Cleaners among wrasses: Phylogenetics and evolutionary patterns of cleaning behavior within Labridae

Accepted Manuscript Cleaners amongst wrasses: phylogenetics and evolutionary patterns of cleaning behavior within Labridae Vikram B. Baliga, Chris J. Law PII: DOI: Reference:

S1055-7903(15)00272-9 http://dx.doi.org/10.1016/j.ympev.2015.09.006 YMPEV 5295

To appear in:

Molecular Phylogenetics and Evolution

Received Date: Revised Date: Accepted Date:

11 July 2015 5 September 2015 8 September 2015

Please cite this article as: Baliga, V.B., Law, C.J., Cleaners amongst wrasses: phylogenetics and evolutionary patterns of cleaning behavior within Labridae, Molecular Phylogenetics and Evolution (2015), doi: http://dx.doi.org/ 10.1016/j.ympev.2015.09.006

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

1

Title: Cleaners amongst wrasses: phylogenetics and evolutionary patterns of cleaning behavior

2

within Labridae

3 4 5 6

Vikram B. Baligaa*, Chris J. Lawa a

Department of Ecology and Evolutionary Biology, Long Marine Laboratory, University of

California Santa Cruz, 100 Shaffer Road, Santa Cruz, CA 95060, USA

7 8 9 10 11 12

*Correspondence:

13

Vikram B. Baliga

14

Center for Ocean Health

15

Long Marine Laboratory

16

100 Shaffer Road

17

Santa Cruz, CA 95060

18

Email: [email protected]

19

Phone: (949) 307-0880

20

21

Abstract

22

Cleaner fishes remove and consume ectoparasites and are often categorized by whether they

23

perform this behavior: 1) predominately as juveniles, 2) facultatively throughout ontogeny, or 3)

24

obligately. Through a literature search, we confirmed that with at least 58 species exhibiting

25

cleaning behavior, the Labridae (wrasses, parrotfishes, and allies) contain the highest diversity of

26

cleaner fishes. In fact, there are 3-4 times as many cleaners within labrids as there are in any

27

other marine group. The distribution and underlying causes of this exceptional diversity have not

28

been determined. Here, we assess the topological and temporal patterns of labrid cleaner

29

evolution. We used maximum likelihood and Bayesian approaches to infer the phylogenetic

30

relationships and divergence times between 320 labrid species (50.7% of nominal species). We

31

then employed stochastic character mapping to infer how and when cleaning behavior evolved.

32

We estimate that cleaning has independently evolved 26-30 times in the Labridae, and all such

33

events likely occurred no earlier than in the late Miocene. Given the current sampling and pattern

34

of transitions, we hypothesize that the majority of facultative or obligate cleaning may have

35

evolved through heterochrony.

36 37 38 39 40 41 42

Keywords: cleaner fishes; Labridae; phylogeny; stochastic character mapping

43

1. Introduction

44

The Labridae (wrasses, parrotfishes, and hogfishes) is a speciose group of marine perciform

45

fishes that occupies diverse ocean habitats worldwide. Labrids are well known for being some of

46

the most common and functionally important inhabitants of coral reef ecosystems, revealing

47

tremendous diversity in morphology and trophic strategies (Wainwright et al., 2004; Bellwood et

48

al., 2006; Price et al., 2011). Labrids feature, among myriad trophic strategies, extreme

49

specializations such as corallivory, planktivory, and molluscivory.

50

One of the most fascinating specializations within the Labridae is cleaning behavior. Cleaner

51

fishes are taxa that remove and consume ectoparasites off other organisms. The evolution of

52

cleaning behavior presents one of the few examples of mutualisms among vertebrates (Bronstein,

53

1994; Poulin and Grutter, 1996). While cleaners typically clean other fishes, they have also been

54

observed to inspect a variety of marine vertebrates and invertebrates (see Grutter, 2010 for a

55

review of cleaner fish behaviors). The presence of cleaners in a habitat can have tremendous

56

ecological consequences. For instance, experimental removal of the bluestreak cleaner wrasse

57

(Labroides dimidiatus) has been shown to affect the behavior, recruitment dynamics, and sizes of

58

client fishes (Waldie et al., 2011).

59

Cleaning is not exclusive to labrids; in fact, at least 18 marine families of fishes include at least

60

one member that cleans. Coté (2000) provides an extensive list of cleaner fishes. According to

61

Coté (2000), 50 species of labrids are documented as cleaners. This is three times as many

62

species as in the next highest group the Gobiidae, within which 14 species of cleaners are

63

recognized. This suggests cleaner fish species richness is not directly proportional to clade

64

diversity, especially when considering the Gobiidae has close to 2,000 extant members.

65

Furthermore, of the various groups of marine fishes in which cleaning is found, the

66

overwhelming majority contain five or fewer species that clean (Coté, 2000). These metrics

67

underscore the exceptional diversity of labrid cleaners, marking labrids as a model clade within

68

which to explore the evolution of cleaning.

69

Cleaner fishes can be categorized by whether they perform the behavior 1) predominately as

70

juveniles, 2) facultatively throughout ontogeny, or 3) obligately (Coté, 2000). Obligate cleaners

71

are more conspicuous and most of what is known about cleaning behavior has been determined

72

through observing species in the obligate cleaner genus Labroides. For example, L. dimidiatus

73

commonly maintains “cleaning stations”, small areas that attract visiting “client” organisms

74

(Youngbluth, 1968). In L. dimidiatus, cleaning interactions often begin with the cleaner fish

75

approaching a potential client and presenting itself by swimming in a vertical oscillatory pattern

76

(Randall, 1958; Gorlick et al., 1978). A receptive client will then pose to solicit cleaning by

77

holding still in the water column, spreading its pectoral and pelvic fins, opening its jaws, and

78

flaring its opercula laterally (Losey, 1972; Coté et al., 1998). The cleaner will dart around the

79

client’s body as it picks off ectoparasites, most commonly gnathiid isopod larvae (Grutter, 1996)

80

that may be embedded in the fins, gills, buccal cavity, and pharyngeal chamber of the client

81

(Grutter, 1996; Coté, 2000; Grutter, 2010).

82

The complexity of behaviors that labrid cleaners exhibit varies widely across the diversity of taxa

83

that clean. Contrary to what is recorded for the obligate cleaner genus Labroides, most labrid

84

cleaners do not perform oscillatory swimming, and many do not hold cleaning stations. What

85

does unify cleaners in the Labridae, however, is 1) the ability to detect, remove, and consume

86

ectoparasites off other taxa, and 2) acknowledgement by client species, who allow cleaners to

87

approach and inspect them.

88

There is growing evidence that cleaners share a variety of morphological characteristics related

89

to feeding. When removing and consuming prey off a substrate, the cleaners Thalassoma

90

lutescens, Larabicus quadrilineatus and Labroides dimidiatus employ low-displacement, fast

91

jaw movements that allow for rapid gape cycles on individually-targeted items (Baliga and

92

Mehta, 2015). Furthermore, the cranial skeletons of cleaner fishes in Thalassoma show reduced

93

vertical gape sizes, smaller bite forces, and jaws with reduced mobility when compared to those

94

of non-cleaner congeners (Baliga and Mehta, 2014). Thus, while it may be reasonable to predict

95

that this feeding specialization, which may involve a variety of behavioral and morphological

96

adaptations, would show relatively few independent origins, the staggering species richness of

97

cleaners in the Labridae compared to that of other marine families is perplexing.

98

How and when cleaning behavior arose in the Labridae has not been established and requires

99

further study with phylogenetic information. While several recent efforts have put forth several

100

sound phylogenetic hypotheses for labrids (Kazancioğlu et al., 2009; Cowman and Bellwood,

101

2011), these phylogenies do not extensively cover all genera in which cleaning is known to

102

occur, notably in the genus Bodianus and close allies. Fortunately, thanks to more recent

103

sampling efforts (e.g. Hubert et al., 2012), we find opportunity here to add further insight to the

104

history and topological organization of the Labridae by incorporating additional species into our

105

analyses. Furthermore, while Coté’s review (2000) was exhaustive, additional taxa in the

106

Labridae have recently been identified as cleaners (e.g. Austrolabrus maculatus by Shepherd et

107

al., 2005). Thus, in order to formally examine cleaning diversity in labrids, a more current

108

literature search is warranted. Armed with an extensive phylogeny of 320 species and a more

109

comprehensive list of cleaners, we map the evolutionary history of cleaning in the Labridae.

110

Our investigation thus involved 1) inference of phylogenetic relationships between 320 labrids,

111

2) an extensive literature search to identify cleaner fishes within the group, and 3) stochastic

112

character mapping to identify evolutionary transitions to cleaning.

113 114

2. Materials and Methods

115

2.1 Phylogenetic Inference and Divergence Time Estimation

116

We reconstructed phylogenetic relationships using a molecular dataset that comprised four

117

mitochondrial (12S, 16S, COI, and CytB) and three nuclear gene regions (RAG2, TMO4c4, and

118

S7), with 5462 total base pairs. We obtained all sequences for 320 labrids and a 24-taxon

119

outgroup from GenBank (see Tables A.1-A.4 for accession numbers and information on genetic

120

sampling).

121

Our taxon sampling included 261 of 519 wrasses, 52 of 100 scarids, and seven of 12 odacids (see

122

Table A.5 and Figure A.1 for more details on sampling methods). While some sources (Nelson,

123

1994; Froese and Pauly, 2015) classify the Scaridae and Odacidae as distinct families, separate

124

from the Labridae, others have found these groups’ phylogenetic origins to be nested within the

125

Labridae (Clements et al., 2004; Westneat and Alfaro, 2005).

126

We thus consider the Labridae to include 631 total species (Froese and Pauly, 2015), and

127

therefore our genetic dataset contained 50.7% of nominal species (90% of nominal genera; Table

128

A.5, Figure A.1). Following previous studies, we used 24 outgroup taxa that comprised members

129

of the Pomacentridae, Cichlidae, Embiotocidae and other perciforms (Kazancioğlu et al., 2009;

130

Cowman and Bellwood, 2011).

131

We aligned each gene sequence separately using the built-in algorithm in Geneious 4.8.5. Each

132

alignment yielded high similarity to those found in previous studies (Kazancioğlu et al., 2009;

133

Alfaro et al., 2009). We then trimmed flanking regions that contained sequences from less than

134

60% of taxa. To identify the best-fitting model of nucleotide substitution for each gene, we used

135

jModelTest 2.0 (Darriba et al., 2012). In each case, we found the best-fit (assessed via AIC and

136

BIC scores) to be a GTR +I +Γ model, or a close variant thereof (Table A.6).

137

Using SequenceMatrix 1.7.8 (Vaidya et al., 2011), we concatenated nucleotide marker datasets

138

into a supermatrix. We partitioned this supermatrix by individual molecular markers and

139

performed a maximum-likelihood (ML) analysis in RAxML (Stamatakis, 2006). We ran a

140

bootstrap analysis under a GTR + Γ model with 1000 pseudoreplicates and used the phylogenetic

141

tree with the best likelihood score to guide further analyses (Supplementary File Tree B.1).

142

We then used BEAST 2.2.1 (Bouckaert et al., 2014) to simultaneously estimate topology, branch

143

lengths, and divergence times in a Bayesian framework. Using a relaxed log normal clock model

144

approach, we partitioned the supermatrix by sequence, and fit a separate model for each partition

145

based on our results from jModelTest.

146

To estimate divergence times, we placed informative parametric priors on nodes of the tree to

147

reflect the somewhat sparsely but available paleontological history of the group (Table 1). We

148

identified descendant members of each node based on the topology of the ML tree. The crown

149

group calibration was based on the K/T boundary extinction, as no full fossil specimens of

150

labrids have been found before this event, and our prior was informed by the 5-95% HPD for

151

crown labrids found by Near et al. (2013). While the fossil history of the Labridae is somewhat

152

sparse, previous studies have described six fossil taxa belonging to the group (see Table 1). For

153

these, we used the estimated age of the fossil as a hard bound on the minimum age of the node,

154

and priors were log-normally distributed. This information assimilated into our analysis of fossil

155

data and historical biogeographical events that have been used by previous studies (Kazancioğlu

156

et al., 2009; Alfaro et al., 2009; Cowman and Bellwood, 2011; Near et al., 2013). One key

157

difference is our extension of the prior related to the closure of the Isthmus of Panama (IoP),

158

which informs the divergence time between Halichoeres pictus and H. dispilus. Traditionally,

159

the closure of the IoP is estimated to have occurred ~3.1-3.5 MYA (Coates and Obando, 1996),

160

but a recent study uncovered evidence that suggests this closure may have occurred 13-15 MYA

161

(Montes et al., 2015). We took a conservative approach to parameterizing our priors for this

162

event by using a normally distributed prior with 3.1-15.9 MYA as the 5-95% intervals.

163

To ensure that each BEAST MCMC sampling converged on the target distribution, we

164

conducted five separate runs, each from a different random starting tree. We ran each MCMC

165

sampler for 200 million generations, sampling every 20,000 generations. We also ran a similar

166

analysis in which the supermatrix was not partitioned, but found that the MCMC runs had great

167

difficulty attaining stationarity, even after 75+ million generations. We assessed convergence via

168

Tracer 1.6 (Rambaut et al., 2014) by plotting likelihood vs. generation and estimating the

169

effective sample size (ESS) of each parameter.

170

Once we discarded the burn-in from each run (the first 15-20%), we combined runs via

171

LogCombiner 2.2.1 (Bouckaert et al., 2014). The combined set included 41,323 trees, which we

172

used to assemble the maximum clade credibility (MCC) tree in TreeAnnotator 2.2.1 (Bouckaert

173

et al., 2014). Within each BEAST run, the ESS of all parameters were generally >200, with the

174

lowest ESS still >100. After we discarded the burn-in and combined the results of all five runs,

175

the ESS of all parameters were >600, but the vast majority of parameters had ESS >2000.

176

2.2 Identifying Cleaner Fishes

177

We conducted an exhaustive literature search to gather information on cleaning behavior within

178

the Labridae. Explicit information for each cleaner species is available in Table A.7, and our

179

categorizations of cleaning follow a modification of Coté (2000). We categorized each species in

180

our genetic dataset to one of four states: 1) non-cleaner, 2) juvenile cleaner, 3) facultative

181

cleaner, or 4) obligate cleaner. Juvenile cleaners are those that clean predominately as juveniles

182

or sub-adults. Facultative cleaner species clean throughout ontogeny, although they do not rely

183

on cleaning behavior as their sole means of food acquisition. Obligate cleaners are notable for

184

depending on cleaning to obtain nearly all sources of food. We used these assigned categories in

185

our SIMMAP analyses (below). These states were designed to be discrete and non-overlapping.

186

We encountered some uncertainty in determining states for only one species (see Results and

187

Table A.7). Species for whom cleaning behavior had not been recorded in the literature were

188

simply assigned to the non-cleaner category.

189

2.3 Inferring the History of Cleaning

190

To simulate the evolutionary history of cleaning behavior on our phylogenetic trees, we used

191

stochastic character mapping (Nielsen, 2002; Huelsenbeck et al., 2003; Bollback, 2006). This

192

method enabled us to 1) sample simulated histories of cleaning evolution, and 2) estimate the

193

temporal patterns of transitions from non-cleaning to cleaning. In our analyses, we pruned all

194

trees to include just the 320 species of labrids; outgroup taxa were removed.

195

In analyzing transitions between states, we performed stochastic character mapping via

196

SIMMAP 1.5 (Bollback, 2006) on a sample of 10,000 trees from the posterior distribution of

197

trees provided by BEAST. To select parameters of the prior distributions for our mapping

198

analyses, we first performed an MCMC analysis using built-in functions in SIMMAP. We

199

sampled values for the parameters of the overall rate prior (i.e. the Г prior) using our MCC tree.

200

Using the prsummary() function (distributed with SIMMAP) in the R 3.1.3 environment (R Core

201

Team, 2014), we used samples from the posterior distribution of these parameters to find the

202

best-fitting parameters for the prior distribution. We then employed these “best fit” priors in

203

samplings of 10 stochastic character maps for each tree in the 10,000 posterior distribution trees.

204

This allowed us to incorporate uncertainty about the topology into our analyses (Huelsenbeck

205

and Rannala, 2003), while sampling character histories in proportion to their posterior

206

probabilities, given the tip states. We imported these maps into the R environment, and used the

207

describe.simmap() function in the phytools package (version 0.4.57; Revell, 2012) to summarize

208

our findings for each 10-map set. We then collected data across the (10,000 total) map

209

summaries to quantify the number and types of state changes, and the relative timing spent in

210

each state.

211

Because state changes (count data) were nearly all Poisson-distributed, we used a Poisson test to

212

test the hypothesis that each set of transitions was greater than zero, and used a Šidák correction

213

to account for multiple testing (Šidák, 1967). To assess whether certain transitions occurred more

214

frequently than others, we tested the hypothesis that mean numerical counts of different

215

transition types differed significantly via ANOVA. We excluded groups whose mean counts did

216

not significantly differ from zero (via the aforementioned Poisson test) from the ANOVA. Since

217

group variances were unequal, we used Tamhane’s T2 to make comparisons between all pairs of

218

groups.

219

To identify the timing of transitions from non-cleaning to cleaning, we employed two separate

220

analyses. In the first analysis, we used the character histories from the above SIMMAP sampling

221

and extracted the most probable state for each node in each tree. We then matched these node

222

states to their ages (in millions of years from the root), and computed summary statistics. In

223

particular, we recorded the 5th, 50th (median), and 95th percentiles of ages for each category, and

224

used the width of the 5th-95th percentile range to make comparisons among groups. This analysis

225

allowed us to compare ages of node states across a span of varying topologies, and using the 5th-

226

95th percentile range in ages reduced bias from extreme outliers. However, assigning each node

227

to its most probable state and then using the set of nodes as the only basis for comparison has

228

two limitations: 1) the general loss of resolution when each node is assigned to its most probable

229

state, and 2) the lack of incorporating changes that may occur along branches. These limitations

230

are especially key in situations where transitions consistently occur (i.e. across many mappings)

231

along relatively long branches; using information only from nodes may thus bias estimations of

232

where state transitions occur towards recency.

233

Thus, in a second analysis, we used SIMMAP to sample 1,000 stochastic character maps on only

234

the MCC tree. We then imported the maps into the R environment, and used the

235

mergeMappedStates() function in phytools to merge together the histories of all three cleaner

236

states. The resulting set of maps thus contained character histories for a binary set of states:

237

“non-cleaning” and “combined cleaning” in which the three cleaning categories were collapsed.

238

We then integrated information across this set of stochastic maps into a Bayesian posterior

239

probability (BPP) for each part of each branch in the tree via the densityMap() function in

240

phytools. We adopted this binary approach on the MCC tree due to the intractability of

241

simultaneously incorporating both topological uncertainty and multi-category complexity into

242

assigning BPPs of cleaning along mapped edges. To estimate the timing of transitions from non-

243

cleaning to cleaning, we identified the earliest point along each branch at which the BPP of being

244

in a cleaning state reached 0.5, and matched such points to their corresponding times. We chose

245

to quantify times corresponding to a BPP of 0.5 because subsequent (i.e. more recent) points

246

along a branch are more likely to be in a cleaning state than not. We only identified points on

247

branches along which there was an overall increase in the posterior probability of cleaning in

248

order to avoid quantifying information on secondary losses of the behavior. A similar analysis in

249

which we initially coded cleaning as a binary trait before running SIMMAP (i.e. bypassing the

250

need to employ the mergeMappedStates() function) yielded nearly identical results.

251

3. Results

252

3.1 Phylogenetic Inference

253

Our Bayesian analysis yielded a well-resolved phylogeny that was largely congruent with those

254

found in previous studies (Fig. 1, Supplemental File Tree B.2). We found the origin of crown

255

labrids to be approximately 62.08 MYA (95% HPD: 57.90-66.67 MYA), which is close to the

256

59.92 MYA (95% HPD: 54.4-66.7 MYA) estimate that Cowman and Bellwood (2011) found,

257

albeit slightly earlier. The ages of major groups in our MCC tree (i.e. those shown in Fig. 1) are

258

highly congruent with those found in previous studies (Alfaro et al., 2009; Kazancioğlu et al.,

259

2009, Cowman and Bellwood, 2011). Additionally, all BEAST runs converged on a MRCA time

260

for Halichoeres dispilus and H. pictus of 5.71 MYA. This time is closer to traditional estimates

261

for the closure of the IoP, and is too recent to fit the estimates of 13-15 MYA put forth by

262

Montes et al. (2015).

263

Within the hypsigenyines, the only major disagreement between our MCC and ML trees was the

264

placement of the MRCA of Achoerodus viridis and Pseudodax moluccanus. In the MCC tree, we

265

found the MRCA to be immediately sister to the group containing Bodianus and close allies with

266

a posterior probability of 0.84 (Fig. 1, Supplemental File Tree B.2). In contrast, our ML tree

267

placed this MRCA as sister to the odacines, Choerodon et al., and Bodianus et al. with bootstrap

268

support of 99 (Supplemental File Tree B.1). The placement of this MRCA on our ML tree is

269

more similar to the topology of the MCC tree produced by Cowman and Bellwood (2011).

270

Another source of disagreement between our MCC and ML trees was within the organization of

271

the pseudocheilines. Cowman and Bellwood (2011) identify Paracheilinus as a monophyletic

272

genus (though only two species were included), sister to the monophyletic Pteragogus. Our

273

analyses incorporated information for six additional Paracheilinus species. Whereas our ML tree

274

yielded a topology largely congruent with that of Cowman and Bellwood, our MCC tree suggests

275

Paracheilinus is part of a paraphyletic group that includes Malapterus reticulatus and placement

276

is closer to the novaculines.

277

The positions of various groups within the julidines is also unclear. Macropharyngodon,

278

Anampses, Pseudojuloides, Hemigymnus, the labricthynes, and Sagittalarva each occupy

279

different positions when comparing our MCC and ML trees. Adding to these complications is the

280

extreme polyphyly that Halichoeres and, to a lesser extent, Coris exhibit.

281

Though the organization of major groups (typically genera) in relation to each other is still

282

problematic, shallower nodes generally had higher support. Topology within each genus was

283

largely congruent between our MCC and ML trees.

284

3.2 Literature Search

285

Our literature search identified 58 species of labrids that are known to engage in cleaning

286

behavior (Table A.7). The vast majority, 43 species (74.1%), were reported to clean

287

predominately as juveniles. Less common are species that engage in cleaning facultatively

288

throughout ontogeny (11 species; 19.0%). The rarest strategy is obligate cleaning (8.6%), which

289

is exclusively found in all five Labroides species.

290

Only in one case (Labropsis polynesica) did we encounter uncertainty in assigning cleaning

291

status. There is little evidence in the literature of this species engaging in cleaning behavior

292

(Randall, 1981). We therefore conservatively coded L. polynesica as a non-cleaner in all

293

SIMMAP runs. We note, however, that since many other Labropsis species engage in cleaning

294

as juveniles (e.g. L. australis), it is possible that L. polynesica shares this characteristic.

295

Our final tally of 58 cleaners (which excludes L. polynesica) thus includes 7 species that have

296

been described to clean since Coté’s (2000) review: Austrolabrus maculatus (Shephard et al.,

297

2005), Bodianus anthiodes (Schiaparelli and Alvaro, 2009), Centrolabrus caeruleus (Azevedo,

298

1999), Halichoeres nigrescens (Sadovy and Cornish, 2000), Halichoeres penrosei (Coni et al.,

299

2007), Halichoeres radiatus (Grossman et al., 2006), Labrus bergylta (Steigen et al., 2015), and

300

Pseudocheilinus hexataenia (Sano et al., 1984). Of these 58 cleaners, 50 appeared in our genetic

301

dataset, and thus a substantial majority of labrid cleaners (86% of known species; 100% of

302

genera) was represented in our phylogenetic analyses.

303

3.3 Transitions between States

304

In stochastic character mappings performed on 10,000 posterior distribution trees (summarized

305

in Figure 2), we found that cleaning evolved from a non-cleaning state on average 28.10 times

306

(SD: 2.43) and was secondarily lost 6.50 times (SD 2.29). Our analysis also shed light on 8.05

307

transitions (SD: 2.34) between different cleaning states, the majority showing the pattern of

308

juvenile cleaner transitioning to facultative cleaner. In Figure 3, a single (and representative)

309

stochastic character map is superimposed on the MCC tree.

310

Within each transition type, the mean and median did not differ appreciably. Furthermore, the

311

median and mode were identical in each case. All mean counts for transitions were significantly

312

greater than zero, except for the following: 1) non-cleaner to obligate cleaner, 2) obligate cleaner

313

to non-cleaner, 3) facultative cleaner to obligate cleaner, and 4) obligate cleaner to facultative

314

cleaner. Notably, these transitions were extremely invariant across mappings on trees; they had

315

medians equal to zero, means near zero and standard deviations equal to or below 0.02 (Figure

316

2). We then excluded these four transition types in an ANOVA on mean transition counts, which

317

showed significant differences between groups (df: 7, F-ratio: 401,686.709; p-value < 0.001). A

318

Tamhane’s T2 test revealed significant differences between each pair of transitions (all p-values

319

< 0.001). Thus, all transitions represented in Figure 4 are not only significantly different from

320

zero, but also are significantly different from each other.

321

3.4 Timing of Transitions to Cleaning

322

After matching node states to their ages in mappings on 10,000 trees, we pooled together node

323

ages, which were measured in millions of years from the root (Figure 5). Given that across our

324

10,000 trees, the root age of the Labridae ranged from 54.91 to 71.54 MYA, each 5-95th

325

percentile range is firmly within the most recent third of the phylogenies. The 5th – 95th

326

percentile age range for each set of nodes was: juvenile cleaners: 45.13 to 67.38 MY; facultative

327

cleaners: 52.02 to 64.61 MY; obligate cleaners: 49.85 to 62.61 MY. Figure 5 also shows the

328

node ages for non-cleaners, for which the 5-95th percentile range was 24.83 to 68.63 MY.

329

Table 2 showcases the results of our within-branches estimation through 1,000 mappings on the

330

MCC tree alone, and Figure 6 shows graphical representations of the estimates in several clades.

331

We found that the earliest evolution of cleaning likely occurred close to or more recently than

332

18.36 MYA, with connecting nodes aged 21.26 and 17.89 MYA. The second-oldest age

333

estimate, leading to cleaning in Pseudodax moluccanus, is found in a completely separate part of

334

the labrid tree: the hypsigenyines (Figs. 1 and 3). This evolution provides an example of how

335

changes in states can occur within long branches, as the node-to-tip distance is 29.19 MY. Our

336

estimation places the evolution of cleaning within this branch around or after 15.20 MYA.

337

Among all of our estimations of cleaning evolution, the median was 6.11 MYA (Fig 6D), the

338

mean was 6.80 MYA, and the standard deviation was 4.91 MY.

339

4. Discussion

340

The present study aimed to determine when and how cleaning behavior evolved in the Labridae.

341

Through a genetic dataset comprising sequences from previously published molecular studies,

342

we first reconstructed the most complete phylogeny of the family to date. After identifying the

343

cleaning status of each taxon in our phylogeny, we inferred the history of cleaning behavior

344

using stochastic mapping methods.

345

4.1 Phylogenetic Analyses

346

Our study provides an extension of previous efforts to resolve relationships within the Labridae

347

(i.e. Cowman and Bellwood 2011) by including genetic data for 44 additional species. With

348

these additional taxa, we attained 55.41% coverage of genes for the ingroup, while sampling

349

slightly more than 50% of all nominal species. These metrics compare favorably to those of

350

Cowman and Bellwood (2011), which achieved 52.4% coverage of genes for 46% of nominal

351

species. Our increases in coverage of sequences and taxa reap the benefits of recent sampling

352

efforts (e.g. Hodge et al., 2011; Hubert et al., 2012; Steinke et al., 2012).

353

Our Bayesian analysis also converged on a MRCA time for Halichoeres dispilus and H. pictus of

354

approximately 5.7 MYA, reasonably close to findings from previous studies (Kazancioğlu et al.,

355

2009; Cowman and Bellwood, 2011). This estimate lends credence to traditional estimates of the

356

closure of the IoP around 3.1-3.5 MYA, and is in line with similar findings in geminate species

357

pairs of echinoids, crustaceans, molluscs, and other fishes (Lessios, 2008).

358

While the topology of our MCC and ML trees lacked congruence in specific areas, the majority

359

of disagreements were among the relative positions of major groups (i.e. the organization of deep

360

nodes). Within most genera, the topologies of the MCC and ML trees were far more similar and

361

well resolved; posterior probabilities and bootstrap support values were generally higher than

362

0.90 and 90%, respectively. The lack of consensus on the organization of major groups is a

363

problem that other studies have identified as well, especially concerning relationships among

364

genera in the julidines (Kazancioğlu et al., 2009; Cowman and Bellwood, 2011).

365

Furthermore, the lack of monophyly in some genera could be a result of sampling. For example,

366

in Paracheilinus, no nuclear markers were included in the analysis of this group, while only one

367

mitochondrial region (COI) was present. While mitochondrial markers typically provide good

368

resolution for estimating the topology of terminal branches, due to their fast mutation rates,

369

mtDNA are not ideal for estimating basal topology. No doubt, through increased species and

370

gene region sampling, future studies will be better equipped to resolve these organizational

371

problems. Regardless, since a number of our stochastic character mappings involved integrating

372

results from maps on 10,000 trees, we folded topological uncertainty into our analyses of

373

cleaning evolution.

374

4.2 The Organization and Inferred History of Cleaning

375

Through stochastic character mapping on 10,000 posterior distribution trees, we infer that

376

cleaning evolved in the Labridae 26 to 30 times. Most of these events occur within the julidines,

377

but many notable examples occur within the hypsigenyines and labrines as well. Cleaning is

378

conspicuously absent in the speciose scarids and some species-rich genera such as Cirrhilabrus,

379

Choerodon, Iniistius, and Paracheilinus. With 50 of 58 known cleaners in the Labridae

380

appearing in the present work, we feel confident that we inferred most of the major evolutions of

381

cleaning herein.

382

The remaining eight species that were not present in our genetic dataset are: Coris sandageri,

383

Halichoeres poeyi, Halichoeres penrosei, Halichoeres zeylonicus, Labropsis micronesica,

384

Labropsis xanthonota, Pseudolabrus luculentus, and Suezichthys aylingi. Of these species,

385

almost all have congeners in our phylogeny that clean. It is thus conceivable that some or many

386

of these missing species are sisters to cleaner congeners in the present study and thus could have

387

evolved cleaning through events that are already included in our analyses. Only in the case of S.

388

aylingi is the only congener present in our phylogeny (S. gracilis) a non-cleaner. Cleaning in S.

389

aylingi could be the result of an additional point of cleaning evolution, but would depend on the

390

position of S. aylingi in the labrid tree, especially if Suezichthys proves to be non-monophyletic.

391

Ultimately, until all of these missing taxa can be incorporated into future phylogenetic analyses,

392

the number of additional evolutions of cleaning beyond the 26-30 described herein remain

393

unknown. Cleaning in the Labridae may have involved up to eight additional independent

394

evolutions, but such assessments remain speculative.

395

In the present study, of all possible transitions from the non-cleaner to a cleaner state, the most

396

common by far involved transitions to the juvenile cleaner state, occurring 23 to 28 times on

397

mappings performed on 10,000 trees. The frequency of this repeated transition across various

398

parts of the tree accounts for most of the extant diversity of labrid cleaner fishes (Fig. 3).

399

Secondary losses were relatively rare (typically 4 to 8 times per tree).

400

Selection towards juvenile cleaning from a non-cleaner origin requires changes to the juvenile

401

life history stage, but its effects on adult morphology and behavior are presently ambiguous. It is

402

possible that these changes to the juvenile stage occur via similar (and perhaps simple) genetic

403

changes, while selection towards facultative or obligate cleaning might require changes that are

404

more extensive. This may account for the higher frequency to juvenile cleaner relative to

405

transitions to facultative or obligate cleaning from a non-cleaner origin. The lack of secondary

406

losses emanating from the facultative or obligate states adds credence to this hypothesis.

407

Whether a similar suite of morphological traits is found in juvenile individuals within this

408

cleaning state has not been extensively tested, although there is some evidence of this in the

409

Thalassoma clade (Baliga and Mehta, 2014). Baliga and Mehta (2014) show that juvenile

410

cleaners in this group consistently exhibit weak bite forces and possess jaws with low mobility

411

when compared to non-cleaner congeners. Upon reaching adulthood, differences in these

412

functional traits begin to vary. In the present study, we infer cleaning in Thalassoma to originate

413

from separate events, indicating that juvenile cleaners in this group are morphologically

414

convergent in the juvenile phase.

415

Transitions between different cleaning states are relatively infrequent, but appear to be important

416

in attaining facultative cleaner or obligate cleaner states. Notably, our analyses revealed that

417

direct jumps from non-cleaner states to obligate cleaner states (and vice versa) were virtually

418

nonexistent. Essentially all transitions to obligate cleaning originated from a juvenile cleaning

419

state. We estimate that the evolution of obligate cleaning was most likely a single event, perhaps

420

as depicted in Fig. 3. Additionally, we found that transitions to facultative cleaner states were

421

nearly twice as common from the juvenile cleaner state as from the non-cleaner state.

422

Essentially, juvenile cleaning in the Labridae presents a fascinating character state that seems to

423

bridge all others (Fig. 4).

424

Of course, the notion that obligate cleaning very likely evolved from a juvenile cleaning state is

425

perhaps unsurprising given that the monophyletic genus Labroides exclusively contains all five

426

species of obligate cleaners. The immediate sister to this group (the monotypic Larabicus) as

427

well as other closely related genera (Labropsis and Diproctacanthus) all contain juvenile

428

cleaners.

429

That the majority of transitions to facultative cleaning and all transitions to obligate cleaning are

430

preceded by a juvenile cleaner state lends us to hypothesize that facultative and obligate cleaning

431

evolved via a heterochronic process. Potentially, cleaning in the adult stage is merely an

432

extension of a juvenile feeding preference. Juvenile morphological traits may be carried over to

433

the adult stage, giving adult obligate and facultative cleaners neotenous characteristics, at least

434

for traits related to foraging, prey-capture, or cleaner-client recognition. Selective pressures to

435

retain morphological features that are conducive to cleaning may be rare, however, which may

436

explain the relative infrequency of these hypothesized cases of arrested development. These

437

hypotheses could be tested via a comparative study that examines the ontogeny of clades of

438

facultative or obligate cleaners, giving insight to morphological trajectories therein.

439 440

4.3 Estimating Temporal Patterns of Cleaning Evolution

441

Through our analyses, we find evidence that cleaning behavior evolved relatively recently. From

442

our within-branch estimation method, we found the earliest transitions to cleaning occurred

443

within the last 20 million years, while the majority of cleaning evolution occurred within the last

444

10 million years (Fig. 6D). These results generally hold up when accounting for varying

445

topological estimates of the tree: at the very least, the earliest cleaning states occurred no deeper

446

than a third of the length of the tree (Fig. 5). Based on our (62.08 MYA) and others’ estimates

447

(ranging from 55 to 68 MYA; Alfaro et al., 2009; Kazancioğlu et al., 2009; Cowman and

448

Bellwood, 2011) of the age of crown labrids, the patterns we observe point to the onset of

449

cleaning evolutions in the mid- to late-Miocene, and continuing through the Pliocene and

450

Pleistocene.

451

The Miocene also marks an especially important era in the diversification of reef fishes. Several

452

groups of reef-associated fishes, including tetradontiforms (Alfaro et al. 2007), chatetodontids,

453

pomacentrids, apogonids, and labrids (Cowman and Bellwood, 2011) appear to have undergone

454

rapid diversification during the mid- to late-Oligocene and early Miocene. In fact several major

455

labrid lineages, including the julidines, scarines, and some hypsigenyines (Bodianus et al.), show

456

significantly higher rates of cladogenesis during these epochs. In particular, Alfaro et al. (2009)

457

provide evidence that around 24 MYA, the julidine rate shifted to nearly double that of the

458

background labrid diversification rate, which led to more than 40% of non-scarine labrid

459

diversity. Furthermore, coral reefs themselves show patterns of diversifying and dominating

460

shallow-water marine systems in the late Eocene through early Miocene (Wood, 1999). Workers

461

have argued that the presence of reefs promoted diversification in fishes by providing habitats of

462

high productivity (Fraser and Currie, 1996), high spatial complexity (Gratwicke and Speight,

463

2005; Lingo and Szedlmayer, 2006), and high ecological complexity. Price et al. (2011) show

464

that reef-associated labrids exhibit markedly faster rates of trophic morphological diversification

465

and occupy a larger area of trophic morphospace than non-reef species. The tremendous

466

increases in the diversity of reef fishes in the Miocene may have provided cleaner fishes the

467

conditions to expand their potential clientele, thereby increasing the viability of this feeding

468

strategy. Given the patterns we observe in the present study, it appears that cleaning behavior

469

presents a possible example of ecological novelty supported and sustained by labrid

470

diversification on coral reef systems.

471

On the other hand, extant labrid cleaners are not constrained to occupying coral reef ecosystems.

472

Some taxa, including Oxyjulis californica, and several Symphodus cleaner species, occur in

473

temperate, seagrass- or kelp-dominated habitats. Ultimately, the factors that promoted or

474

constrained cleaning evolution in the recent past remain unclear, and may not be homogenous

475

across reef and non-reef habitats. Conceivably, an explosion in either ectoparasite taxonomic

476

diversity and/or population sizes during the late Oligocene or early Miocene could have provided

477

requisite opportunity for sustained directional selection on traits related to cleaning in labrids.

478

Additionally, shifts in climate occurring during these epochs may have contributed to rapid and

479

pronounced restructuring of marine ecological organization, creating space for novel dietary

480

strategies. Whether any or all of these factors contributed to the recurring evolution of cleaning

481

behavior throughout the Labridae can be addressed through future studies that incorporate

482

information on paleoclimatic events or invertebrate diversity.

483

4.4 Additional Remarks

484

One confound in our analyses is the simple assumption that a lack of observation of cleaning in

485

our “non-cleaner” state is a true representation of the behavioral repertoire of non-cleaner

486

species. In the 15 years since Coté’s review of cleaning, eight additional labrid species have been

487

identified to perform the behavior and none of these eight species was newly-described. This

488

indicates that future observations of cleaning are possible among putatively non-cleaner taxa.

489

While the feeding and social behaviors of many labrids have been extensively documented, this

490

group contains more than 600 taxa, many for which only a paucity of information exists. As

491

scientists continue to document field observations of labrid ecology, additional taxa (including

492

species herein classified as non-cleaners) may be identified to clean.

493

We issue an additional caveat about our within-branch time estimates of cleaning evolution. One

494

disadvantage of our approach is that as the BPP of being in a cleaning state approaches 0.5,

495

uncertainty reaches a maximum, since uncertainty is proportional to p*(1-p). However, we argue

496

that in integrating over our 1,000 character maps on the MCC tree, evolutionary transitions

497

between discrete states often occurred within short branch lengths. Of the 20 node-to-tip

498

transitions to cleaning, 12 occurred over branch lengths shorter than 10 million years (Fig. 6,

499

Table 2). Thus, the relative duration of remaining in a state of high uncertainty in many cases

500

was short. Here, our approach gave us a way to incorporate changes along a branch into our time

501

estimates (thus reducing the recency bias) and provided a comparative metric that could be

502

applied across all transitions to cleaning.

503

5. Conclusions

504

Through topological and temporal analyses of labrid evolution, we infer that cleaning evolved 26

505

to 30 times in various lineages, leading to an astounding extant diversity of cleaner fishes in this

506

group. Our estimates suggest that these evolutionary transitions began to occur in the mid- to

507

late-Miocene, with the majority occurring within the last 10 MY. Furthermore, we find that

508

direct transitions from non-cleaning to either facultative or obligate cleaning are relatively rare.

509

Transitions to these states are much more common from a juvenile cleaning state, which lends us

510

to hypothesize that some evolutions of facultative or obligate cleaning may involve

511

heterochrony.

512 513

Acknowledgements

514

We would like to thank Rita S. Mehta for insightful comments and discussion over the course of

515

this project. We also thank the instructors of the 2015 Bodega Bay Workshop in Applied

516

Phylogenetics for training and advice given to VB Baliga. Support towards the attendance of this

517

workshop was provided by the Society for Integrative and Comparative Biology Grant-in-Aid of

518

Research Award. We also wish to thank Peter T. Raimondi, Bruce E. Lyon, and Peter C.

519

Wainwright for helpful discussion on the evolution of cleaning in fishes. All fish illustrations in

520

Figure 3 were done by CJ Law.

521

Appendix

522

A. Supplementary Materials – Supplementary tables (A1-A7) and figures (A1) containing

523

additional data or findings

524

B. Supplementary Tree Files:

525

B.1 – Maximum Likelihood (ML) tree, obtained via RAxML

526

B.2 – Maximum Clade Credibility (MCC) tree, obtained via BEAST

527

References

528 529 530

Alfaro, M.E., Brock, C.D., Banbury, B.L., Wainwright, P.C., 2009. Does evolutionary innovation in pharyngeal jaws lead to rapid lineage diversification in labrid fishes? BMC Evol. Biol. 9, 255. doi:10.1186/1471-2148-9-255

531 532 533

Alfaro, M.E., Santini, F., Brock, C.D., 2007. Do Reefs Drive Diversification in Marine Teleosts? Evidence From the Pufferfish and Their Allies (Order Tetraodontiformes). Evolution. 61, 2104– 2126. doi:10.1111/j.1558-5646.2007.00182.x

534 535

Azevedo, J.M.N., Cepeda, H.I.R., Rodrigues, J.B., 1999. NOTES ON THE BIOLOGY OF Centrolabrus caeruleus AZEVEDO, 1999 (TELEOSTEI: LABRIDAE). Arquipél. Life Mar. Sci. 1999, 27–36.

536 537

Baliga, V.B., Mehta, R.S., 2014. Scaling patterns inform ontogenetic transitions away from cleaning in Thalassoma wrasses. J. Exp. Biol. 217, 3597–3606. doi:10.1242/jeb.107680

538 539 540

Baliga, V.B., Mehta, R.S., 2015. Linking Cranial Morphology to Prey Capture Kinematics in Three Cleaner Wrasses: Labroides dimidiatus, Larabicus quadrilineatus, and Thalassoma lutescens. J. Morphol. doi:10.1002/jmor.20425.

541 542 543

Barbu, L., Guinand, C., Bergmüller, R., Alvarez, N., Bshary, R., 2011. Cleaning wrasse species vary with respect to dependency on the mutualism and behavioural adaptations in interactions. Anim. Behav. 82, 1067–1074. doi:10.1016/j.anbehav.2011.07.043

544 545 546

Bannikov, A., Carnevale, G., 2010. Bellwoodilabrus landinii n. gen., n. sp., a new genus and species of labrid fish (Teleostei, Perciformes) from the Eocene of Monte Bolca. Geodiversitas. 33, 201–220. doi:10.5252/g2010n2a2

547 548 549

Bannikov, A.F., Sorbini, L., 1990. Eocoris bloti, a new genus and species of labrid fish (Perciformes, Labroidei) from the Eocene of Monte Bolca, Italy. Studi e Ricerche sui Giacimenti Terziari di Bolca, Museo Civico di Storia Naturale, Verona. 6, 133–148.

550 551 552

Barber, P.H., Bellwood, D.R., 2005. Biodiversity hotspots: evolutionary origins of biodiversity in wrasses (Halichoeres: Labridae) in the Indo-Pacific and new world tropics. Mol. Phylogenet. Evol. 35: 235– 253. doi:10.1016/j.ympev.2004.10.004

553 554 555

Bellwood, D.R., 1990. A new fossil fish Phyllopharyngodon longipinnis gen. et sp. nov. (family Labridae) from the Eocene, Monte Bolca, Italy. Studi e Ricerche sui Giacimenti Terziari di Bolca, Museo Civico di Storia Naturale, Verona. 6, 149–160.

556 557 558

Bellwood, D.R., Schultz, O., 1991. A review of the fossil record of the parrotfishes (Labroidei: Scaridae) with a description of a new Calotomus species from the Middle Miocene (Badenian) of Austria. Ann. Naturhist. Mus. Wien. 92, 55–71.

559 560

Bellwood, D.R., Wainwright, P.C., Fulton, C.J., Hoey, A.S., 2006. Functional versatility supports coral reef biodiversity. Proc. R. Soc. B Biol. Sci. 273, 101–107. doi:10.1098/rspb.2005.3276

561 562 563

Bertoncini, Á.A., Machado, L.F., Barreiros, J.P., Hostim-Silva, M., Verani, J.R., 2009. Cleaning activity among Labridae in the Azores: the rainbow wrasse Coris julis and the Azorean blue wrasse Centrolabrus caeruleus. J. Mar. Biol. Assoc. 89, 859. doi:10.1017/S002531540900040X

564 565

Bollback, J.P., 2006. SIMMAP: stochastic character mapping of discrete traits on phylogenies. BMC Bioinformatics 7, 88. doi:10.1186/1471-2105-7-88

566 567

Bronstein, J.L., 1994. Conditional outcomes in mutualistic interactions. Trends Ecol. Evol. 9, 214–217. doi:10.1016/0169-5347(94)90246-1

568 569 570

Bouckaert, R., Heled, J., Kühnert, D., Vaughan, T., Wu, C-H., Xie, D., Suchard, MA., Rambaut, A., & Drummond, A. J. (2014). BEAST 2: A Software Platform for Bayesian Evolutionary Analysis. PLoS Computational Biol. 10, e1003537. doi:10.1371/journal.pcbi.1003537

571 572 573

Clements, K.D., Alfaro, M.E., Fessler, J.L., Westneat, M.W., 2004. Relationships of the temperate Australasian labrid fish tribe Odacini (Perciformes; Teleostei). Mol. Phylogenet. Evol. 32, 575–87. doi:10.1016/j.ympev.2004.02.003

574 575 576

Coates, A.G. and Obando, J.A. 1996. The Geologic Evolution of the Central American Isthmus. p. 21. In: Jackson, J.B.C., Budd, A.F. and Coates, A.G. (eds.) Evolution and environment in Tropical America. University of Chicago Press, Chicago.

577 578

Coni E.O.C., Nunes, J.d.A.C.C., Sampaio, C.L.S., 2007. Halichoeres penrosei (Labridae), a sporadic cleaner wrasse. Mar. Biodivers. Rec. 1, e82. doi:10.1017/s1755267207008494

579 580

Coté, I.M., 2000. Evolution and Ecology of Cleaning Symbioses in the Sea. Oceanogr. Mar. Biol. an Annu. Rev. 38, 311–355.

581 582

Coté, I.M., Arnal, C., Reynolds, J.D., 1998. Variation in posing behaviour among fish species visiting cleaning stations. J. Fish Biol. 53, 256–266.

583 584 585

Cowman, P.F., Bellwood, D.R., 2011. Coral reefs as drivers of cladogenesis: expanding coral reefs, cryptic extinction events, and the development of biodiversity hotspots. J. Evol. Biol. 24, 2543–62. doi:10.1111/j.1420-9101.2011.02391.x

586 587

Darriba D, Taboada GL, Doallo R, Posada D. 2012. jModelTest 2: more models, new heuristics and parallel computing. Nat. Methods. 9, 772. doi:10.1038/nmeth.2109

588 589

Fraser, R.H., Currie, D.J., 1996. The species richness-energy hypothesis in a system where historical factors are thought to prevail: coral reefs. Am. Nat. 148, 138-159.

590

Froese, R. and Pauly, D. (2015). FishBase, Version (5/2015). www.fishbase.org.

591 592

Gorlick, D.L., Atkins, P.D., Losey, G.S., 1978. Cleaning Stations as Water Holes, Garbage Dumps, and Sites for the Evolution of Reciprocal Altriusm? Am. Nat. 112, 341–353.

593 594 595

Grossman, A., Sazima, C., Bellini, C., Sazima, I., 2006. Cleaning Symbiosis Between Hawksbill Turtles and Reef Fishes at Fernando de Noronha Archipelago, off Northeast Brazil. Chelonian Conserv. Biol. 5, 284-288. doi:10.2744/1071-8443(2006)5[284:csbhta]2.0.co;2

596 597

Grutter, A., 1996. Parasite removal rates by the cleaner wrasse Labroides dimidiatus. Mar. Ecol. Prog. Ser. 130, 61–70.

598

Grutter, A.S., 2010. Cleaner fish. Curr. Biol. 20, 547–549. doi:10.1016/j.cub.2010.04.013

599 600 601

Hodge, J.R., Read, C.I., van Herwerden, L., Bellwood, D.R., 2012. The role of peripheral endemism in species diversification: evidence from the coral reef fish genus Anampses (Family: Labridae). Mol. Phylogenet. Evol. 62, 653–63. doi:10.1016/j.ympev.2011.11.007

602 603 604 605

Hubert, N., Meyer, C.P., Bruggemann, H.J., Guérin, F., Komeno, R.J.L., Espiau, B., Causse, R., Williams, J.T., Planes, S., 2012. Cryptic diversity in Indo-Pacific coral-reef fishes revealed by DNA-barcoding provides new support to the centre-of-overlap hypothesis. PLoS One. 7, e28987. doi:10.1371/journal.pone.0028987

606 607

Huelsenbeck, J.P., Nielsen, R., Bollback, J.P., 2003. Stochastic Mapping of Morphological Characters. Syst. Biol. 52, 131–158. doi:10.1080/10635150390192780

608 609

Huelsenbeck, J.P., Rannala, B., 2003. DETECTING CORRELATION BETWEEN CHARACTERS IN A COMPARATIVE ANALYSIS WITH UNCERTAIN PHYLOGENY. Evolution. 57, 1237–1247.

610 611 612

Kazancioglu, E., Near, T.J., Hanel, R., Wainwright, P.C., 2009. Influence of sexual selection and feeding functional morphology on diversification rate of parrotfishes (Scaridae). Proc. R. Soc. B Biol. Sci. 276, 3439–46. doi:10.1098/rspb.2009.0876

613 614 615

Lessios, H.A. 2008. The Great American Schism: Divergence of Marine Organisms After the Rise of the Central American Isthmus. Annu. Rev. Ecol. Evol. Sys. 39, 63-91. doi:10.1146/annurev.ecolsys.38.091206.095815

616 617

Lingo, M.E. and S.T. Szedlmayer. 2006. The influence of habitat complexity on reef fish communities in the northeastern Gulf of Mexico. Environ. Biol. Fishes. 76:71-80.

618 619

Losey, G.S., 1972. The Ecological Importance of Cleaning Symbiosis The Ecological Importance of Cleaning Symbiosis. Copeia. 1972, 820–833.

620 621 622

Montes, C., Cardona, A., Jaramillo, C., Pardo, A., Silva, J.C., Valencia, V., Ayala, C., Perez-Angel, L.C., Rodriguez-Parra, L.A., Ramirez, V., Nino, H., 2015. Middle Miocene closure of the Central American Seaway. Science. 348, 5–8.

623 624 625 626

Near, T.J., Dornburg, A., Eytan, R.I., Keck, B.P., Smith, W.L., Kuhn, K.L., Moore, J.A., Price, S.A., Burbrink, F.T., Friedman, M., Wainwright, P.C., 2013. Phylogeny and tempo of diversification in the superradiation of spiny-rayed fishes. Proc. Natl. Acad. Sci. 110, 12738–12743. doi:10.5061/dryad.d3mb4

627

Nelson, J.S., 1994. Fishes of the world. Third edition. John Wiley & Sons, Inc., New York.

628 629

Nielsen, R., 2002. Mapping mutations on phylogenies. Syst. Biol. 51, 729–39. doi:10.1080/10635150290102393

630 631

Poulin, R., Grutter, A.S., 1996. Proximate Symbioses : Cleaning and Adaptive Explanation. Bioscience 46, 512–517.

632 633 634

Price, S.A., Holzman, R., Near, T.J., Wainwright, P.C., 2011. Coral reefs promote the evolution of morphological diversity and ecological novelty in labrid fishes. Ecol. Lett. 14, 462–9. doi:10.1111/j.1461-0248.2011.01607.x

635 636

R Development Core Team (2014). R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. Available at: http://www.R-project.org.

637 638

Randall, J.E., 1958. A Review of the Labrid Fish Genus Labroides, with Descriptions of Two New Species and Notes on Ecology. Pac. Sci. 12, 327–347.

639 640

Randall, J.E., 1981. Revision of the Labrid Fish Genus Labropsis with Descriptions of Five New Species. Micronesica. 17, 125–155.

641 642

Rambaut A, Suchard MA, Xie D & Drummond AJ (2014) Tracer v1.6, Available from http://beast.bio.ed.ac.uk/Tracer

643 644

Revell, L.J., 2012. phytools: an R package for phylogenetic comparative biology (and other things). Methods Ecol. Evol. 3, 217–223. doi:10.1111/j.2041-210X.2011.00169.x

645

Sadovy, Y., Cornish, A.S., 2000. Reef fishes of Hong Kong. Hong Kong University Press, Hong Kong.

646 647

Sano, M., Shimizu, M., Nose, Y., 1984. Food habits of teleostean reef fishes in Okinawa Island, southern Japan. University of Tokyo Bulletin, no. 25. University of Tokyo Press, Tokyo, Japan.

648 649 650

Schiaparelli, S., Alvaro, M.C., 2009. Incidental cleaning of crinoids by juveniles of Bodianus anthioides (Bennett, 1831) (Labridae) in the Red Sea. Coral Reefs. 28, 839–839. doi:10.1007/s00338-0090517-5

651 652 653

Schultz, O., Bellwood, D.R., 2004. Trigonodon oweni and Asima jugleri are different parts of the same species Trigonodon juleri, a Chiseltooth Wrasse from the lower and middle Miocene in the Central Europe (Osteichthyes, Labridae, Pseudodacinae). Ann. Naturhist. Mus. Wien. 105, 287–305.

654 655

Shepherd, S.A., Teale, J., Muirhead, D., 2005. Cleaning Symbiosis Among Inshore Fishes at Althorpe Island, South Australia and Elsewhere. Trans. R. Soc. South Aust. 129, 193–201.

656 657

Šidák, Z. K. (1967). Rectangular confidence regions for the means of multivariate normal distributions. J. Am. Stat. Assoc. 62, 626-633.

658 659 660

Stamatakis, A., 2006. RAxML-VI-HPC: maximum likelihood-based phylogenetic analyses with thousands of taxa and mixed models. Bioinformatics. 22, 2688–90. doi:10.1093/bioinformatics/btl446

661 662 663 664

Steigen, A., Karlsbakk, E., Plarre, H., Watanabe, K., Øvergård, A.C., Brevik, Ø., Nylund, A., 2014. A new intracellular bacterium, Candidatus Similichlamydia labri sp. nov. (Chlamydiaceae) producing epitheliocysts in ballan wrasse, Labrus bergylta (Pisces, Labridae). Arch. Microbiol. 197, 311–318. doi:10.1007/s00203-014-1061-4

665 666

Steinke, D., Zemlak, T.S., Hebert, P.D.N., 2009. Barcoding nemo: DNA-based identifications for the ornamental fish trade. PLoS One. 4, e6300. doi:10.1371/journal.pone.0006300

667 668

Vaidya, G., Lohman, D.J., Meier, R., 2011. Cladistics multi-gene datasets with character set and codon information. Cladistics. 27, 171–180.

669 670 671

Wainwright, P.C., Bellwood, D.R., Westneat, M.W., Grubich, J.R., Hoey, A.S., 2004. A functional morphospace for the skull of labrid fishes: patterns of diversity in a complex biomechanical system. Biol. J. Linn. Soc. 82, 1–25. doi:10.1111/j.1095-8312.2004.00313.x

672 673 674

Waldie, P.A., Blomberg, S.P., Cheney, K.L., Goldizen, A.W., Grutter, A.S., 2011. Long-Term Effects of the Cleaner Fish Labroides dimidiatus on Coral Reef Fish Communities. PLoS One. 6, e21021. doi:10.1371/journal.pone.0021201

675 676

Westneat, M.W., Alfaro, M.E., 2005. Phylogenetic relationships and evolutionary history of the reef fish family Labridae. Mol. Phylogenet. Evol. 36, 370–90. doi:10.1016/j.ympev.2005.02.001

677

Wood, R., 1999. Reef Evolution. Oxford University Press, New York.

678 679

Youngbluth, M.J., 1968. Aspects of the Ecology and Ethology of the Cleaning Fish, Labroides phthirophagus (Randall). Z. Tierpsychol. 25, 915–932.

680

681 682

Figure Captions

683

Figure 1. Maximum clade credibility tree from Bayesian MCMC analyses. Tree is simplified

684

to show only the relationships between major groups within the Labridae. Blue bars show 95%

685

HPD intervals for node ages. Nodes are labeled with support values in the following order:

686

Bayesian posterior probabilities (BPP)/bootstrap support (BS). Dashes (-) indicate no nodal

687

support in the ML tree. Unlabeled nodes have Bayesian posterior probabilities (BPP) ≥ 0.90 and

688

bootstrap support (BS) ≥ 90%. Triangles at the tips indicate that a clade is collapsed and

689

simplified. Tip labels denote genus or group names with the proportion of species sampled in this

690

study in parentheses. The genera Halichoeres and Coris are polyphyletic, and proportions are not

691

shown for these groups due to the difficulty of assigning taxon placements for non-sampled

692

species. Vertical text and box shading delineate major groups.

693 694

Figure 2. State changes in stochastic character mappings on 10,000 Bayesian posterior

695

distribution trees. Summary statistics and histograms of state changes in stochastic character

696

mappings performed in SIMMAP on 10,000 trees from the posterior distribution obtained via

697

BEAST. Row names indicate the “from” state, and column headings list the “to” state. Within

698

each box, sample means (
̅ ) are listed first, sample standard deviations (s) second, and sample

699

medians (Md) third. Modes are not listed, as they are equivalent to medians. State changes that

700

amounted to gains in cleaning behavior (i.e. all non-cleaner to cleaner transitions) are

701

summarized in All Gains, while losses (i.e. from cleaner to non-cleaner) are summarized in All

702

Losses. All Cleaner Transitions integrates information across all transitions between different

703

cleaner states. Bold sample means are significantly greater than zero, as determined by Šidák-

704

corrected Poisson tests.

705 706

Figure 3. Maximum clade credibility tree with simulated history of cleaning. A single

707

stochastic character map out of 1,000 is laid over the topology of the MCC tree. Grey colors

708

correspond to non-cleaners; orange to juvenile cleaners; green to facultative cleaners; and purple

709

to obligate cleaners. Circles at terminal tips indicate cleaning status for extant taxa, while branch

710

colors depict simulated history. On each node, a pie chart shows the relative Bayesian posterior

711

probability of each character state. Key nodes are numbered (refer to in Table 2 for additional

712

details). Illustrations of fishes encircling the phylogeny are not systematically representative of a

713

particular age class or sex.

714 715

Figure 4. Mean number of state changes in stochastic character mappings on 10,000

716

Bayesian posterior distribution trees. A visual representation of the relative frequencies of

717

state changes summarizing simulated character histories on 10,000 trees. The widths of the

718

arrows to/from each state are approximately proportional to the mean count of transitions

719

between connecting states. Transition counts that were not significantly different from zero are

720

not depicted.

721 722

Figure 5. Distributions of node ages in stochastic character mappings on 10,000 Bayesian

723

posterior distribution trees. After stochastic character mapping was applied to each of 10,000

724

trees from the Bayesian posterior distribution of trees, each node within each tree was assigned

725

to its most probable state. The age of each node (measured in millions of years from the root)

726

was extracted from its tree. Node ages are pooled and represented in a histogram. Horizontal bars

727

below the histogram show values for the 5th, 50th (median), and 95th percentile of ages within

728

each category. Abbreviations: NON – non-cleaner; JUV – juvenile cleaner; FAC – facultative

729

cleaner; OBL – obligate cleaner

730 731

Figure 6. Estimating temporal patterns of cleaning evolution within branches. (A-C):

732

Graphical representations of the Bayesian posterior probabilities (BPP) of cleaning in three

733

example clades: A) Symphodus et al. (facultative & juvenile cleaners), B) Bodianus et al.

734

(juvenile cleaners) and C) the labrichthynes (juvenile & obligate cleaners). On the topology of

735

each clade, darker red colors indicate higher BPPs of being in a cleaning state for a node or a

736

portion of a branch. On the accompanying plots, BPPs and ages for nodes and tips are shown

737

using dark grey circles, and BPPs along branches are presented in light grey. Each blue dashed

738

line indicates the time at which a branch attains a BPP of 0.50. Such times were only estimated

739

along branches in which there was an overall increase in the BPP of cleaning, in order to avoid

740

estimating timings of secondary losses. (D): Timing of these events across all clades in the 320-

741

speces MCC tree, with a boxplot and summary statistics. See Table 2 for further details.

742

743

744 745 746 747 748 749 750 751

Table 1: Fossil and biogeographic information used for divergence time estimation in BEAST Group

Fossil or Event

Age (MY)

Distribution

Prior (5-95%)

Source

Root (crown Labridae)

K⁄T boundary

66.2∗

Normal

62.2-70.3

Near et al., 2013

Hypsigenyines

Phyllopharyngodon longipinnis

50†

Lognormal

51.5–63.1

Bellwood, 1990

Labridae (-Hypsigenyines)

Eocoris bloti

50†

Lognormal

51.5–63.1

Bannikov & Sorbi

Bellwoodilabrus landinii

50†

51.5–63.1

Bannikov & Carne

Pseudodax moluccanus ⁄Achoerodus viridis

Trigondon jugleri

14†

Lognormal

15.1–44.0

Schultz & Bellwoo

Calotomus⁄Sparisoma

Calotomus preisli

14†

Lognormal

15.1–44.0

Bellwood & Schul

Halichoeres dispilus⁄H. pictus

Isthmus of Panama

9.5‡

Normal

3.1-15.9

Coates & Obando Barber & Bellwoo Montes et al., 201

Bolbometopon muricatum ⁄Cetoscarus bicolor

Bolbometopon sp.

5†

Lognormal

6.1–11.1

Bellwood & Schul

We placed parametric prior distributions on the MRCA of lineages as specified above. ∗ The K/T boundary is often used as a guide for the estimation of the age of the crown group; no fossil labrids have been found before this event † Minimum age for the fossil ‡ Estimation for this biogeographic event incorporates information from traditional sources that estimate the closure of the Isthmus of Panama to have occurred 3.1-3.5 MYA as well as a recent study that presents evidence for the closure instead occurring 13-15 MYA.

752

Table 2: Estimated ages of transitions to cleaning along branches

Group Labrichthynes

Taxoni Node 1

Pseudodax

Node 21

Bodinanus et al.

Node 15

Halichoeres V et al.

Node 3

Bodinanus et al.

Node 18

Halichoeres V et al.

Node 5

Labrus

Node 7

Bodinanus et al.

Node 21

Coris et al.

Node 26

Bodinanus et al.

Node 20

Pseudocheilinus

Node 23

Bodinanus et al.

Node 17

Symphodus et al.

Node 9

Thalassoma et al.

Node 29

Halichoeres IV et al.

Node 28

Ctenolabrus Symphodus et al. Symphodus et al. Thalassoma et al. Thalassoma et al. Austrolabrus Symphodus et al. Thalassoma et al. Coris et al. Thalassoma et al. Thalassoma et al. Labrus Thalassoma et al.

Node 14 Node 12 Node 24 Node 33 Node 36 Node 24 Node 11 Node 31 Node 27 Node 34 Node 35 Node 8 Node 32

Agei 21.2 6 29.1 9 19.5 7 19.7 5 18.0 7 13.6 6 11.1 3 15.4 5 19.0 2 13.0 2 15.3 2 14.3 1 12.1 9 10.2 9 11.1 4 9.96 5.21 8.28 7.42 5.97 5.26 4.75 3.82 3.43 2.36 2.02 0.79 0.79

Taxonj Node 2

Estimate 18.36

Pseudodax moluccanus

Agej 17.8 9 0

Node 16

7.53

14.42

Node 4

3.12

13.25

Node 19

13.14

Node 6

11.4 9 7.59

Labrus mixtus

0

9.78

Bodianus anthiodes

0

9.74

Coris picta

0

9.51

Bodinaus scrofa

0

8.30

Pseudocheilinus hexataenia

0

8.26

Bodianus speciosus

0

8.03

Node 10

0.91

6.87

Node 30

3.07

6.64

Halichoeres nigrescens

0

5.58

Ctenolabrus rupestrus Node 13 Pseudolabrus miles Thalassoma lunare Thalassoma pavo Austrolabrus maculatus Symphodus tinca Thalassoma lutescens Coris julis Thalassoma cupido Thalassoma lucasanum Labrus berygylta Thalassoma duperrey

0 4.52 0 0 0 0 0 0 0 0 0 0 0

5.21 4.80 4.23 3.73 3.02 2.69 2.39 1.94 1.90 1.20 1.01 0.73 0.40

15.20

10.04

753 754 755 756 757 758

Group names correspond to those in Fig 1. Node names listed in either Taxon column correspond to labeling in Fig 3. All ages are in millions of years before the present. Estimated ages indicate the earliest time along branches between each pair of taxa at which the Bayesian posterior probability (BPP) reaches 0.50. Such ages were only estimated along branches in which there was an overall increase in the BPP of cleaning, in order to avoid estimating timings of secondary losses.

759 760 761 762 763 764 765 766

Highlights: • • • • •

We infer phylogenetic relationships between 320 species of labrids Stochastic character mapping suggests cleaning evolved at least 26–30 times The transition from non-cleaner to juvenile cleaner is the most common state change Extant facultative or obligate cleaners may have evolved via heterochrony The earliest cleaners likely appeared in the late-Miocene

Figure 1 /-

Odacidae (7/12)

0.15/-

0.84/-

/71 /70

0.67/77

Pseudocheilinus I (4/7) 0.87/79

Pseudocheilinus II (1/7) 0.88/44

0.84/-

Cirrhilabrus (7/48)

Pseudocheilines

0.76/-

Scaridae

Pteragogus (2/10)

Labrines

/80

Choerodon et al. (11/25) Achoerodus (1/2) Pseudodax (1/1) Bodianus et al. (24/58) Lappanella (1/2) Labrus (4/4) Tautoga (1/1) Ctenolabrus (1/1) Acantholabrus (1/1) Tautogolabrus (1/1) Symphodus et al. (13/13) Cheilines (13/21) Leptoscarus (1/1) Calotomus (2/5) Cryptotomus (1/1) Nicholsina (2/2) Sparisoma (12/15) Bolbometopon (1/1) Cetoscarus (1/2) Hipposcarus (1/2) Chlorurus (7/18) Scarus (24/52)

Hypsigenynes

Outgroup Anchichoerops (1/1) Lachnolaimus (1/1)

Paracheilinus et al. (9/18)

0.45/0.45/-

0.60/36 /63

/21

/77

/29 /-

//-

//0.50/0.48 /-

0.80/-

0.42/0.77//-

70

60

50

40

30

20

10

0 MYA

Julidines

Novaculines (11/37) Cheilio (1/1) Doranotus (1/1) Suezichthys (1/12) Pseudolabrines (14/24) Stethojulis (6/10) Coris et al. (17) Halichoeres I (1) Leptojulis (1/5) Parajulis (1/1) Halichoeres II (3) Halichoeres III (1) Macropharyngodon (8/12) Halichoeres IV et al. (28) Anampses (11/12) Pseudojuloides (3/11) Halichoeres V et al. (14) Ophthalmolepis (1/1) Hemigymnus (2/3) Labrichthynes (10/14) Halichoeres VI (1) Sagittalarva (1/1) Thalassoma et al. (28/30)

0.74/-

Figure 2

Juvenile Cleaner

NonCleaner

x: 25.42 s: 2.43 Md: 25

8

NonCleaner

%4

Facultative Cleaner x: 2.68 s: 1.63 Md: 3

24

0 10 20 30 40 50 # of transitions

x: 0 s: 0 Md: 0

100

8

00

x: 28.10 s: 2.26 Md: 28

%4

% 50

% 12

0

Obligate Cleaner

10 20 30 40 50 # of transitions

0

x: 5.35 s: 2.21 Md: 5

36

0 10 20 30 40 50 # of transitions

0

0 10 20 30 40 50 # of transitions

All Gains x: 6.29 s: 2.30 Md: 6

16

Juvenile Cleaner

18

% 8

% 9

% 18

0

0

0

0 10 20 30 40 50 # of transitions

84

Facultative Cleaner

x: 0.22 s: 0.46 Md: 0

x: 1.35 s: 1.33 Md: 1

36

0 10 20 30 40 50 # of transitions

100

% 42

% 18

% 50

0

0

0

0 10 20 30 40 50 # of transitions

100

Obligate Cleaner

0 10 20 30 40 50 # of transitions

x: 1.03 s: 0.68 Md: 1

x: 0 s: 0 Md: 0

0 10 20 30 40 50 # of transitions

80

x: 0.29 s: 0.48 Md: 0

0 10 20 30 40 50 # of transitions

100

% 50

% 40

% 50

0

0

0

0 10 20 30 40 50 # of transitions

16

%8

x: 6.50 s: 2.29 Md: 6

0

0 10 20 30 40 50 # of transitions

All Losses

0 10 20 30 40 50 # of transitions

x: 0.02 s: 0.02 Md: 0

x: 0.01 s: 0.02 Md: 0

0 10 20 30 40 50 # of transitions 16 16

%8 % 8

x: 8.05 s: 2.34 Md: 8

0 0 0

10 20 30 40 50 0 10# of 20transitions 30 40 50 count

All Cleaner Transitions

Figure 3

Wetmorella albofasciata Cheilinus fasciatus Oxycheilinus bimaculatus Cheilinus trilobatus Cheilinus oxycep halus Cheilinus Leptoscaruchlorourus s vaigiensis Sparisom Sparisom a strigatum a creten Sparis se Spari oma radian Spa soma tuiu s Sparirisoma ato piranga Spa soma a marium Spa risoma urofrena Spa risoma frondos tum Spa risoma chrysop um terum Sp risom rubri Sp arisom a axil pinne Nic arisom a viridlare Nic holsin a am e Cr hols a us plum Ca ypto ina d ta C loto tomu entic Hipalotommus s s rose ulata S po us pin us S caru sca ca iden S caru s p rus rolin s Sc caru s sc sittac longicus S ar s h hle us eps S car us yps ge S car us flav elo li Sc caru us r spin ipec pter a S s iv us tora us lis S ca rus ov ula S ca rus g ice tus Sc car rus qu lobic ps Sc ar us fe oy ep s i u S a s c S ca ru s ha tivu Sc ca rus s is taen me s ar rus g er iop leo us c ua i te n o c ru fre el am s na est a tu inu ia s s

Wetmorella nigropinnata Cheilinus undulatus Epibulus insidiators bicu Oxycheilinus cele digramma Oxycheilinusunifasciatus us ctatus Oxycheilin un op ardi es caerule Anamps Anampses lenn icus raph es geoges cuvier Anamps us mps in na A femin pses halus Anam hrysocep eatus c n des pses ses li Anam Anamp meleagri icus a pses oguin istii Anam es ne es tw ns mps nampss elega tus A Ana e adus us mp s il Ana hoeres otosp tus n ta c Hali oeres s bivitiensiss re ch sil tu Hali lichoe s bra radiaalus Ha oere eres eph noti oc ar lsi lich ho Ha Halic cyan res g icho eyi s oe s n po s e oer lich ere es pilu s lich Ha licho hoer dis ictu lis s Ha Ha lic re s p cia s a e H o ere so ctu lich ho s in ica i Ha alic oereemic iforntava s H ich s al a nu i l s c s i s Ha ere lis ide ras ern s ho yju lo ce v ru is lic Ox oju es s senthu tral ca Ha ud oid de xa us si lor e i l e s Ps doju julo hus is a lyn bico atu lis i o t s o eu ud an op p s id ra Ps se tac abr psis oide dim cto P c L o r e o br ab es s p pr La L roid ide b ro La ab L

Di

La L br La ab oid r ro e La abi ide s p c h He He bric us s ru thi m m ht qu b ro Go igy igym hys ad rola pha m n u r il b g m n Th Th n in ia u ala al G pho us mus f ilin ea tus s ss ass om sus e asc ea tus om om ph c la ia tu p a Th a sa a a osu er ter tus s Th ala n sc s ule us ala ss T ss cta e va u Th oma hala om ehensio rius s a n a Th lass amb ssom newlena is ala om lyc a to e s Th som a luc epha pavoni ala a r as lu Tha Tha sso ob an m e u Thalasso lasso ma cur tsonm las ma ma pid i Thasoma purpu loxumo las trilo reu T T hala som ba m Tha halass ssom a viretum lass a o oma ma du lunarns T Thala halass geniv perre e ittatu y ssom oma ru a Thala gram eppe m Thalassoma h maticu llii m Thala ssoma luebraicum ssom tesce Thalass Thalassoa hardwic ns ke ma ja oma Thalas quinquevittnsenii Thalassosoma bifasciaatum tum ma no Thalassomronhanum Thalassom a ballieui a se Halichoeres ptemfasciata maculipinna Sagittalarva inor Ophthalmolepis lineonata lata Coris aurilineata Coris pictoides Halichoeres prosopeion Halichoeres solorensis Halichoeres tenuispinis Halichoeres arguss uru Halichoeres leuc nigrescens Halichoeres binotopsis s ere ho Halic aceus es papilion tus Halichoer eres margina us Halicho eres melanur ondi Halicho eres richm fieldi Halichoeres brownensis o tu Halich Coris baropterus a s chlo stigmus e hoere podo Halic hoeres rgaritac sus Halic eres mas nebuloiatus s re ho min Halic alichoe oeres cellatudia H u ch bio Hali oeres res clasimus ch oe tis tus Hali Halich orna osme us m s ere es c po us cho oer sma rys is Hali Halich mela res chs irid us es oe ere nth s oer lich ho xa illu lich Ha Halic leuco s lap r titusis Ha e a s res er ip n ti oe cho n b ose oa s lich Hali godo negr n chttatu us Ha n do u t y n ary do go og rna ro ph go yn an o off ris cro ryn har cy on ge ag eri Ma pha crop don ngoddon ele kuit nusis cro Ma yngo ary ngo on m on tula lar us Ma ar ph ry od od or pu at ii ph cro ha ng ng s h ca cul feld ra cro Ma crop ary hary ere es s ma rtz leu h r i a p Ma Ma rop crop licho oe s tr s h no h re re a ac a a M M H alic oe oe s cy H ich ich uli l l j Ha Ha pto Le

4

5 3 2

36

35 34

6

1

33

32

s s u tu ce ia la id io m rov ban i d b b r us r u o lo i os ar us gh ico en th Sc car rus s tr orst r gna S ca ru s f ige io is s S ca ru s n ras inn du S ca ru p tip rdi a is S ca rus al so dem ns s S ca rus rus oe ane ino es S ca ru us jap orh toid S hlo rur us icr tra C hlo rur s m apis rsi C hlo ruru s c we ri C hlo uru s bo eke r m C hlor uru ble icolo ricatu C hlor urus s b mu C hlor caru pon iata C etos eto fasc C lbomnella tus Bo ppa mix is La brus virid la s La brus meru lta ercu La brus ergy elanoc La brus b us m oletus La mphod rus ex ni lo b Sy ntrola s bail rleini Ce p ho du s dode Sym phodu tinca Sym phodus cinereus Sym phodus stratus Sym phodus ro iterraneus Sym phodus med latus Sym odus ocel Symph odus roissali Symph odus melops Symph brus trutta Centrola s caeruleus Centrolabru Tautoga onitis Ctenolabrus rupestris Tautogolabrus adspersus Acantholabrus palloni Anchichoerops natalensis Lachnolaimus maximus Odax pullus Odax cyanoallix Heteroscarus acro ptilus Olisthops Siphonogncyanomelas Neooda athus argyrophan es Haletta x balteatus Xipho semifascia Choercheilus typu ta Choe odon monoss tigma Cho rodon azu Cho erodon fa rio Cho erodon sciatus Cho erodon venustus Ch erodo cepha Ch oerod n scho lotes Ch oerod on rub enleinii Chooerodoon cau escens Bo ero n oli terom Bo dian don a gaca a Po dian us un ncho nthus De lylep us ox imacu rago latu D cod ion yce C ecod on p russ phalu s s B lept on uell elli Boodianicus pmela aris s B dia us ar m B odi nu bim rae a Bo odia anus s ruf acu B di nu p us latu s B od anu s d ulc B od ian s ipl hell Bo odiaianu us e spec otae us B d n s c io n Bo odi ian us per lanc sus ia B d a u s d h B od ia nu s b ola itio eri Se od ian nus s di ilun tus m ian us di cty ula ic us m an nn tu os a e a a s sy xi so ph lla tho us ris ra x pu lc he r

8 10

7

31

11

9

29

12 13

30

14

28

15

22

17 18 21

26

23

25

24

19

s fa tu s s ro ea idu s sc ilin ok ide anu us tr ny io cc an us ta th lu s is d i i a n u s a n mo i r i d i u s n s e n i a Bo d ian us ax v pt ine ta s Bo od ian od dus cry bo tetra llatu nia B od ud ro us am s ce tae B se oe og us linu s o xa dus P ch rag og ei inu he ani nia s v A e g ch il ae Pt tera udo che ilinu us e ctot P se udo che ilin s o ki s P se do che ilinu boc sali b or P u o e e u h Ps seud doc rus l flavid torum s P seu ilab us cot itu inatus P irrh ilabr us s xquis arg s e rim trali C rh br Cirirrhilaabruss rub riven ura s C rhil bru rub ople iatu Cirirrhila brus cyan mitaen C rhila brus s he on s Cir rhila eilinu walt entosu Cir ach ilinus lam s Parrache ilinus ficyaneu eri Pa rache ilinus ccosk s Pa rache inus m vianali fla Pa heil s nteri c u Para cheilin us carpelatus Para cheilin s angu s Para eilinu ulatu idotus c h ti c Para rus re acrolep te p Mala culoides m taeniourus Nova ulichthys s Novachtys splendenps dice Xyric un m Xyrichtys novacula Xyrichtyss martinicensis Xyrichty o Iniistius pav Iniistius verrens Iniistius aneitensis Cymolutes torquatus Cymolutes praetextatus Cheilio inermis

Pa ra ju Ho lis lo g Ho y C poe m or c lo n gy o i i m su C s at lept no s o la e su an ris nt rus Co C s nu j ica C C o r is or i do la ul i s Co oris ris f bu s ay liatutus ris ca lav lbif gu s do ud ovi ron la rs im tt s Ps Co Cori om acu ata eu s ac la do Pse r i C s u Ps coris do oris ga cuvi ula eu c e Ps doc aura oris formimar ri eu do oris ntio blee osad cor he fas ke is y tero cia ri Ste t a p a Ste tho m tho juli Cor ash tera ju s is ir Ste lis ba albo pic oi Ste thoju ndan vittatata li th ojuli s ba ensis Ste th lt s oju Ste tril eata Au thoju lis inte ineata Eup strolabrulis strig rrupta etric iv hthy s macu enter s latu No an Pictilatolabrus gustipess tetric brus u N la Pseud otolabrus ticlaviuss olabru fucico s la Pseud olabrueoethinus s siebol Pseud Pseudo olabrus ga di yi Pseudolablabrus fuentesi Notolabrusrus guentheri gymnogen is Notolabru Pseudolabrus s parilus biseriali Pseudolabrus miless Suezichthys gracilis Doratonotus megalepis

27

20

16

Figure 4

NonCleaner

2.7

25.4 0.2

Facultative Cleaner

6.3

5.4

Juvenile Cleaner 1.4

0.3 1.0

Obligate Cleaner

Figure 5

Percent

12

6

0 5th

median

95th

NON JUV FAC OBL

0

10

20

30

40

Time from Root (MY)

50

60

70

Figure 6

A

B Symphodus et al.

Bodianus et al. Clepticus parrae Bodianus bimaculatus Bodianus rufus Bodianus pulchellus Bodianus diplotaenia Bodianus speciosus Bodianus eclancheri Bodianus perditio Bodianus solatus Bodianus bilunulatus Bodianus dictynna Bodianus diana Bodianus mesothorax Bodianus axillaris Semicossyphus pulcher Bodianus scrofa Bodianus trilineatus Bodianus tanyokidus Bodianus anthioides 1

Symphodus melanocercus Centrolabrus exoletus Symphodus bailloni Symphodus doderleini Symphodus tinca Symphodus cinereus Symphodus rostratus Symphodus mediterraneus Symphodus ocellatus Symphodus roissali Symphodus melops Centrolabrus trutta Centrolabrus caeruleus Tautoga onitis Ctenolabrus rupestris Tautogolabrus adspersus Acantholabrus palloni 1

0.5

20

15

10

5

0

BPP of Cleaning

0.5

0

25

20

Time (MYA)

15

10

5

0

BPP of Cleaning

0

Time (MYA)

C

D The Labrichthynes Hemigymnus melapterus Hemigymnus fasciatus Labrichthys unilineatus Larabicus quadrilineatus Labroides rubrolabiatus Labroides phthirophagus Labroides pectoralis Labroides dimidiatus Labroides bicolor Labropsis polynesica Labropsis australis Diproctacanthus xanthurus 1

18.36

20

9.76

15

10 Time (MYA)

0.5

30

25

20

15

10

Time (MYA)

5

0

0

BPP of Cleaning

6.11

5

2.54 0.40

0

*Graphical Abstract (for review)

Juvenile Cleaner 1.0

25.4 0.3

6.3

NonCleaner

5.4

1.4

0.2 2.7

Facultative Cleaner

Obligate Cleaner