Received: 20 July 2017 | Revised: 11 May 2018 | Accepted: 29 May 2018

DOI: 10.1111/geb.12780

RESE ARCH PAPER

The global structure of marine cleaning mutualistic networks

Juan Pablo Quimbayo1,2* | Mauricio Cantor1* | Murilo S. Dias3 | Alexandra S. Grutter4 |

Simon Gingins5,6 | Justine H. A. Becker4 | Sergio R. Floeter1

1
Departamento de Ecologia e
Zoologia, Universidade Federal de Santa Abstract
Catarina, Florianópolis, Brazil Aim: We studied the underlying biotic and abiotic drivers of network patterns in ma‐
2
Reef System Ecology and Conservation
rine cleaning mutualisms (species feeding upon ectoparasites and injured tissues of
Laboratory, Universidade Federal
Fluminense, Niterói, Brazil others) at large spatial scales.
3
Departamento de Ecologia, Universidade de Location: Eleven marine biogeographical provinces.
Brasília, Brasília, Brazil
Time period: 1971–2018.
4
School of Biological Sciences, The
University of Queensland, St Lucia, Major taxa studied: Reef fish and shrimps.
Queensland, Australia Methods: We combined field and literature data to test whether recurrent patterns
5
Department of Collective Behaviour, Max in mutualistic networks (nestedness, modularity) describe the distributions of marine
Planck Institute for Ornithology, Radolfzell,
Germany cleaning interactions. Nested network structures suggest that some cleaner species
6
Department of Biology, Universität interact with many clients while the others clean fewer, predictable subsets of these
Konstanz, Konstanz, Germany
clients; modular network structures suggest that cleaners and clients interact within
Correspondence defined, densely connected subsets of species. We used linear mixed models to eval‐
Juan Pablo Quimbayo, Departamento de
uate whether the life‐history traits of cleaners contribute to the emergence of these
Ecologia e Zoologia, Universidade Federal de
Santa Catarina, Florianópolis, Brazil. patterns locally and whether environmental and geographical factors influence the
Email: [email protected]
network structures.
Results: Marine cleaning networks were more nested than modular. Nestedness was
FUNDING INFORMATION
National Council for Scientific and prevalent in communities with dedicated cleaners (ones that feed exclusively by
Technological Development, CNPq Brazil, cleaning), whereas communities with only facultative cleaners (ones that clean op‐
Grant/Award Numbers: 403740/2012‐6,
153797/2016‐9, 150784/2015‐5 and portunistically) were generally unstructured. Cleaner type and taxa were the only
305358/2015‐4; Australian Research traits shaping networks, with dedicated fish cleaners contributing disproportionally
Council; University of Queensland;
Australian Museum. more than facultative cleaners and shrimps to the emergence of nestedness. Although
cleaner species seem concentrated around the tropics and biodiversity centres, we
did not detect an influence of environmental and geographical factors on network
structures.
Main conclusions: Dedicated species are key in shaping the structure of marine
cleaning mutualistic networks. By relying exclusively on cleaning to feed, dedicated
cleaners interact with most of the available clients and form the network core,
whereas the opportunistic facultative species tend to clean the most common cli‐
ents. We hypothesize that trophic niche variation and phenotypic specialization are
major drivers of this asymmetry in marine mutualisms. Our study strengthens the
links between biotic interactions at the community level and the distribution of spe‐
cies and specializations at larger spatial scales.

*
J.P.Q. and M.C. should be considered joint first authors.

Global Ecol Biogeogr. 2018;1–13. wileyonlinelibrary.com/journal/geb © 2018 John Wiley & Sons Ltd | 1
2 | QUIMBAYO et al.

KEYWORDS
cleaner–client, dedicated–facultative cleaners, ecological network, marine mutualism,
modularity, nestedness, resource partitioning

1 | I NTRO D U C TI O N called “dedicated” cleaners (Vaughan et al., 2017). Such varying

degrees of specialization in cleaning, as well as the distribution of
Every species is involved in intricate webs of ecological interac‐ cleaning interactions, can provide insights into the fundamental and
tions. A persistent challenge in ecology is to unravel the interplay realized foraging niche of cleaner species within marine communi‐
between ecological and evolutionary processes shaping these inter‐ ties (Floeter, Vázquez, & Grutter, 2007; Sazima et al., 2010). Cleaning
actions. Recent years have seen a surge in the study of the resultant mutualisms are influenced by multiple characteristics of cleaners
ecological networks, aiming to disentangle the underlying drivers and clients, such as morphology, size, behaviour, trophic group and
of their structure (e.g., Andreazzi, Thompson, & Guimarães, 2017; abundance (Baliga & Mehta, 2015; Cheney, Grutter, Blomberg, &
Bascompte, Jordano, Mélián, & Olesen, 2003; Fontaine et al., 2011) Marshall, 2009; Floeter et al., 2007; Quimbayo, Dias, Schlickmann,
and reveal the implications of this structure for individual fitness, & Mendes, 2017). Nevertheless, the contribution of such traits to
population structure and community dynamics (e.g., Bascompte & shaping the mutualistic network at the community level remains
Jordano, 2007; Pascual & Dunne, 2006). Although the bulk of the poorly understood.
work comes from studies at the local community scales, the latest In the last decade, we began to understand that marine cleaning
research has been tackling the timely challenge of assessing general interactions can be distributed asymmetrically within a community
assembling principles of ecological networks at large spatial scales because individual species show varying degrees of specialization
(e.g., Martín‐González et al., 2015; Zanata et al., 2017). Mutualisms in cleaning interactions (Barbu, Guinand, Bergmüller, Alvarez, &
are at the forefront of this trend. Bshary, 2011; Guimarães, Sazima, et al., 2007; Sazima et al., 2010).
We can now appreciate how recurrent structural patterns of mu‐ A notable resultof such heterogeneity is the so‐called nested pat‐
tualistic networks vary at macroecological scales (e.g., Schleuning tern, in which the interactions of low‐connected species tend to be
et al., 2012; Trøjelsgaard, Jordano, Carstensen, & Olesen, 2015) a subset of interactions of highly connected species (Bascompte et
owing to the environment (e.g., Sebastián‐González, Dalsgaard, al., 2003; Guimarães, Sazima, et al., 2007). However, the previous
Sandel, & Guimarães, 2015), phylogeny (e.g., Rezende, Lavabre, studies were restricted in their geographical scope, leaving three
Guimarães, Jordano, & Bascompte, 2007) and the degree of phe‐ crucial knowledge gaps: (a) How widespread these network patterns
notypic specialization of the interacting species (e.g., Bascompte are at large spatial scales; (b) which main life‐history traits explain
& Jordano, 2007; Vázquez, Chacoff, & Cagnolo, 2009). However, the disproportionately high cleaning services of some species at the
most attention has been given to mutualisms in terrestrial environ‐ local scale; and (c) which, if any, environmental and geographical fac‐
ments, particularly animal–plant interactions describing pollination tors relate to the emergence of these network patterns. Therefore,
and seed dispersal. We know considerably less about animal–ani‐ it remains uncertain whether nestedness is a recurrent pattern in
mal mutualisms in the ocean (but see Guimarães, Sazima, dos Reis, marine cleaning networks and, if so, the nature of the underlying
& Sazima, 2007; Ollerton, McCollin, Fautin, & Allen, 2007; Sazima, drivers. Exploration of how the structure of local networks var‐
Guimarães, dos Reis, & Sazima, 2010; Thompson, Adam, Hultgren, ies across communities globally can shed light into the ecological
& Thacker, 2013). For example, little is known about how marine and evolutionary mechanisms of the phenotypic specializations of
mutualistic networks are structured at large spatial scales and species (e.g., Martín‐González et al., 2015; Schleuning et al., 2012;
whether such structures are explained by similar drivers to the ter‐ Zanata et al., 2017). Here, we compiled a comprehensive empirical
restrial mutualisms. dataset on cleaning interactions in 28 marine communities distrib‐
In the marine environment, cleaning is a conspicuous mutualistic uted across 11 marine biogeographical provinces world‐wide, with
interaction, in which a cleaner species benefits from feeding upon three goals in mind.
ectoparasites and tissue from the body of another, client species First, we describe the structure of marine cleaning mutualistic
(Côté, 2000). Cleaning mutualism can also influence local abun‐ networks on a large spatial scale. As observed in animal–plant mu‐
dance and species diversity by promoting population and commu‐ tualisms, we expected that cleaning networks can be either modular,
nity health (e.g., Bshary, 2003; Waldie, Blomberg, Cheney, Goldizen, in which interactions are compartmentalized into sets of cleaner and
& Grutter, 2011). In tropical and subtropical waters, c. 259 species client species that interact more often with each other than with the
of two taxa (teleost fish and decapod crustaceans) engage in clean‐ rest (e.g., Olesen, Bascompte, Dupont, & Jordano, 2007; Zanata et
ing behaviour (Vaughan, Grutter, Costello, & Hutson, 2017). Some al., 2017), or nested, in which interactions are hierarchically orga‐
species do so only temporarily (during the juvenile stage) or oppor‐ nized so that some cleaner species interact with many client species
tunistically, and are hereafter called “facultative” cleaners; other and others tend to interact with fewer, proper subsets of these cli‐
species specialize in cleaning throughout their lifetime, hereafter ents (e.g., Bascompte et al., 2003; Guimarães, Sazima, et al., 2007).
QUIMBAYO et al. | 3

Second, given that cleaner species differ in their efficiency to attract dead tissue (such as skin and scales) and/or mucus (see Johnson &
clients (e.g., Cheney et al., 2009), we sought to identify which, if any, Ruben, 1988).
life‐history traits influence the distribution of cleaning interactions The rest of the data were from a systematic review of the lit‐
and so contribute to structure networks locally. Specifically, we erature. We first gathered all the studies cited in the comprehen‐
tested whether the level of exclusivity to mutualism (i.e., whether sive, recent reviews on marine cleaning interactions by Côté (2000)
cleaners are facultative or dedicated; see Guimarães, Rico‐Gray, et and Vaughan et al. (2017). Then, we used electronic databases and
al., 2007; Thompson et al., 2013) along with a range of other biolog‐ search engines (Web of Science, Scopus and Google Scholar) to look
ical traits (taxa, advertising coloration, water column position, body for primary studies and grey literature that provide cleaning inter‐
and group size) are associated with the number of client species with action data as supplement or in tables and figures. We used com‐
which they interact. We hypothesize that networks are nested in the binations of the following search terms, in English, Portuguese and
presence of cleaners (either fish or shrimp) that are specialized and Spanish: cleaning interactions, marine cleaning mutualism, cleaning
efficient, such as those that are dedicated, small‐bodied, coloured, symbiosis, shrimp cleaners, fish cleaners, marine clients, facultative
form larger groups and dwell at the seafloor. These traits (or a com‐ cleaners, obligate cleaners and dedicated cleaners. We considered
bination thereof) could improve the cleaning service and the detec‐ only studies with more than a single cleaner species because we
tion of clients by cleaners or vice versa (see also Cheney et al., 2009; were interested in depicting the local cleaner–client network struc‐
Quimbayo, Dias, et al., 2017; Vaughan et al., 2017). Alternatively, we ture rather than acquiring a comprehensive description of cleaner
hypothesize that modular cleaning networks can emerge if there species distribution. Finally, we minimized the potential effects of
is resource‐use partitioning amongst morphologically and/or be‐ varying sampling effort across studies by focusing on the incidence,
haviourally distinct cleaners. For instance, when fish and shrimp not the strength, of the cleaning interactions. That is, we converted
cleaners co‐occur, shrimps may target clients associated with the all quantitative interactions into qualitative ones to focus only on the
seafloor (Quimbayo, Nunes, et al., 2017) that might not be the focus topologies of the network.
of fish cleaners in the water column. Finally, we test whether the
structure of emergent marine cleaning networks varies at a macro‐
2.2 | Network structure
ecological scale, where environmental and geographical factors (sea
temperature, primary productivity, isolation and distance from a cen‐ We described the mutualistic interactions between cleaner
tre of biodiversity) are associated with species turnover and the com‐ and client species in each site using binary two‐mode networks
position of marine communities and thus indirectly affect the biotic (Boccaletti, Latora, Moreno, Chavez, & Hwang, 2006). A mutual‐
interactions therein (e.g., Pellissier et al., 2017). We hypothesize that istic network was defined by an adjacency matrix M, in which the
cleaning networks can be nested in less‐productive sites distant from element mij = 1 when the cleaner species i was empirically ob‐
their biodiversity centre, where there may be less clients available, served interacting with the client species j, and mij = 0 otherwise.
and that networks can be modular in productive and connected sites In the network depiction, nodes representing cleaners were linked
where clients abound. to nodes representing clients (Supporting Information Figure S1).
We first calculated the connectance of all networks (Supporting
Information Table S1) as the proportion of realized cleaning interac‐
2 | M ATE R I A L S A N D M E TH O DS tions given the total possible interactions (Boccaletti et al., 2006).
Then, we evaluated the global structure of each local network
using metrics that describe two common, and generally compet‐
2.1 | Data sampling
ing, structural patterns of mutualistic networks: nestedness and
We collected and compiled data on cleaning interactions amongst modularity.
marine species from 28 marine communities in 11 marine biogeo‐ Nestedness describes an asymmetric, hierarchical distribution of
graphical provinces (defined for reef fish fauna; see Kulbicki et interactions amongst species (e.g., Bascompte et al., 2003); here it
al., 2013). These were the Caribbean, the Southwestern, Central, indicates that some marine species clean most of the client species
North and Eastern Atlantic, the Western Indian, the Central Indo‐ available locally, whereas others tend to clean subsets of these cli‐
Pacific and the Southwestern, Central, Northeastern and Tropical ents (Guimarães, Sazima, et al., 2007). Modularity describes a com‐
Eastern Pacific (Figure 1 Supporting Information Tables S1 and S2). partmentalized distribution of interactions amongst species (e.g.,
Out of the 28 sites, we empirically recorded cleaning interactions Olesen et al., 2007); here it is indicative of the degree of specificity
in seven of them (Curaçao, Fernando de Noronha, St Paul’s Rocks, of the cleaner–client interactions (e.g., Martín‐González et al., 2015),
Trindade, Príncipe, Red Sea and Galápagos Islands) through direct as a modular network would contain highly connected and almost
observations during dedicated SCUBA dive surveys. We considered non‐overlapping subsets (modules) representing cleaner species
a cleaning interaction as the directed physical contact between in‐ interacting more often with a subset of the available pool of client
dividuals, involving bites from one species (cleaner) on another (cli‐ species. We quantified nestedness with the metric NODF based on
ent), which can result in the removal of ectoparasites, injured or the concepts of overlap and decreasing fill of the adjacency matrix
4 | QUIMBAYO et al.

(Almeida‐Neto, Guimarães, Guimarães, Loyola, & Ulrich, 2008). We 2.4 | Influence of biological traits of cleaners on
quantified modularity with the metric Q that measures the difference local network structure
between the observed fraction of links connecting species in the
To assess the extent to which cleaner traits influence the struc‐
same module and the fraction expected by chance (Newman, 2006)
ture of mutualistic networks, we used linear mixed models (LMMs)
using an algorithm modified for two‐mode networks (Dormann &
to relate their life‐history traits to their individual nestedness and
Strauss, 2013).
modularity contributions (see below). We considered the relation‐
We assessed the significance of nestedness and modularity
ship between cleaning efficiency and behaviour, functional and
using a null model approach. We built null distributions of NODF
morphological traits (Figure 1; Supporting Information Table S3) as
and Q‐values for each local network by randomizing their observed
follows:
cleaning interactions for 1,000 iterations, while constraining the
observed size (i.e., same number of cleaner and client species) and
1. Taxa (fish or shrimp): species of these taxonomic groups vary
connectance. We used an algorithm that randomizes interactions
both morphologically (e.g., vertebrate vs. invertebrate) and in
amongst species based on the empirical observations (i.e., row and
the types of clients they target (Côté, 2000).
column sums; Bascompte et al., 2003). Each cell of the theoretical
2. Type of cleaner (dedicated or facultative): this trait indicates the
matrices had a probability of being filled that was proportional to
level of dependence and/or specialization on cleaning interac‐
the observed number of interactions of both cleaners and clients,
( P
) tions for feeding, given that dedicated cleaners depend exclu‐
defined as: cij = 12 Ci + Rj , where Pi = number of cleaners that in‐
P
sively on cleaning activities to obtain food, whereas facultative
teracted with the client i (row sums); Pj = number of client species
cleaners perform the activity opportunistically or only during ju‐
cleaned by the cleaner j (column sums); C = number of cleaner spe‐
venile stages (Vaughan et al., 2017).
cies (columns); and R = number of client species (rows). The observed
3. Coloration (number of body colours): distinctive colours that con‐
nested or modular structure of a local network was considered sig‐
trast with the background are associated with conspicuousness
nificant when its empirical NODF and Q‐values, respectively, lay
and signalling of the status of the cleaner to clients (Cheney et al.,
outside of the 95% confidence intervals of their corresponding null
2009).
distributions. Finally, we used unpaired Student’s t tests to compare
4. Body size (maximum total length reported in centimetres; Froese
network connectance and the cleaner : client ratio between commu‐
& Pauly, 2017): cleaners species of different sizes may target dif‐
nities with only facultative cleaners and communities also containing
ferent clients (Baliga & Mehta, 2015; Côté, 2000; Floeter et al.,
dedicated cleaners.
2007).
5. Water column position (bottom/medium/top): cleaners more as‐
2.3 | Contribution of cleaner species to the sociated with the bottom could have different clients available
network structure from cleaners that stay in medium and higher positions of the
water column (Johnson & Ruben, 1988; Quimbayo, Nunes, et al.,
To estimate the contribution of cleaner species to the network
2017).
structure, we evaluated the influence of each species on the
6. Group size [solitary, pair, small (3–20), medium (20–50) or large
emergence of the nested and modular patterns. After control‐
(>50 individuals)]: the number of individual cleaners can influence
ling for local differences in the observed number of interactions
the time and efficiency of their cleaning service, in which larger
across cleaner species, we defined whether the overall nested‐
groups are typically quicker to clean a client than smaller groups
ness and modularity of the network were changed by the pres‐
(Côté, 2000).
ence of each cleaner. For each species in a given network, we
compared the NODF and the Q of the entire network with the
NODF and Q‐values, respectively, obtained by randomizing only
the interactions of that target species (Saavedra, Stouffer, Uzzi, 2.5 | Influence of environmental and geographical
& Bascompte, 2011). To randomize the interactions of the tar‐ factors on network structure
get species, we used the same null model described above. The
cleaner species i was deemed a strong contributor to nestedness We used LMMs to evaluate whether environmental and geographi‐

or modularity when, respectively, the NODF and the Q‐value of cal factors that influence local species composition could indirectly

the network whose interactions of the species i were randomized influence the structure (nestedness and modularity) of cleaning mu‐

were consistent and close to the NODF and Q of the original net‐ tualistic networks (see Pellissier et al., 2017). To allow comparisons

work. We calculated the Z‐scores of NODF and Q to compare all amongst localities, we standardized both the nestedness (NODF)

individual cleaners of each network; positive Z‐scored NODF and and the modularity (Q) of each network as Z‐scores (e.g., Sebástián‐

Z‐scored Q of a given species indicated that it contributed to an in‐ González et al., 2015): Z = [(Observed – Meannull)/SDnull], where

crease in the nestedness and modularity of the entire network, re‐ Observed was the empirical values of NODF or Q of a given network,

spectively, whereas negative values indicate otherwise (Saavedra Meannull was the average NODF or Q values of all null model matrices,

et al., 2011). and SDnull was their standard deviation. We used the Bio‐ORACLE
QUIMBAYO et al. | 5

F I G U R E 1 Mutualistic interactions between client and cleaner species. Illustrative examples of the six biological traits of cleaners
considered: taxa (fish/shrimp), type (dedicated/facultative), coloration, body size, water column position (bottom to top) and group size. (a)
Elacatinus phthirophagus cleaning the head of a great Barracuda (Sphyraena barracuda) off Fernando de Noronha archipelago. (b) White‐striped
shrimps (Lysmata grabhami) cleaning the mouth of a Brow Moray (Gymnothorax unicolor) off Ascension Island. (c) Bodianus rufus cleaning a
black margate (Anisotremus surinamensis) off Fernando de Noronha archipelago. (d) Thalassoma noronhanum cleaning a squirrelfish (Holocentrus
adscensionis) off Rocas Atoll. (e) Labroides rubrolabiatus cleaning the grouper Cephalopholis argus off Moorea Island. (f) Bodianus insularis
cleaning a Pomacanthus paru off Ascension Island. Photographs by S. R. Floeter (a, c, e), J. P. Quimbayo (d) and J. Brown (b, f) [Colour figure can
be viewed at wileyonlinelibrary.com]

database (Tyberghein et al., 2012) to obtain data for two environ‐ clorophyll a concentration) estimated from satellite imaging. We
mental factors in each locality: annual mean sea surface tempera‐ then measured three geographical factors for each locality: latitude,
ture (in degrees Celsius) and primary productivity (mean surface isolation and distance from centres of biodiversity. Isolation was
6 | QUIMBAYO et al.

the minimum distance (in kilometres) from mainland (or the nearest variables (set as continuous fixed factors) and the marine biogeo‐
reef when appropriate); and the distance from biodiversity centres graphical provinces as random factors to control for variation be‐
was measured considering the Caribbean for the Atlantic localities tween provinces. Additionally, we considered the number of cleaner
(Floeter et al., 2008) and the Indo‐Australian Archipelago (Kulbicki species in each locality as a fixed factor to account for differences in
et al., 2013) for the rest. the sampling effort between studies. We evaluated the significance
of all predictors with likelihood ratio tests (LRTs), dropped non‐sig‐
nificant (p > 0.05) individual predictors from the full model and cal‐
2.6 | Construction and validation of linear models
culated significant differences in model fit based on χ2 distributions.
We built LMMs with a Gaussian distribution for both biological traits Before building LMMs, we examined potential collinearity
and environmental/geographical factors. To analyse the biological amongst independent variables (predictors) using Pearson correla‐
traits, we combined data of all cleaner species (units of analysis) tions, in which r < 0.60 was considered a cut‐off for keeping pre‐
from all the 28 localities and built two LMMs, one whose depend‐ dictors in the models (Supporting Information Figures S2 and S3).
ent variable was the contribution of cleaner species to nestedness After the model fitting, we calculated the variance inflation factor
and another with the contribution of cleaner species to modularity. (VIF) to ensure that predictors were not correlated with each other
In both models, all the six life‐history traits described above were (the final model presents low multicollinearity, with VIF < 2; Tables 1
independent variables (set as fixed factors), and localities were set and 2). Based on geographical coordinates of each locality and model
as a random effect variable (i.e., random intercept model). To ana‐ residuals, we estimated Moran’s I index using the inverse of the
lyse the environmental and geographical factors, we combined data Euclidean distance amongst localities as weights. Overall, Moran’s
of all 28 localities (units of analysis) and built two other LMMs, one I indicated no spatial autocorrelation in the model residuals (nest‐
with Z‐scored NODF and the other with Z‐scored Q as dependent var‐ edness: I = −0.007 ± 0.018, p = 0.53; modularity: I = −0.03 ± 0.018,
iables, all environmental and geographical factors as independent p = 0.26; Supporting Information Figures S4 and S5); therefore, it

TA B L E 1 Effects of biological traits on

Nestedness contribution Modularity contribution
the individual contribution of cleaner
Biological traits Estimate LTR p‐value Estimate LTR p‐value VIF species to nestedness and modularity of
mutualistic networks based on LMMs
Taxa 4.42 0.04** 3.18 0.07* 1.68
Fish 0.49 1.09
Shrimp −0.35 1.67
Cleaner type 17.26 < 0.01** 0.13 0.71 1.79
Facultative 0.88 1.2
Dedicated 1.88 1.08
Coloration 0.14 0.82 0.36 0.05 0.16 0.69 1.29
Body size −0.02 1.81 0.18 −0.01 0.27 0.6 1.88
Water column 3.19 0.2 0.58 0.64 1.9
position
Bottom 0.07 1.07
Medium 0.08 1.35
Top 0.62 1.27
Group size 1.6 0.81 2.49 0.64 1.85
Large 0.42 1.34
Medium 0.09 1.21
Small 0.27 1.01
Pair 0.49 1.15
Solitary 0.50 1.38
Sample size (n) 140 140
Local variability 0.5 0.61
(random effect
SD)
Model residual 1.04 1.55

Note. Traits are considered as fixed factors and locality as a random factor. The variance inflation
factor (VIF) is shown for individual predictors of both models. **p <0.05; *marginal significance
(0.05 < p <0.10) estimated through likelihood ratio tests (LRTs). SD = Standard Deviation.
QUIMBAYO et al. | 7

TA B L E 2 Effects of environmental and

Z‐scored nestedness Z‐scored modularity
geographical factors on the large‐scale
structure of marine cleaning mutualistic Factors Estimate LRT p‐value Estimate LRT p‐value VIF
networks based on linear mixed‐effect
Local cleaner 0.44 3.65 0.06 −0.15 0.81 0.37 1.12
models
richness
Environmental
Sea surface 0.1 0.03 0.86 0.04 0.01 0.91 1.74
temperature
Primary −0.59 1.74 0.19 −0.02 0.004 0.95 1.06
productivity
Geographical
Latitude −0.02 0.97 0.32 −0.01 0.3 0.58 1.27
Isolation −1.05 2.34 0.13 −0.57 1.4 0.23 2.45
Distance from 0.81 2.43 0.12 0.32 0.8 0.37 1.4
biodiversity
centre
Sample size (n) 28 28
Marine 0.0001 0.0001
biogeographi‐
cal province
variability
(random effect
SD)
Model residual 6.32 0.32

Note. LRT = likelihood ratio test. Marine biogeographical provinces were considered as a random
factor, whereas local cleaner richness and all environmental and geographical variables were consid‐
ered as fixed factors (sea surface temperature, primary productivity, latitude, isolation and distance
from biodiversity centre). The variance inflation factor (VIF) is shown for individual predictors of
both models.

was not necessary to use spatial models to control for this source per locality showed a slightly increasing trend towards eastern longi‐
of variation nor to correct the degrees of freedom of our models. tudes (Figure 2b). Dedicated cleaners were more commonly found in
All analyses were performed in the R environment, v. 3.2.4 (R Core the Caribbean, Southwestern Atlantic and Central Pacific, whereas
Team, 2016). facultative cleaners were distributed more homogeneously across all
marine biogeographical provinces (Figure 2b; Supporting Information
Table S2). Nevertheless, we note that these distributions reflect the
3 | R E S U LT S network sizes, not cleaner diversity per se; that is, the number of
cleaners here indicate the species sampled in each network, which
We recorded 480 client species and 85 cleaner species (Supporting might reflect the number of studies carried out in each biogeograph‐
Information Table S1), out of which 86% were facultative (n = 62 ical province but does not necessarily represent their total cleaner
fish, n = 11 shrimps) and 14% were dedicated cleaners (n = 9 fish, species richness.
n = 3 shrimps; Supporting Information Table S2). Our database Half of the localities contained both dedicated and facultative
comprised about a third of all marine cleaner species (Vaughan et cleaners (n = 14), and the other half contained only facultative clean‐
al., 2017) and contained 50% of all species recognized as dedi‐ ers (n = 14; Figure 2a; Supporting Information Figure S1). The ratio
cated cleaners and 28% of the facultative cleaners (Supporting of cleaner species per client species was lower in localities with both
Information Table S4). Cleaner shrimps may be under‐represented types of cleaners than in facultative‐only localities (Student’s t test,
in the literature due to difficulties in observing such cryptic, crev‐ t = −5.536, df = 25.15, p <0.001; Figure 3a). However, the network
ice‐living and often nocturnal organisms. There were more cleaner connectance (the proportion of realized cleaning interactions) was
species sampled in the Atlantic (n = 88) than in the Pacific (n = 46) similar between localities with and without dedicated cleaners
and Indian (n = 6) ocean basins (Figure 2a; Supporting Information (t = 0.190, df = 23.64, p = 0.851; Figure 3b; Supporting Information
Table S2). Table S1).
The number of cleaners per locality did not show a latitudinal Overall, cleaning networks were more often nested (15 out 28
trend, yet observations were mostly in tropical waters (−20, +20° localities) than modular (5 out 28 localities; Figure 4; Supporting
latitude; Figure 2b). Despite large variability, the number of cleaners Information Figure S1), and these metrics showed no strong
8 | QUIMBAYO et al.

F I G U R E 2 Distribution of marine cleaning mutualistic networks. (a) The 28 sampled localities, with pie charts representing the
proportion of dedicated and facultative cleaner species and sizes proportional to the total richness of sampled cleaner species (full details
in Supporting Information Table S1). (b) Distribution of number of cleaner species in the networks of localities along the longitudinal and
latitudinal gradients (blue = localities with both dedicated and facultative; yellow = facultative only) and across marine biogeographical
provinces (Central, Northeastern and Tropical Eastern Pacific; Caribbean, Southwestern, Central, North and Eastern Atlantic; Western
Indian; Central Indo‐Pacific and Southwestern Pacific; see Supporting Information Tables S1 and S2). (c) Representative cleaning networks
of the localities (dashed boxes in the map) with both dedicated (blue) and facultative cleaners (yellow) and localities with only facultative
cleaners. Cleaners are linked to client species (grey) by binary links whenever they were observed interacting. Cleaning networks of all
localities are in Supporting Information Figure S1 [Colour figure can be viewed at wileyonlinelibrary.com]

latitudinal or longitudinal trends (Supporting Information Figure contributed more than fish to the emergence of the few modular
S6). Networks with dedicated cleaners had high NODF values, networks (marginally significant effect of taxa; Table 1; Supporting
and most of them (79%) were more nested than expected by Information Figure S8), none of the other biological traits was re‐
chance (Figure 4a). In contrast, most networks with only faculta‐ lated to the contribution of cleaners to modularity (Table 1). Finally,
tive cleaners were not nested (71%; Figure 4b). The opposite was the emergence of nestedness and modularity was unrelated to both
found for modularity: facultative‐only localities tended to have environmental (temperature and primary productivity) and geo‐
higher Q‐values (Figure 4d) than localities with dedicated cleaners graphical (latitude, isolation and distance from biodiversity centres)
(Figure 4c), and we emphasize that most of the Q‐values were not factors (Table 2).
different from the null expectancy. Only three out of 28 locali‐
ties were both nested and modular (all with dedicated cleaners;
Figure 4). 4 | D I S CU S S I O N
Nested networks were characterized by a core of cleaners in‐
teracting with most clients, along with more peripheral cleaners in‐ Our study reveals that the biological traits of species are key in
teracting with the highly connected clients (Figure 2c). Dedicated structuring mutualisms in marine communities. At macroecologi‐
cleaners were often, but not always, central species in the networks cal scales, marine cleaning mutualistic networks with dedicated
(Figure 2c; Supporting Information Figure S1). Cleaner type was the cleaners repeatedly display a hierarchical, nested architecture, in
most important biological trait, followed by taxa, in explaining the contrast to the generally unstructured networks containing only
positive contribution of cleaners to nestedness across all networks facultative cleaners. Cleaner taxa play a minor role in structur‐
(Table 1). Dedicated cleaners showed significantly higher contri‐ ing these networks, and the effects, if any, of environmental and
butions to the emergence of nestedness than facultative cleaners geographical factors were not detected. Nestedness in cleaning
(Figure 1 and Table 1; Supporting Information Table S3), whereas interactions being mainly dependent on dedicated fish species im‐
fish cleaners contributed more than shrimps (Table 1; Supporting plies that the level of interaction specificity drives the asymmetry
Information Figure S7). All other traits (coloration, position in the of the distribution of biotic interactions, (Guimarães, Rico‐Gray,
water column, body and group sizes) did not significantly contribute et al., 2007; Thompson et al., 2013) at both local and large spatial
to the emergence of nestedness (Table 1). Although shrimp cleaners scales.
QUIMBAYO et al. | 9

preference for rock/reef crevices where clients are more limited.

However, in most communities, shrimps share clients with cleaner
fish of similar life history, such as small‐bodied species that form
small groups near the seafloor (e.g., Figure 1a; Johnson & Ruben,
1988; Quimbayo, Nunes, et al., 2017). Indeed, the typically nested
cleaning networks imply that resource use amongst cleaners often
overlaps in such a way that some cleaners interact with many cli‐
ents and others interact only with predictable subsets of the most
interactive clients.

4.2 | Dedicated mutualists shape marine

cleaning networks
A key finding is that such nested cleaning networks primarily oc‐
curred in marine communities containing dedicated cleaners. The
level of dependence between the interacting species can result in
markedly different network structures, which is evident in intimate
and non‐intimate biological interactions. High‐interaction intimacy
networks, such as ant–myrmecophyte and gobies–shrimp symbi‐
oses, have high reciprocal specialization and are therefore highly
compartmentalized (Blüthgen, Menzel, Hovestadt, Fiala, & Blüthgen,
2007; Guimarães, Rico‐Gray, et al., 2007; Thompson et al., 2013). In
contrast, low‐interaction intimacy networks, such as seed dispersal
and cleaning mutualisms, contain relatively loose interactions and
therefore have overall low specificity and tend to be nested (e.g.,
F I G U R E 3 Proportion of species and mutualistic interactions
Bascompte et al., 2003; Guimarães, Sazima, et al., 2007). Our study
in communities containing different cleaner types. (a) Number
of sampled cleaner species per client species. (b) Network further suggests that the level of reliance on the interactions itself
connectance in all mutualistic networks of ecological communities (i.e., being dedicated or not to it) can also influence the network
with dedicated and facultative cleaners (n = 14) and in localities structure and promote nestedness.
with only facultative cleaners (n = 14). Network connectance is the Dedicated species adopt cleaning as an exclusive foraging strat‐
proportion of realized links (observed cleaning interactions) given
egy, whereas facultative species are opportunists (Côté, 2000;
the possible number of interactions. Box plots indicate the median
(horizontal line), mean (diamond), first and third quartiles (box), Vaughan et al., 2017). Therefore, dedicated cleaners target most
range of values excluding outliers (whiskers) and the observed data of the available clients, becoming highly connected species at the
(grey circles) [Colour figure can be viewed at wileyonlinelibrary. core of the network. In contrast, facultative cleaners exploit other
com] food sources and may clean fewer clients as opportunities arise,
for instance, clients that are common or use the same habitat. This
asymmetry in foraging specificity leads to nestedness (Guimarães,
4.1 | Marine cleaning networks are more nested
Sazima, et al., 2007), which we found to be prominent in communi‐
than modular
ties where dedicate and facultative cleaners co‐occur. In localities
We detected non‐random mutualistic interactions in marine com‐ with only facultative cleaners, their use of clients overlaps more be‐
munities around the world. Heterogeneity in local distributions of cause they clean when juveniles or when predation risk is low (Côté,
cleaning services reflects variation in the use of clients as food re‐ 2000; Vaughan et al., 2017) and therefore eventually and sporad‐
sources by cleaners. Modular networks suggest marked niche parti‐ ically interact with the entire pool of clients, hindering the emer‐
tioning and specialization (e.g., Olesen et al., 2007), whereas nested gence of nestedness.
networks suggest hierarchical niche overlap and varying degrees Through behavioural and morphological adaptations, dedicated
of generalization and specialization in mutualistic interactions (e.g., cleaners might be able to provide a better cleaning service, attract
Guimarães, Sazima, et al., 2007). The lack of modularity in most more clients, and thus take over a large part of the cleaning activi‐
of the cleaning networks indicates that specific groups of clean‐ ties that might otherwise be performed by opportunistic facultative
ers interacting with specific groups of clients are rare. Moreover, cleaners. The connectance of the networks with dedicated cleaners
we found only a minor contribution of shrimps in promoting mod‐ is very similar to the connectance of facultative‐only networks, al‐
ular networks, suggesting that in some localities shrimps might though in the latter the number of cleaners is higher. This suggests
use a slightly different set of clients from the fish cleaners, pos‐ that few dedicated cleaners can clean similar proportions of clients
sibly resulting from the the nocturnal habits of shrimps and their cleaned by many facultative species (i.e., the cleaning service is more
10 | QUIMBAYO et al.

F I G U R E 4 Network topological metrics of marine cleaning networks. Nestedness (NODF) and Modularity (Q) of cleaning networks
in localities with both dedicated and facultative cleaner species (a, c) and in localities with only facultative cleaners (b, d). Red dotted
lines represent mean NODF and Q‐values. Significantly nested or modular networks display empirical values (circles) outside the 95%
confidence intervals generated by null models (whiskers). Localities are ordered by decreasing latitude [Colour figure can be viewed at
wileyonlinelibrary.com]

efficient where dedicated species are present; Quimbayo et al., 2006; Cheney et al., 2009). They also show more cleaning‐specific
2012; Sazima et al., 2010). Moreover, where dedicated species are cognitive skills (Gingins & Bshary, 2016) and higher escape perfor‐
missing, more species engage in cleaning (Quimbayo, Schlickmann, mance than facultatives (Gingins, Roche, & Bshary, 2017). Finally,
Floeter, & Sazima, 2018); for example, in Cape Verde and São Tomé their usual small bodies may facilitate interaction with a large range
Island, where multiple cleaner species stations are the norm and of clients (Vaughan et al., 2017). Although these traits may make
parasite‐loaded clients seem vehemently to “request” cleaning from dedicated cleaners more attractive to clients, our results showed
facultative cleaners (Quimbayo et al., 2012). However, where ded‐ that colour, body size and behaviour do not explain the distribution
icated Labroides cleaners occur, few opportunities are left for the of cleaning interactions at the community level. This corroborates
less‐effective facultatives (Barbu et al., 2011). Morphological and the apparent lack of relationship between cleaning and body mor‐
behavioural adaptations of dedicated cleaners seem fundamental phology (Arnal et al., 2006; Côté, 2000) and suggests that life‐his‐
for their overall effectiveness. tory traits other than specificity in cleaning play a minor role, if any,
Dedicated cleaners often have coloured bodies (Cheney et in structuring cleaning networks. Nevertheless, we acknowledge
al., 2009), with dark lateral stripes associated with yellow or blue, that our coarse, first‐order measures of behaviour and morphology
strongly contrasting with background (Arnal, Verneau, & Desdevises, might have failed to capture subtle differences that could influence
QUIMBAYO et al. | 11

cleaning behaviour; we assessed coloration patterns as the number et al., 2015; Rezende et al., 2007), but it is a promising avenue for re‐
of human‐visible colours, and the variation in body size turned out solving whether evolutionary pathways of teleost fish and shrimps
to be too low. Quantifying more accurate proxies of cleaner perfor‐ modulate cleaning mutualisms.
mance will help to illuminate the influence of the traits of cleaners
on mutualistic networks.
5 | CO N C LU S I O N S
4.3 | Cleaning at large spatial scales
Dedicated, more efficient cleaner species shape the global struc‐
Cleaning mutualism occurs in all ocean basins, forming networks ture of marine cleaning mutualism. The presence of such obligate
that, although structured in different ways, are similar in containing mutualists modulates trophic niche variation amongst species in
a relatively low richness of cleaners. Although one caveat here is the marine communities, driving the asymmetry in marine cleaning at
inherent incompleteness of our database, the number of cleaners local and large spatial scales. The rarity of modular networks sug‐
around the globe, especially the dedicated ones, is strikingly small: gests that cleaning specialization is not as marked at the community
only 16 fish species and eight shrimp species from two families each level as it is at the species level. Instead, the recurrence of nested
(Vaughan et al., 2017; Supporting Information Table S4). After con‐ networks in communities with dedicated cleaners suggests that the
trolling for the total local cleaner richness, we found no geographical partitioning of clients as resources happens hierarchically, in which
or environmental influence on the structure of the resulting clean‐ dedicated cleaners tend to interact with many clients and faculta‐
ing networks. This contrasts with terrestrial mutualisms, in which tives tend to interact with the most common clients. Phenotypic
climatic seasonality and past climate stability can lead to modular or specialization, and the resultant trophic niche variation, are key
nested networks (e.g., Schleuning et al., 2014; Sebástián‐González et underlying mechanisms that shape the structure of mutualistic net‐
al., 2015). works in the ocean.
Marine cleaning interactions and dedicated cleaners are appar‐
ently more common towards tropical latitudes. In animal–plant mu‐
tualisms, tropical ecosystems have lower network specialization (i.e., AC K N OW L E D G E M E N T S
low modularity and high nestedness) than temperate counterparts.
We thank all the researchers who have made their data available on‐
Avian seed dispersal in temperate regions tends towards modules of
line; A. W. Aued, R. Lamb and R. Bshary for sharing data, pictures
frugivorous species specialized on particular fruiting plants, suggest‐
and videos on cleaning interactions; M. M. Pires, T. C. Mendes, J.
ing niche partition (Dalsgaard et al., 2017). In contrast, as obligate
R. Paula, C. Sorte and three anonymous reviewers for insight‐
frugivory is more common among tropical birds, the consumption
ful comments on the manuscript; and the Marine Macroecology
of fruiting plants in these regions overlaps more, leading to nested,
and Biogeography Laboratory for logistical support. This study
non‐modular networks (e.g., Bascompte et al., 2003; Schleuning et
received funds from PELD/ILOC (National Council for Scientific
al., 2014). Likewise, our nestedness results support low network
and Technological Development, CNPq Brazil 403740/2012‐6).
specialization in marine cleaning mutualism in tropical and subtrop‐
Coordination for the Improvement of Higher Education Personnel,
ical areas (Quimbayo et al., 2018). If evolutionary processes can in‐
CAPES, Brazil provided a doctoral scholarship for J.P.Q.; CNPq Brazil
fluence the specificity of ecological interactions (Thompson et al.,
provided research fellowships to M.C. (153797/2016‐9), M.S.D.
2013), low resource‐use specialization in tropical areas could have
(150784/2015‐5) and S.R.F. (305358/2015‐4); The Australian
followed the evolution of obligate mutualists (Dalsgaard et al., 2017;
Research Council and The University of Queensland funded A.S.G.
Schleuning et al., 2014), here the dedicated cleaners.
The Australian Museum funded J.H.A.B.
The number of dedicated cleaner species is not directly propor‐
tional to the regional species pool or to their clade diversity (Baliga
& Law, 2016), probably owing to the rarity of trait combinations that DATA AC C E S S I B I L I T Y
favour cleaning behaviour over evolutionary time: small body size
Data can be found at the Zenodo data repository: https://doi.
(Baliga & Law, 2016), aposematic colours (Cheney et al., 2009) and
org/10.5281/zenodo.1253028.
signalling behaviour to the clients (Côté, 2000). Dedicated clean‐
ers are restricted to only two fish genera (Labroides and Elacatinus)
and four shrimps genera (Lysmata, Ancylomenes, Periclimenes and REFERENCES
Urocaridella), which share these cleaning‐facilitating traits (Sims et Almeida‐Neto, M., Guimarães, P., Guimarães, P. R. Jr, Loyola, R. D., &
al., 2014; Taylor & Hellberg, 2005). If such traits are conserved along Ulrich, W. (2008). A consistent metric for nestedness analysis in eco‐
phylogenies and influence the set of clients for the bearer, they could logical systems: Reconciling concept and measurement. Oikos, 117,
ultimately shape mutualistic networks (e.g., Martín‐González et al., 1227–1239. https://doi.org/10.1111/j.0030-1299.2008.16644.x
Andreazzi, C. J. N., Thompson, P. R., & Guimarães, P. R. Jr. (2017).
2015; Schleuning et al., 2014). The evaluation of phylogenetic sig‐
Network structure and selection asymmetry drive coevolution in
nals in mutualistic networks remains challenging because it requires species‐rich antagonistic interactions. The American Naturalist, 190,
phylogenies on both sides of the interaction (e.g., Martín‐González 99–115. https://doi.org/10.1086/692110
12 | QUIMBAYO et al.

Arnal, C., Verneau, O., & Desdevises, Y. (2006). Phylogenetic relation‐ Gingins, S., Roche, D. G., & Bshary, R. (2017). Mutualistic cleaner fish
ships and evolution of cleaning behaviour in the family Labridae: maintains high escape performance despite privileged relationship
Importance of body colour pattern. Journal of Evolutionary Biology, with predators. Proceedings of the Royal Society B: Biological Sciences,
19, 755–763. https://doi.org/10.1111/j.1420-9101.2005.01059.x 284, 20162469. https://dx.doi.org/10.1098/rspb.2016.2469
Baliga, V. B., & Law, C. J. (2016). Cleaners among wrasses: Phylogenetics Grutter, A. S. (1996). Parasite removal rates by the cleaner wrasse
and evolutionary patterns of cleaning behavior within Labridae. Labroides dimidiatus. Marine Ecology Progress Series, 130, 61–70.
Molecular Phylogenetics and Evolution, 94, 424–435. https://doi. https://doi.org/10.3354/meps130061
org/10.1016/j.ympev.2015.09.006 Guimarães, P. R., Rico‐Gray, V., Oliveira, P. S., Izzo, T. J., dos Reis, S. F., &
Baliga, V. B., & Mehta, R. S. (2015). Linking cranial morphology to prey Thompson, J. N. (2007). Interaction intimacy affects structure and
capture kinematics in three cleaner wrasses: Labroides dimidia‐ coevolutionary dynamics in mutualistic networks. Current Biology, 17,
tus, Larabicus quadrilineatus, and Thalassoma lutescens. Journal of 1797–1803. https://doi.org/10.1016/j.cub.2007.09.059
Morphology, 276, 1377–1391. Guimarães, P. R., Sazima, C., dos Reis, S. F., & Sazima, I. (2007). The
Barbu, L., Guinand, C., Bergmüller, R., Alvarez, N., & Bshary, R. (2011). nested structure of marine cleaning symbiosis: Is it like flowers and
Cleaning wrasse species vary with respect to dependency on the mu‐ bees? Biology Letters, 3, 51–54.
tualism and behavioural adaptations in interactions. Animal Behaviour, Hillebrand, H. (2004). On the generality of the latitudinal diver‐
82, 1067–1074. https://doi.org/10.1016/j.anbehav.2011.07.043 sity gradient. The American Naturalist, 163, 192–211. https://doi.
Bascompte, J., & Jordano, P. (2007). Plant‐animal mutualistic networks: org/10.1086/381004
The architecture of biodiversity. Annual Review of Ecology Evolution Johnson, W. S., & Ruben, P. (1988). Cleaning behavior of Bodiunus rufus,
and Systematics, 38, 567–593. https://doi.org/10.1146/annurev. Thalassoma bifasciatum, Gobiosoma evelynae, and Periclimenes
ecolsys.38.091206.095818 pedersoni along a depth gradient at Salt River Submarine Canyon.
Bascompte, J., Jordano, P., Melián, C. J., & Olesen, J. M. (2003). The St. Croix. Environmental Biology of Fishes, 23, 225–232. https://doi.
nested assembly of plant–animal mutualistic networks. Proceedings org/10.1007/BF00004913
of the National Academy of Sciences USA, 100, 9383–9387. https://doi. Kulbicki, M., Parravicini, V., Bellwood, D. R., Arias‐Gonzàlez, E., Chabanet,
org/10.1073/pnas.1633576100 P., Floeter, S. R., … Mouillot, D. (2013). Global biogeography of reef
Blüthgen, N., Menzel, F., Hovestadt, T., Fiala, B., & Blüthgen, N. (2007). fishes: A hierarchical quantitative delineation of regions. PLoS One, 8,
Specialization, constraints, and conflicting interests in mutualistic e81847. https://doi.org/10.1371/journal.pone.0081847
networks. Current Biology, 17, 341–346. https://doi.org/10.1016/j. Luiz, O. J., Madin, J. S., Robertson, D. R., Rocha, L. A., Wirtz, P., & Floeter,
cub.2006.12.039 S. R. (2012). Ecological traits influencing range expansion across
Boccaletti, S., Latora, V., Moreno, Y., Chavez, M., & Hwang, D. U. (2006). large oceanic dispersal barriers: Insights from tropical Atlantic reef
Complex networks: Structure and dynamics. Physics Reports, 424, fishes. Proceedings of the Royal Society B: Biological Sciences, 279,
175–308. https://doi.org/10.1016/j.physrep.2005.10.009 1033–1040.
Bshary, R. (2003). The cleaner wrasse, Labroides dimidiatus, is a Martín‐González, A. M., Dalsgaard, B., Nogués‐Bravo, D., Graham, C.
key organism for reef fish diversity at Ras Mohammed National H., Schleuning, M., Maruyama, P. K., … Azevedo, S. M. (2015). The
Park. Egypt. Journal of Animal Ecology, 72, 169–176. https://doi. macroecology of phylogenetically structured hummingbird–plant
org/10.1046/j.1365-2656.2003.00683.x networks. Global Ecology and Biogeography, 24, 1212–1224. https://
Cheney, K. L., Grutter, A. S., Blomberg, S. P., & Marshall, N. J. (2009). doi.org/10.1111/geb.12355
Blue and yellow signal cleaning behavior in coral reef fishes. Current Newman, M. E. J. (2006). Modularity and community structure in
Biology, 19, 1283–1287. https://doi.org/10.1016/j.cub.2009.06.028 networks. Proceedings of the National Academy of Sciences United
Côté, I. M. (2000). Evolution and ecology of cleaning symbiosis in the sea. States of America, 103, 8577–8582. https://doi.org/10.1073/
Oceanography and Marine Biology: an Annual Review, 38, 311–355. pnas.0601602103
Dalsgaard, B., Schleuning, M., Maruyama, P. K., Dehling, D. M., Sonne, Olesen, J. M., Bascompte, J., Dupont, Y. L., & Jordano, P. (2007). The mod‐
J., Vizentin‐Bugoni, J., … Rahbek, C. (2017). Opposed latitudinal pat‐ ularity of pollination networks. Proceedings of the National Academy
terns of network‐derived and dietary specialization in avian plant– of Sciences USA, 104, 19891–19896. https://doi.org/10.1073/
frugivore interaction systems. Ecography, 40, 1395–1401. https://doi. pnas.0706375104
org/10.1111/ecog.02604 Ollerton, J., McCollin, D., Fautin, D. G., & Allen, G. R. (2007). Finding
Dormann, C. F., & Strauss, R. (2013). Detecting modules in quantitative NEMO: Nestedness engendered by mutualistic organization in
bipartite networks: The QuaBiMo algorithm. Methods in Ecology and anemonefish and their hosts. Proceedings of the Royal Society
Evolution, 5, 90–98. B: Biological Sciences, 274, 591–598. https://doi.org/10.1098/
Floeter, S. R., Rocha, L. A., Robertson, D. R., Joyeux, J. C., Smith‐Vaniz, rspb.2006.3758
W. F., Wirtz, P., …Bernardi, G. (2008). Atlantic reef fish biogeography Pascual, M., & Dunne, J. A. (2006). Ecological networks: Linking structure to
and evolution. Journal of Biogeography, 35, 22–47. dynamics in food webs. New York: Oxford University Press.
Floeter, S. R., Vázquez, D. P., & Grutter, A. S. (2007). The macroecology of Pellissier, L., Albouy, C., Bascompte, J., Farwig, N., Graham, C., Loreau,
marine cleaning mutualisms. Journal of Animal Ecology, 76, 105–111. M., … Rohr, R. (2017). Comparing species interaction networks along
https://doi.org/10.1111/j.1365-2656.2006.01178.x environmental gradients. Biological Reviews, 93, 785–800. https://
Fontaine, C., Guimarães, P. R., Kéfi, S., Loeuille, N., Memmott, J., doi.org/10.1111/brv.12366
van Der Putten, W. H., … Thébault, E. (2011). The ecologi‐ Quimbayo, J. P., Dias, M. S., Schlickmann, O. R. C., & Mendes, T. C. (2017).
cal and evolutionary implications of merging different types Fish cleaning interactions on a remote island from the Tropical
of networks. Ecology Letters, 14, 1170–1181. https://doi. Eastern Pacific. Marine Biodiversity, 47, 603–608.
org/10.1111/j.1461-0248.2011.01688.x Quimbayo, J. P., Floeter, S. R., Noguchi, R., Rangel, C. A., Gasparini, J.
Froese, R., & Pauly, D. (2017). FishBase. World wide web electronic publi‐ L., Sampaio, C. L. S., … Rocha, L. A. (2012). Cleaning mutualism in
cation. Retrieved from www.fishbase.org. Accessed 13 March 2018. Santa Luzia (Cape Verde Archipelago) and São Tomé Islands, Tropical
Gingins, S., & Bshary, R. (2016). The cleaner wrasse outperforms other Eastern Atlantic. Marine Biodiversity Records, 5, e118. https://doi.
labrids in ecologically relevant contexts, but not in spatial discrim‐ org/10.1017/S175526721200108X
ination. Animal Behaviour, 115, 145–155. https://doi.org/10.1016/j. Quimbayo, J. P., Nunes, L. T., Ozekoski, R., Floeter, S. R., Morais, R. A.,
anbehav.2016.02.022 Fontoura, L., … Sazima, I. (2017). Cleaning interactions at the only
QUIMBAYO et al. | 13

atoll in the South Atlantic. Environmental Biology of Fishes, 100, 865– Vaughan, D. B., Grutter, A. S., Costello, M. J., & Hutson, K. S. (2017).
875. https://doi.org/10.1007/s10641-017-0612-3 Cleaner fishes and shrimp diversity and a re‐evaluation of clean‐
Quimbayo, J. P., Schlickmann, O. R. C., Floeter, S. R., & Sazima, I. (2018). ing symbioses. Fishes and Fisheries, 18, 698–716. https://doi.
Cleaning interactions at the southern limit of tropical reef fishes in org/10.1111/faf.12198
the West Atlantic. Environmental Biology of Fishes, 101, 1195–1204. Vázquez, D. P., Chacoff, N. P., & Cagnolo, L. (2009). Evaluating multiple
R Core Team (2016). R: A language and environment for statistical com‐ determinants of the structure of plant–animal mutualistic networks.
puting. R Foundation for Statistical Computing, Vienna, Austria. URL Ecology, 90, 2039–2046. https://doi.org/10.1890/08-1837.1
https://www.R-project.org/. Waldie, P. A., Blomberg, S. P., Cheney, K. L., Goldizen, A. W., & Grutter,
Rezende, E. L., Lavabre, J. E., Guimarães, P. R., Jordano, P., & Bascompte, A. S. (2011). Long‐term effects of the cleaner fish Labroides dimidia‐
J. (2007). Non‐random coextinctions in phylogenetically struc‐ tus on coral reef fish communities. PLoS One, 6, e21201. https://doi.
tured mutualistic networks. Nature, 448, 925–928. https://doi. org/10.1371/journal.pone.0021201
org/10.1038/nature05956 Zanata, T. B., Dalsgaard, B., Passos, F. C., Cotton, P. A., Roper, J. J.,
Saavedra, S., Stouffer, D. B., Uzzi, B., & Bascompte, J. (2011). Strong Maruyama, P. K., … Franklin, D. C. (2017). Global patterns of interac‐
contributors to network persistence are the most vulnerable tion specialization in bird–flower networks. Journal of Biogeography,
to extinction. Nature, 478, 233–235. https://doi.org/10.1038/ 44, 1891–1910. https://doi.org/10.1111/jbi.13045
nature10433
Sazima, C., Guimarães, P. R. Jr, dos Reis, S. F., & Sazima, I. (2010). What
makes a species central in a cleaning mutualism network? Oikos, 119,
1319–1325. https://doi.org/10.1111/j.1600-0706.2009.18222.x B I O S K E TC H E S
Schleuning, M., Fründ, J., Klein, A. M., Abrahamczyk, S., Alarcón,
Juan Pablo Quimbayo is a postdoctoral researcher at Universidade
R., Albrecht, M., … Dalsgaard, B. (2012). Specialization of mu‐
tualistic interaction networks decreases toward tropical lati‐ Federal Fluminense and an associate researcher at the Marine
tudes. Current Biology, 22, 1925–1931. https://doi.org/10.1016/j. Macroecology and Biogeography Laboratory in Brazil. He is in‐
cub.2012.08.015 terested in both local and large‐scale ecological questions, and
Schleuning, M., Ingmann, L., Strauß, R., Fritz, S. A., Dalsgaard, B.,
his research focuses on fish assemblages, marine mutualisms and
Dehling, D. M., … Dormann, C. F. (2014). Ecological, historical
and evolutionary determinants of modularity in weighted seed‐ functional diversity.
dispersal networks. Ecology Letters, 17, 454–463. https://doi. J.P.Q., M.C., M.S.D., and S.R.F. conceived the idea. J.P.Q. and
org/10.1111/ele.12245 M.C. analysed the data, and M.S.D. contributed to coding and to
Sebástián‐González, E., Dalsgaard, B., Sandel, B., & Guimarães, P. R.
interpretation of results. J.P.Q., S.G., A.S.G., J.A.H.B. and S.R.F.
(2015). Macroecological trends in nestedness and modularity of
seed‐dispersal networks: Human impact matters. Global Ecology and collected data, and J.P.Q. compiled data from the literature. J.P.Q.
Biogeography, 24, 293–303. https://doi.org/10.1111/geb.12270 and M.C. wrote the paper and led the revisions, and M.S.D.,
Sims, C. A., Riginos, C., Blomberg, S. P., Huelsken, T., Drew, J., & Grutter, S.R.F., S.G. and A.S.G. critically reviewed several versions of the
A. S. (2014). Cleaning up the biogeography of Labroides dimidiatus
manuscript. All authors gave final approval for publication and
using phylogenetics and morphometrics. Coral Reefs, 33, 223–233.
agree to be held accountable for the work performed therein.
https://doi.org/10.1007/s00338-013-1093-2
Taylor, M. S., & Hellberg, M. E. (2005). Marine radiations at small geo‐
graphic scales: Speciation in neotropical reef gobies (Elacatinus).
S U P P O R T I N G I N FO R M AT I O N
Evolution, 59, 374–385. https://doi.org/10.1554/04-590
Thompson, A. R., Adam, T. C., Hultgren, K. M., & Thacker, C. E. (2013). Additional Supporting Information may be found online in the sup‐
Ecology and evolution affect network structure in an intimate ma‐
porting information tab for this article.
rine mutualism. The American Naturalist, 182, E58–E72. https://doi.
org/10.1086/670803
Trøjelsgaard, K., Jordano, P., Carstensen, D. W., & Olesen, J. M. (2015).
Geographical variation in mutualistic networks: Similarity, turnover How to cite this article: Quimbayo JP, Cantor M, Dias MS, et
and partner fidelity. Proceedings of the Royal Society B: Biological al. The global structure of marine cleaning mutualistic
Sciences, 282, 20142925.
networks. Global Ecol Biogeogr. 2018;00:1–13. https://doi.
Tyberghein, L., Verbruggen, H., Pauly, K., Troupin, C., Mineur,
org/10.1111/geb.12780
F., & De Clerck, O. (2012). Bio‐ORACLE: A global environ‐
mental dataset for marine species distribution modelling.
Global Ecology and Biogeography, 21, 272–281. https://doi.
org/10.1111/j.1466-8238.2011.00656.x