phylogeographic transition in northern New Guinea.

Cleaner signal structure

We found significant variation in the body stripe width and tail stripe shapes among sites (Table 2). Similarities among the body stripe widths of individuals were observed for Japan and Heron Island, and PNG and Lizard Island (Fig. 3a), contributing to the positive relationship between morphological and geographical distance (P = 0.00001). In contrast, body stripe widths of individuals from Maldives and Seychelles were significantly different, even though these populations share the same Round tail stripe shape (Fig. 1a). Interestingly, PNG and Lizard Island were the only two sites with individuals that had the Hook tail shape; this may have led to similarities in their body stripes, and therefore, the sites being grouped in the CDA (Fig. 3a).

Variation in the body stripe pattern across L. dimidiatus’ distribution has been previously documented. Bleeker (1851, as cited by Randall 1958) described L. paradiseus because it differed from L. dimidiatus by having what we termed a Round tail shape, but it was later deemed only a variant and synonymised with L. dimidiatus (Randall 1958). The paradiseus form had been documented in Japan (Kuwamura 1981), the Indian Ocean, Marshall Islands and Philippines, while the dimidiatus form, or Fan tail stripe, was found mainly in the Pacific Ocean (Randall 1958). The Half-Hook and Hook tail shapes categorised in the present study could well have been recorded above as the paradiseus form since we also found them in similar areas (i.e. PNG, Japan). Overall, our findings support the observations of Randall (1958) and Kuwamura (1981).

Generally, the body stripe width and tail stripe shape of L. dimidiatus are geographically structured by ocean basin. Fan, Half-Hook and Hook tail stripe shapes are found only in Pacific Ocean sampled sites, while the Round tail shape is mainly found in the Indian Ocean, with the exception of a few individuals in Japan (Fig. 1a). The significant difference observed between the tail stripe shapes (Fig. 3b) is due to differences in body stripe width. The PCA scatterplots illustrate that Pacific Ocean individuals, with mainly Fan, Half-Hook and Hook tail shapes, have wider body stripes than Indian Ocean individuals with Round tail stripes (Fig. 2a, b); therefore, the body stripe width seems to be loosely linked to the tail shape.

Variations observed in the cleaner signal between sampled sites could be due to local environmental pressures or stochastic drift. Local selection pressures leading to variations in the cleaner signal of L. dimidiatus among sites may come from local environmental differences, in combination with the visual systems of the client fish assemblage at each site (discussed further in Marshall and Vorobyev 2003). For the Caribbean cleaner gobies, Lettieri et al. (2009) found that the visual sensory systems of client fish varied across sampled sites and suggested clients may have selected for contrasting body colour patterns in the cleaner gobies. Thus, heterogeneous environments and disparate assemblages of client fish visual sensory systems between our sampled sites could be driving regional variation in the cleaner signal of L. dimidiatus. A similar argument could be made for plastic phenotypic responses due to local abiotic or biotic triggers. However, whether the observed variation in body marking is genetically inherited or environmentally induced is unknown. Regardless of what process explains the morphological differentiation among populations, it is likely that the total scope for variation in the cleaner signal is limited; after all, maintaining a local optimal range permitted by the cleaner mutualism is imperative to the cleaner fish’s fitness (Stummer et al. 2004). Further experimental work to determine both the heritability of the cleaner signal and fitness consequences for differing phenotypes would be required to resolve these competing hypotheses.

Phylogenetic and population genetic structure

By including a number of sequences from the two closely related cleaner fish species, L. bicolor and L. pectoralis, the polyphyletic pattern of L. dimidiatus’ evolutionary history was revealed (Fig. 1c). Monophyly of the Labroides genus has been well supported (Westneat and Alfaro 2005; Cowman et al. 2009), but our results call into question the current taxonomy of L. dimidiatus. Whereas L. pectoralis and L. bicolor are supported in the mtDNA gene tree as monophyletic assemblages, their presence within clade 2 renders L. dimidiatus polyphyletic. The type locality for L. dimidiatus is Mauritius (Randall 1958). While we do not have samples from this locality, we do have sequence data for individuals from nearby Seychelles. These data are nested within clade 1, suggesting that the L. dimidiatus from clade 2 represent a unique monophyletic assemblage, centred in the western Pacific. Peripheral populations rendering widespread species paraphyletic are not uncommon phenomena (Feldman and Spicer 2006; Hull et al. 2008; Mulcahy 2008; Drew and Barber 2009; Gaither et al. 2011).

The reciprocal monophyly between the two clades and the sympatric occurrence of clades in northern PNG resulted in high haplotype and nucleotide diversity at this site relative to other sites (Table 1) and is a pattern highly suggestive of secondary contact (as in Gaither and Rocha 2013). Secondary contact at the Indo-Pacific junction is noted for several marine organisms (Bay et al. 2004; Crandall et al. 2008a; Carpenter et al. 2011; Gaither et al. 2011). Phylogeographic breaks between West Papua (Teluk Cenderawasih) and populations to the west of this location have been identified in other marine taxa, including giant clams (DeBoer et al. 2008), mantis shrimp (Barber et al. 2006) and fish (Carpenter et al. 2011). The point of secondary contact that we observe, however, is to the east of West Papua; genetic disjunctions have been noted in this same general location in giant clams (T Huelsken pers. comm.), blue sea stars (ED Crandall pers. comm.) and butterflyfish (McMillan et al. 1999). Whether or not reproductive isolation accompanies the deep mtDNA divergence is unknown.

Within clades, we see that clade 1, being widespread across the Indo-Pacific and Indian Ocean, shows substantial genetic structure among sites (and an isolation-by-distance pattern), whereas clade 2, containing populations distributed across Eastern Australia, Fiji and PNG, seems to be panmictic (Table 3). Signals of genetic isolation among broadly distributed populations could be due to either an absence of connectivity or local selection against migrants. We acknowledge that the results here are drawn from a single mtDNA locus and that inclusion of more loci might clarify evolutionary processes. For example, Drew et al. (2008) looked at L. dimidiatus from Fiji (our clade 1) and Indonesia (our clade 2) with mtDNA and resolved the same phylogenetic pattern we did here. However, when the same samples were examined using the nDNA RAG2 locus, the clear bifurcation was reduced to a polytomy. This could have resulted from the phylogenetic break occurring too rapidly to be resolved with the slower nDNA or from a real difference in patterns of inheritance. Nonetheless, the strength of the data set described here is based on the cohesive morphological and mtDNA sampling across the species’ range; most genetic surveys overlook quantitative treatment of morphological variation, whereas our combined analyses of both sources of data raise many interesting questions for future research on L. dimidiatus.

Contrasting biogeographic patterns of phylogeny and morphology

We found strong structuring in both the cleaner signal morphology and mtDNA phylogeography of L. dimidiatus. However, the two data sets do not conform in biogeographic structure. Instead the two highly divergent clades show a break at PNG (Fig. 1a), with individuals from PNG showing morphological uniformity. This pattern parallels with butterflyfish, whereby divergent mtDNA lineages of butterflyfish co-occur in eastern PNG and no correlation was found between morphology, mtDNA genotype and mate choice (McMillan et al. 1999). Conversely, Drew et al. (2008) found that the colour morphs of L. dimidiatus individuals in Fiji included a yellow morph and a normal morph (dark stripe with blue colouration). However, there were no genetic differences between the two morphs.

The broad-scale contrast between morphological attributes and putatively neutral mtDNA variation that we observe in L. dimidiatus suggests that this is not a simple case of two cryptic species with morphological divergence accompanying genetic divergence (i.e. stochastic drift). It seems likely that populations are matching their environment either through selection to local conditions or through phenotypic plasticity. What is known is that the body stripe width must remain within an optimal range permitted by the cleaning mutualism. Therefore, we expect strong selection to maintain the fidelity of the communication signal and lead to evolutionary stasis of the cleaner signal.

Although the phenotypic evolution of L. dimidiatus is most likely influenced by its ecological interaction as a cleaner, we have found far greater intraspecific genetic divergence within L. dimidiatus than previously recognised. Further analyses, particularly including multiple unlinked loci and manipulative experiments, will increase our understanding of selective pressures on the cleaner signal, the evolutionary history of the Labroides genus and the taxonomic status of L. dimidiatus.

Acknowledgments

ASG and CR were funded by the Australian Research Council and The University of Queensland. JAD was supported by the National Science Foundation, JD and CT MacArthur Foundation of the Encyclopaedia of Life and Columbia University. For specimen collections and data, we thank JH Choat, R Robertson, L van Herwerden, A Anderson, MA Johnson, M Gauthier and N Okuda, as well as many other volunteers and research assistants for help with the collection of fish. We thank the Coral Reef Ecology and Ecological and Evolutionary Genetics Laboratories for helpful discussions and laboratory support. CAS thanks C Mills for the title and S Calandra for continued support and advice. The authors would also like to thank two anonymous reviewers for their helpful comments and suggestions.

Supplementary material

338_2013_1093_MOESM1_ESM.doc (24 kb)
Supplementary material 1 (DOC 24 kb)
338_2013_1093_MOESM2_ESM.doc (74 kb)
Supplementary material 2 (DOC 75 kb)
338_2013_1093_MOESM3_ESM.eps (1.2 mb)
Supplementary material 3 (EPS 1192 kb)

Copyright information

© Springer-Verlag Berlin Heidelberg 2013