You are on page 1of 18

F I S H and F I S H E R I E S , 2013, 14, 60–76

A comparative analysis of chemically induced sex reversal in


teleosts: challenging conventional suppositions

Alistair M Senior & Shinichi Nakagawa

Department of Zoology, University of Otago, PO Box 56, Dunedin 9054, New Zealand

Abstract Correspondence:
Environmental sex reversal (ESR) occurs when extreme environmental factors Alistair M Senior,
Department of
overpower predetermined sexual development. Scientific theory is beginning to
Zoology, University of
acknowledge the potential roles of sex-reversed individuals in influencing population Otago, PO Box 56,
dynamics and driving the evolution of sex-determination mechanisms. ESR is a Dunedin 9054, New
phenomenon that has been widely observed in fish and can be induced by exposing Zealand
individuals to exogenous hormones. However, reports of the susceptibility of fish to Tel.:
+64 (0) 3 479 5046
hormonally induced ESR vary greatly – a concept we termed ‘inducibility’. It has been
Fax:
suggested that variation in inducibility can be attributed to biological differences +64 (0) 3 479 7584
among species of different taxonomic groups (i.e. phylogenetic effects are present). E-mail:
Here, we provide the first quantitative test of this theory, which was achieved by senal826@student.
combining published data with phylogenetic trees, using phylogenetically controlled otago.ac.nz

comparative analysis. Our results confirm that a great amount of variation exists in Received 21 Jul 2011
the reported inducibility of fish. However, species and taxonomic relationships were Accepted 24 Oct 2011
responsible for trivial portions of variation. Rather, we found that sampling
(measurement) errors in combination with methodological differences across studies
accounted for much more variation in inducibility than taxonomy did. Given that our
analysis contains representatives from over 25% of all teleost orders, we conclude
that inducibility is not a taxonomically constrained trait in teleosts. Therefore, we
suggest that the sex-determination mechanisms of most fish are uniformly plastic.
Further, we propose that ESR occurs relatively regularly over evolutionary time in
many teleost species, playing a vital role in the maintenance of homomorphic sex
chromosomes in this taxonomic group.

Keywords Endocrine-disrupting chemicals, environmental sex reversal, hormone,


meta-regression, phylogenetic signal, sex determination

Introduction 61
Methods 62
Data collection 62
Estimating inducibility from experiments 63
Coding estimates of I for PMMs 66
Coding explanatory variables for PMMs 66
Phylogeny construction 66
Collating and analysing estimates of inducibility 67
Calculating measurement error, variance components and phylogenetic heritability 67
Results 68

60 DOI: 10.1111/j.1467-2979.2011.00446.x  2011 Blackwell Publishing Ltd


A comparative study of sex reversal A M Senior and S Nakagawa

Variance components and phylogenetic analyses 68


Model estimates 68
Discussion 69
Phylogenetic insights 69
Wider evolutionary insights 71
Ecological insights 72
Acknowledgements 72
References 72
Supporting information 74
Appendix 75

determination within a species (Yamamoto and


Introduction
Kajishima 1968; Mair et al. 1987; Liu et al. 2008;
In many fish species, as in mammals and birds, sex is Orban et al. 2009).
determined by somatic chromosomes (Devlin and In most cases, the induction of sex reversal via
Nagahama 2002). However, unlike mammals and exposure to environmental hormones or EDCs is
birds, in fish, it is possible for environmental factors favourable so long as sex reversal is complete (i.e.,
to overpower the sex-determining effects of somatic all the individuals within the treated population
chromosomes, leading to environmental sex reversal become one sex). However, the procedure of induc-
(ESR). In some instances, ESR may be invoked by a ing sex reversal itself is not straightforward. For
naturally occurring environmental factor, for exam- example, suboptimal or excessive hormonal dosages
ple, an unusually high temperature (Baroiller et al. may result in the occurrence of intersex individuals
1995, 1999; Grossen et al. 2011). Alternatively, (individuals with malformed/mixed gonads) or high
ESR can result from environmental extremes that levels of mortality (Pandian and Sheela 1995;
have a purely anthropogenic cause, such as artifi- Devlin and Nagahama 2002). As a result, in an
cially manipulated temperature or the presence of attempt to optimize the procedure of sex reversal,
high levels of exogenous hormones and endocrine- hundreds of studies testing the sex-reversing effects
disrupting chemicals (EDCs). Inducing sex reversal in of varying hormonal dosages in numerous fish
fish via chemical exposure has been exploited by species have been published (Devlin and Nagahama
both the aquaculture industry and research scien- 2002; Cnaani and Levavi-Sivan 2009). Yet, many
tists for many years (Pandian and Sheela 1995; of these published studies report conflicting results
Devlin and Nagahama 2002; Pandian and Kiran about the susceptibility of test species to ESR
Kumar 2002; Cnaani and Levavi-Sivan 2009). induced by the application of exogenous chemicals
The aquaculture industry uses ESR to maintain – a concept that will be referred to as inducibility
single-sex populations of fish, which are favourable throughout this article.
for two reasons: (i) some species such as Nile This vast array of published material has
Tilapia (Oreochromis niloticus, Cichlidae) grow more spawned several qualitative reviews on the subject
rapidly if kept in single-sex populations and (ii) of induced sex reversal in fish (Pandian and Sheela
reproduction is far easier to control if males and 1995; Devlin and Nagahama 2002; Pandian and
females are segregated (Cnaani and Levavi-Sivan Kiran Kumar 2002; Cnaani and Levavi-Sivan
2009). Additionally, in some ornamental species 2009). In their review, Pandian and Sheela
such Siamese fighting fish (Betta splendens, Osphro- (1995) suggest that inducibility (susceptibility to
nemidae), individuals of one sex are more valuable sex reversal by exogenous hormones) is a taxonom-
than the other (Piferrer and Lim 1997; Devlin and ically constrained trait, i.e. inducibility varies more
Nagahama 2002). Physiologists have also found it between, than within taxonomic families. Specifi-
useful to exploit the phenomenon of ESR in fish. cally, they suggest that the intensity of treatment
Sex reversal leads to an unusual combination of required to induce sex reversal resulting in a single-
opposed genetic and phenotypic sex, which may be sex population increases in the following taxonomic
exploited to infer the mechanisms underlying sex order: Cichlidae < Cyprinodontidae < Anabantidae

 2011 Blackwell Publishing Ltd, F I S H and F I S H E R I E S , 14, 60–76 61


A comparative study of sex reversal A M Senior and S Nakagawa

< Poeciliidae < Salmonidae < Cyprinidae (Pandian number or as a percentage of males to females
and Sheela 1995). accompanied by a sample size to which this
Given advancements in comparative and analyt- percentage applies, and (iv) ESR must be induced
ical techniques, coupled with the publication of via either a dietary (chemical mixed with food) or a
numerous teleost phylogenies (Mank et al. 2006; bath (chemical mixed with tank water) route of
Fraga et al. 2007; Li et al. 2008; Mayden et al. chemical administration.
2009), it is now possible to combine the vast wealth We then extracted data from each experiment in
of published experimental data on induced ESR in the remaining 155 papers (see Supporting Infor-
fish. Here, we provide the first quantitative test of mation for a complete list). Our experimental level
the theory put forward by Pandian and Sheela was defined as follows. If a study exposed a single
(1995). Our primary aim is to use phylogenetically species to one chemical using one method of
controlled comparative analysis (or phylogenetic exposure (e.g., Nile Tilapia exposed to testosterone
mixed-effects meta-regression; Hadfield and Nakag- via immersion in tanks containing dilutions of the
awa 2010) to examine whether published data hormone) and the experimental populations are
support Pandian and Sheela’s suggestion, i.e. induc- exclusively composed of single or mixed genotypic
ibility is a taxonomically constrained trait. sex, then this is a single experiment; otherwise, the
paper contains multiple experiments (see Fig. 1 for
a schematic diagram of our definition). Splitting
Methods
data down to this experimental level provided us
with a total of 246 experiments (each of which
Data collection
was become a single data point in our final
We began by collecting studies that examined the analysis, i.e. the experimental level was the unit
sex-reversing effects of a varying range of ESR- of analysis).
inducing chemical dosages on teleosts from two
main sources. Firstly, Devlin and Nagahama (2002)
provide a list of 155 studies examining sex reversal
in fish. We took this list to be a comprehensive list of
publications on the subject of experimentally
induced fish sex reversal up to and including
2001. Secondly, we carried out an online search
using the ISI Web of Knowledge database with the
search terms ‘sex reversal’, ‘fish’ and ‘teleost’ for
papers published between 2002 and 2011. This
search produced a total of 98 additional studies, one
of which was a review paper (Orban et al. 2009).
Searching Orban et al.’s (2009) references lead us to
six further studies (total number of studies
found = 259).
Of these 259 studies, we rejected 104 that did not
meet the following four inclusion criteria: (i) the
paper must include a controlled experimental study
on a captive group, either in a laboratory or natural
setting, which attempted to induce ESR in a teleost
species by administering a single sex reversal-
inducing chemical (we rejected combinations of
chemicals, unless the mixture was used to aid
administration; e.g. hormone with ethanol), (ii) the
paper must specify the chemical, fish species, route
of chemical administration and the length of time
over which the chemical was administered, (iii) the Figure 1 Conceptual flow diagram depicting how
paper must report the sex ratio of the experimental the experimental level within an article is defined in our
and control groups after treatment, either as a direct analyses.

62  2011 Blackwell Publishing Ltd, F I S H and F I S H E R I E S , 14, 60–76


A comparative study of sex reversal A M Senior and S Nakagawa

Splitting studies under these criteria still allowed each experiment (essentially a dosage response).
some variables to vary at the experimental level. For Box provides an example of this process, based on
example, within an experiment, the number of days simplified data for clarity.
for which fish were exposed to a chemical could vary
(e.g., 10, 15, 20 days of exposure). In such cases,
Box
when data were collated using phylogenetic meta-
regression models (PMMs), the mean value of vari- Below is an example of some hypothetical data,
ables, which vary at the experimental level, was used. to clarify our procedures for extracting data. Box
Our procedure of extracting data at the experi- Table 1 contains some hypothetical data typical
mental level also meant that it was necessary for of the results reported by the published studies
some experiments to share control data. For exam- that we collected. Presented are the results of
ple, a study may have compared the effects of two three experiments that tested increasing dosages
steroid hormones (e.g. testosterone and estradiol) on of a masculinizing chemical on groups of fish
sex ratio in a single species but only had a single where each group contains ten individuals.
control group. In this case, because the study varied Experiments 1 and 2 are examples of mixed
the chemical, it would be split into two experiments genotypic sex experiments. At dosage zero, both
(one with the results of testosterone on sex ratio and groups have equal numbers of males and
the other with the effects of estradiol on sex ratio). females. Experiment 3 is an example of data
This control data point would be in our analysis from a mono-genotypic sex experiment and at
twice (once for each experiment, it should be noted dosage zero, this experiment reports only female
that pseudo-replication originating from this proce- fish. In all experiments, as the level of hormone a
dure was controlled in our models). Control data group of fish is exposed to increases, the number
were shared between experiments in this way if the of males in the group increases. It is obvious
study was split due to using different chemicals or from visual inspection of the data that Experi-
routes of chemical administration. However, if the ment 1 reports a greater inducibility (I) than
study was split because it varied the subject species Experiment 2, because complete sex reversal of
or subjected both populations of single and mixed the group is reported at a dosage of 1 in
genotypic sex to induced sex reversal, control data Experiment 1, but not until dosage 10 in
were not shared. In these cases, it is conceivable Experiment 2.
that control groups would produce different sex To quantify the differences in reported I, we
ratios between different species, and control groups model the results of each experiment individu-
would certainly produce different sex ratios between ally in the statistical environment R (package
mono and mixed genotypic sex populations. brglm) using a binomial generalized linear model
with a logit-link function. Essentially, we fit the
sex ratio as the response variable and the log10
Estimating inducibility from experiments
dosage as the explanatory variable. Specified in
To estimate inducibility (I), we reanalysed the data R as follows:
(the number of males and females in control and
brglm ðcbind ðMales; FemalesÞ
treatment groups) from each published experiment
 log10 ðDosage þ 1Þ; family ¼ binomialÞ
separately. Using the R (version 2.13.0, R Develop-
ment Core Team 2011) function brglm in the This analysis produces a single slope estimate
package brglm (Kosmidis 2007), we fitted a bias- for each experiment, which is the response of the
reduced binomial generalized linear model (GLM) group sex ratio to exposure to an increasing
with a logit-link function. The response variable dosage of chemical (essentially a dosage re-
was the proportion of males in the exposed group at sponse). Of the two mixed-sex experiments, this
the end of the treatment; the explanatory variable estimate is largest for Experiment 1 (GLM
was the log10 dosage of the sex-reversing chemical est. = 10.3), rather than for Experiment 2
that the group was exposed to (or more precisely, (GLM est. = 2.9). However, overall the largest
log10 (x + 1) with x being the original dosage). We estimate is that for experiment 3 (GLM
used the parameter estimate for the effect of dosage est. = 92). The reason Experiment 3 has such
on sex ratio and its SE (both on the logit scale), as a large estimate of I is because the intercept
the estimate of I and its uncertainty, respectively for

 2011 Blackwell Publishing Ltd, F I S H and F I S H E R I E S , 14, 60–76 63


A comparative study of sex reversal A M Senior and S Nakagawa

(a) (b)
1.0

4
Proportion of Males (logit scale)

0.8

Proportion of Males
0.6

0.4

–2

0.2

–4

0.0

0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
log10 Dosage +1 log10 Dosage +1

Box Figure 1 The response of group sex ratio to exposure to an increasing dosage of chemical in Experiment 1 (solid line),
Experiment 2 (dashed line) and Experiment 3 (dotted line), as estimated by a binomial generalized linear model (GLM).
These estimates are our measures of inducibility (I). (a) Model estimates directly plotted on the logit scale as given by
the GLM output, these are the values of I that we used in our analyses, (b) the same model estimates back-transformed to
the original scale (proportion of males).

Box (Continued) Box Table 1 Hypothetical data, typical of that reported


by the studies included in our comparative analyses. It
value (number of males at zero) is much lower shows the number of males and females in a group of
than it is in other experiments. This difference in fish exposed to a dosage of masculinizing chemical.
slope estimates owing to differing intercept
values, highlights the importance of splitting Experiment Dosage Males Females
up data at our experimental level (see Methods).
When these model estimates are graphed on the
1 0 5 5
logit scale (Box Fig. 1a), the differences between 0.1 7 3
them are clearly visualized. The estimates of sex 1 10 0
ratio response to increasing dosage on the logit 10 10 0
scale are our estimation of I reported by each 2 0 5 5
experiment. For our real dataset, we repeated 0.1 6 4
this process for all 222 experiments, compress- 1 7 3
ing the data from each experiment to a model 10 10 0

slope to obtain 222 separate estimates of I. 3 0 0 10


When estimates are back-transformed (Box 0.1 7 3
Fig. 1b), it becomes clear how these estimates 1 10 0
10 10 0
reflect the difference in I reported by the exper-
iments.

I), suggesting that the model had not converged.


Of the 246 experiments, which we reanalysed, 24 Once these experiments had been removed, we were
were removed because the data provided were left with 222 estimates of I, across 52 species within
insufficient or the sample size was below our cut- 11 orders (see Fig. 2), from 139 published studies
off (n = 3) for analysis using a GLM. Thus, these (see Data S1). Finally, we split our 222 estimates
data produced an extremely large slope (estimate of into those experiments that induce ESR via a dietary

64  2011 Blackwell Publishing Ltd, F I S H and F I S H E R I E S , 14, 60–76


(a) (Dietary) (b) (Bath)
N N
Family Family
Huso huso x A. ruthenus 2 Cobitidae Misgurnus mizolepis 1
Acipenseridae
Acipenser brevirostrum 1
Anguilla anguilla 4 Puntius conchonius 1
Anguillidae
Anguilla japonica 1 Cyprinidae
Cobitidae M. anguillicaudatus 1 Carassius auratus grandoculis 1
Tinca tinca 2
Danio rerio 3 Heteropneustidae Heteropneustes fossilis 4
Cyprinidae Puntius gonionotus 1
Cyprinus carpio 15 Ictaluridae Ictalurus punctatus 1
Carassius a. auratus 2
Bagridae Pelteobagrus fulvidraco 2 Salmo salar 1
Ictalurus punctatus 5
Ictaluridae
Ictalurus furcatus 2 Salvelinus alpinus 1
Salmo salar 6
Salvelinus namaycush 3 Oncorhynchus kisutch 3
Salvelinus fontinalis 1 Salmonidae
Salmonidae
Salvelinus alpinus 1 Oncorhynchus tshawytscha 3
Oncorhynchus masou 1
Oncorhynchus mykiss 19 Oncorhynchus masou 1
Esox masquinongy 1
Esocidae
Esox lucius 2 Oncorhynchus mykiss 4
Sebastidae Sebastes schlegeli 2
Cyclopteridae Cyclopterus lumpus 1 Osphronemidae Betta splendens 2
Osphronemidae Betta splendens 5
C. nigrofasciatum 2 Cichlasoma nigrofasciatum 1
Tilapia zillii 1

 2011 Blackwell Publishing Ltd, F I S H and F I S H E R I E S , 14, 60–76


Oreochromis aureus 10 Pseudocrenilabrus multicolor 3
Cichlidae O. mossambicus x O. niloticus 1 Cichlidae
Oreochromis mossambicus 11 Oreochromis aureus 1
Oreochromis niloticus 22
Oreochromis spilurus 2 Oreochromis niloticus 7
Perca flavescens 2
Percidae
Sander lucioperca 1 Pomoxis nigromaculatus 3
Serranidae Epinephelus malabaricus 1 Centrarchidae
Pomoxis nigromaculatus 2 Lepomis macrochirus 1
Centrarchidae Micropterus salmonidae 4
Lepomis macrochirus 2 Moronidae Morone chrysops x Morone saxatilis 2
Moronidae Dicentrarchus labrax 10
Siganidae Siganus guttatus 1 Paralichthydae Paralichthys olivaceus 1
Paralichthydae Paralichthys olivaceus 9
Xiphophorus helleri 3 Poeciliidae Poecilia reticulata 4
Poeciliidae Poecilia sphenops 2
Poecilia reticulata 3 Adrianichthyidae Oryzias latipes 4

–2 0 4 –6 0 22
Estimate of Inducibility (I) Estimate of Inducibility (I)

Figure 2 Forest plots of mode inducibility (I) per species (filled circles, random effects) and the meta-analytic mean I across all species (filled squares, fixed effect). (a) The dietary
administration dataset as given by ModelD0. (b) The bath administration dataset as given by ModelB0. Horizontal bars for each mode represent 95% credible intervals (CI). The number of
experiments that estimate is based on (N) and the phylogeny included in the analysis of that dataset are presented alongside each estimate.

65
A comparative study of sex reversal A M Senior and S Nakagawa
A comparative study of sex reversal A M Senior and S Nakagawa

route of administration (n = 172) and those that continuous fixed effects. Variable 6 was coded as
use a bath treatment (n = 50). These two data sets the number of exposures per day i.e. thrice
were analysed separately. Bath dosages (mL L)1 of daily = 3, daily = 1, every 2 days = 0.5, weekly =
water) and dietary dosages (mg kg)1 of food) are 0.14 and so forth. In some instances, authors did
not directly comparable as the ratio of dosage units not state the frequency of exposures; in such cases,
is non-commensurate i.e. mg kg)1 and mL L)1 are we assumed that administration frequency was
not equivalent. That is, the interval between values daily. The age of fish at the beginning of treatment
of the explanatory variable over which our initial was reported as both categorical (e.g. juvenile) and
GLMs were fitted is much larger for bath treatments continuous (14 days post-hatching) forms in the
and thus we would expect a greater estimate of I studies in our data set. Therefore, we initially chose
from bath treatments. to record all ages in categorical form (egg, sac-fry,
We also collected the following parameters asso- swim-up fry, fry, juvenile and adult) and then
ciated with each experiment: (i) the article from convert them to a continuous variable with four
which that experiment originated, (ii) the species, (iii) values, where 0 = egg (or gravid female for live
the type of chemical used, (iv) the direction of bearing species), 0.33 = fry (of any kind),
induced change, (v) whether the exposed population 0.66 = juvenile and 0.99 = adult.
was of mixed or single genotypic sex, (vi) the length of
exposure (number of days), (vii) the age at which fish
received their first treatment and (viii) the frequency
Phylogeny construction
with which fish were exposed to hormones.
Pagel’s k is an estimate of the amount of variation
within a trait, which can be accounted for by the
Coding estimates of I for PMMs
shared evolutionary history of species in an analysis
Each estimate of I was analysed as a single data (Pagel 1999) and hence is the estimate we use in our
point in our PMMs. Because we fitted number of comparative study. To estimate Pagel’s k, we
males relative to females as the response variable, in constructed two phylogenies of the 52 species left
our initial analysis, studies that induced feminiza- in our analysis, one for the dietary dataset (see
tion produced negative slopes. Therefore, all slopes Fig. 2a) and one for the bath dataset (see Fig. 2b).
associated with a feminization study were multiplied These phylogenies were produced by combining
by negative one, making the slopes for masculini- published phylogenies, to work out the taxonomic
zation and those of feminization comparable. relationships across all of the species in our analysis.
We obtained studies, which had phylogenetic rela-
tionships of species at a genus level and used these
Coding explanatory variables for PMMs
papers to construct species level relationships (Lyde-
Each estimate of I in our final analysis was ard et al. 1995; Aoyama et al. 2001; Ruber et al.
associated with eight explanatory variables (or 2004; Takehana et al. 2005; Mank and Avise
predictors), which were fitted as fixed or random 2006a; Borsa et al. 2007; Koedprang et al. 2007).
factors in our analyses. Those variables that were The relationships among genus were then con-
fitted as random categorical factors were: (i) the structed using studies containing the information on
article from which the experiment originated (to family-level relationships (Birstein and DeSalle
control for multiple experiments from within the 1998; Crespi and Fulton 2004; Near et al. 2004,
same article) and (ii) the species. The fixed factors 2005; Orrell and Carpenter 2004; Sloss et al. 2004;
that we fitted were: (i) the type of chemical used, Sparks and Smith 2004; Mank and Avise 2006b;
coded as hormone, inhibitor or other, (ii) the Craig and Hastings 2007; Ku et al. 2007; He et al.
direction of change, (iii) whether the exposed 2011). Finally, once we had constructed trees at
population was mixed or mono-sex, (iv) the number these lower taxonomic order, we grafted together
of days fish were exposed to chemicals for, (v) the our family-level trees in Newick format using the
age at which fish received their first treatment and topology from papers, which specify the relation-
(vi) the frequency with which fish were exposed to ships between teleost families at the order level
chemicals. (Dettai and Lecointre 2005; Mank et al. 2005; Mank
Fixed variables 1, 2 and 3 were fitted as categor- and Avise 2006a; Saitoh et al. 2006; Sullivan et al.
ical and variables 4, 5 and 6 were fitted as 2006; Chen et al. 2007a,b; Fraga et al. 2007;

66  2011 Blackwell Publishing Ltd, F I S H and F I S H E R I E S , 14, 60–76


A comparative study of sex reversal A M Senior and S Nakagawa

(a) (b)
40 40

30 30
Precision (1/SE)

Precision (1/SE)
20 20

10 10

0 0
–60 –40 –20 0 20 40 60 –60 –40 –20 0 20 40 60
Estimate of inducibility (I) via diet Estimate of inducibility (I) via bath

Figure 3 Funnel plots of the estimates of inducibility (I), (given by our initial re-analysis of published experiments
using a binomial GLM) plotted against the precision of that estimate (calculated as 1/SE given by a binomial GLM).
(a) Estimates of I from dietary treatments. (b) Estimates of I from bath treatments. Although these funnel plots appear to
display publication bias, they cannot be interpreted in a conventional manner (see Results for an explanation.

Jondeung et al. 2007; Holcroft and Wiley 2008; Li and a thinning interval of 1000, which gave us
et al. 2008; Mayden et al. 2009). 1000 estimates from each MCMC chain. An inverse
gamma prior for random effects and residuals was
specified for all of our analyses (in MCMCglmm,
Collating and analysing estimates of inducibility
V = 1, nu = 0.002 with V being variance and nu
Our 222 estimates of I were analysed using the R the degree of belief parameter in an inverse Wishart
function MCMCglmm in the package MCMCglmm distribution; Hadfield 2010). We ran three indepen-
(Hadfield 2010). We carried out phylogenetic dent chains for each model. Of the three chains, we
mixed-effects meta-regression (PMMs; Hadfield and present parameter estimates, from that which pro-
Nakagawa 2010) by fitting estimates of I (model duced the lowest DIC.
slopes from binomial GLM) as the response variable To test our model and parameter convergence,
and exploring our eight explanatory variables, to we ran model diagnostics on all PMMs. We began
see which best explain the variation in I estimates by checking for autocorrelation between subse-
observed between experiments. In our PMMs, each quent samples within each MCMC chain for both
estimate of I was weighted based on its precision random and fixed effects. We found that autocor-
(measurement error variance, which was the square relation between subsequent lags was consistently
of SE procured by the GLMs). low (<0.1), suggesting that our MCMC chains
Our initial intercept models (ModelD0 for dietary mixed well. To check the consistency of estimates
data and ModelB0 for bath data) consisted of three produced by our models, we also used Gelman-
random factors alone (article of origin, test species Ruben diagnostics on the three independent
and phylogeny; see Table 1). We then ran a full model MCMC chains run for each model and found
for each data set (ModelDFULL and ModelBFULL), which that results were consistent (point estimates of
contained all of our recorded fixed effects alongside the potential scale reduction factor < 1.1), sug-
the three random effects in the intercept models. gesting that our three independent chains had
Finally, we explored models with combinations of converged.
those fixed effects to find the model favored by
deviance information criterion (DIC); models with
Calculating measurement error, variance
the lowest DIC are preferred (Lunn et al. 2000).
components and phylogenetic heritability
Interactions between fixed effects were not explored
owing to the number of variables present in the In our PMMs, a weight was assigned to each
dataset. estimate of I in the form of sampling (measurement)
We ran all PMMs for fifteen million iterations, error variance. Sampling error variance was calcu-
with a burn-in period of fourteen million iterations lated by squaring the SE of each estimate of I, which

 2011 Blackwell Publishing Ltd, F I S H and F I S H E R I E S , 14, 60–76 67


A comparative study of sex reversal A M Senior and S Nakagawa

was produced by the initial modelling of published family and I exists, then species ID and/or phylog-
results. eny would account for a large proportion of the
The proportion of variation explained by various variation in estimates of I. That is, we would detect
model components (see Table 1), was calculated as a phylogenetic signal. However, both intercept
follows. We began by calculating the total amount models suggest that species and phylogeny account
of variance in each dataset, then subtracted the for <1% of the variation in estimates of I respec-
percentage accounted for by sampling error, fixed tively, hence little phylogenetic signal exists (Pagel’s
effects and partitioned the remaining variation k: dietary data < 0.001 and bath data < 0.001;
among the random effects and the residual variation Pagel 1999; see Table 1). Therefore, there appears
as estimated by each model (see Table A2). to be little or no difference in the inducibility of the
Pagel’s k is an estimate of the degree of phylo- different teleost species in our analyses (see Fig. 2).
genetic signal present within a given trait (Pagel Rather than biological differences accounting for
1999). Pagel’s k, which is equivalent to phyloge- variance in I, it appears that sampling error
netic heritability H2 (Lynch 1991), was calculated variance accounts for most of the variation in the
as given by Equation (1), where r2Phylogeny is the published literature (see Table 1). Thus, the varia-
model estimate for the phylogeny, r2Species is the tion in I that we report is primarily driven by
model estimate for species ID, r2Article is the model differences in experimental design between studies
estimate for the Article ID and r2Residual is the model (i.e., variation in sample size and the number of
estimate for residual variance. different dosages (e.g. 0, 0.1 or 0, 0.1, 1) experi-
mentally tested).
r2Phylogeny
Pagel’s; k ¼
r2Phylogeny þ rSpecies þ r2Article
2 þ r2Residual
ð1Þ
Model estimates
Our intercept model (ModelD0) suggests that, on
Results
average, fish have an inducibility of 1.741 via
dietary treatments (ModelD0 I: mode = 1.741,
Variance components and phylogenetic analyses
mean = 1.706, 95% credible interval (CI) =
Our initial reanalysis of estimates of I using bino- 1.199–2.129, see Fig. 2). This estimate means that,
mial GLMs showed considerable variation among on average, if we expose a mixed-sex population
the published data from both dietary and bath (1:1 males/females) to an increase in dosage of an
datasets, some of which were associated with a large
N
SE (Fig. 3). If conventionally interoperated, the
funnel plots (Fig. 3) appear to show a sign of
publication bias (i.e. asymmetries in funnels; Egger Mono Sex 32
et al. 1997). Because these plots display model
slopes (I) estimates (rather than conventional effect
sizes), we expect these values to be positive. Initially,
Mixed Sex 140
negative slopes resulted from feminization experi-
ments where the exposed population became
female-skewed. However, we multiplied these values
by negative one to make them comparable to Contrasts
(Mono - Mixed)
masculinization experiments. In this instance, neg-
ative slopes (i.e., I) can be interpreted as an
0 6
incidence of ‘counterintuitive sex change’. For Estimate of Inducibility (I)

example, an androgen (e.g., testosterone) is applied,


Figure 4 The mode estimates of inducibility (I) and 95%
but the exposed population becomes female-biased.
credible interval (CI) of mono and mixed genotypic sex
We began by constructing an intercept model for
groups of fish based on our dietary dataset, estimated
each dataset (diet data = ModelD0 and bath from the DIC favoured model (ModelDDIC), alongside the
data = ModelB0), which fitted article identity, spe- number of experiments that the estimate is based on.
cies identity and our phylogeny as random effects The contrast (pair-wise comparison) between estimates is
(see Table 1). If a relationship between taxonomic also presented.

68  2011 Blackwell Publishing Ltd, F I S H and F I S H E R I E S , 14, 60–76


A comparative study of sex reversal A M Senior and S Nakagawa

Table 1 The proportion of total variance (r2) in estimates of I explained by fixed effects (r2FE ), random effects (r2Article ,
r2Species , and r2Phylogeny ) and sampling error variance (r2SE ) explained by our null, DIC favored and full models.

Fixed effects
Data set Model name in model r2FE (%) r2Article (%) r2Species (%) r2Phylogeny (%) r2SE (%) r2Residual (%) DIC

Diet ModelD0 NA NA 0.024 0.004 0.007 87.74 12.16 738.7


ModelDDIC Genotypic sex 0.581 0.013 0.007 0.008 87.74 11.60 721.7
ModelDFULL All 0.229 0.014 0.009 0.009 87.74 11.95 735.7

Bath ModelB0 NA NA 0.000 0.031 0.000 35.05 64.92 353.9


ModelBFULL All 8.156 0.138 0.000 0.000 35.05 56.62 339.1

ESR-inducing chemical from 0 mg kg)1 of food to significant. This lack of statistical significance is
9 mg kg)1 of food, approximately 35% of the owing to large CIs surrounding our estimates,
individuals will reverse sex. Our intercept model which are probably a result of the limited sample
for bath data (ModelB0) suggests that via bath size in the bath dataset (See Table A1).
treatments fish have an inducibility of 6.231
(ModelB0 I: mode = 6.231 mean = 7.258, 95%
Discussion
CI = 1.906–12.896, see Fig. 2). This estimate can
be interpreted as on average, if we expose a mixed- We set out to quantitatively test the hypothesis
sex population (1:1 males/females) to an increase in proposed by Pandian and Sheela (1995) that
dosage of an ESR-inducing chemical from a con- susceptibility to induced sex reversal by exposure
centration of 0 mL L)1 of water to 9 mL L)1 of to exogenous hormones (inducibility, I) is a taxo-
water, nearly 50% of individuals will reverse sex; i.e. nomically constrained trait within teleosts. We
complete sex reversal (although it should be noted combined estimates of inducibility across species,
that, owing to the smaller sample size of the bath from the published literature with a phylogenetic
data set, there is a large credible interval surround- tree by using phylogenetically controlled mixed-
ing this estimate). effects meta-regressions (PMMs). Contrary to what
With regard to the methods used to induce ESR, Pandian and Sheela (1995) proposed, our results
for the dataset containing dietary treatments, DIC suggest that inducibility is a relatively uniform trait
favours a model that only contains the genotypic among those taxa included in our analyses at all
sex of the exposed population as a fixed factor and taxonomic levels (Fig. 2).
the three random effects in the intercept model
(ModelDDIC, see Table 1). The ModelDDIC suggests
that significantly more individuals switch sex in
Phylogenetic insights
populations of a single genotypic sex than in
populations of mixed genotypic sex, for the same Our results suggest that once variation in sample
increase in the dosage of sex-reversing chemical size and within-study methodology has been ac-
(see Fig. 4; ModelDDIC [MONOSEX – MIXEDSEX] I: mean = counted for, there is little evidence for phylogenetic
1.692 mode = 1.567, 95% CI = 0.86–2.579, see signal in the inducibility of those species included in
Table A1). Other than the difference between mixed our analyses (see Table 1). Recently, a similar
and mono-genotypic sex treatment groups, our full analysis was published by Guénard et al. (2011),
model for the dietary dataset (ModelDFULL) did not which looked at the relationship between phylogeny
detect any significant differences between any and susceptibility of species to toxic chemicals. In
moderator variables (see Table A1). stark contrasts to what our study finds, Guénard
For the dataset containing estimates of inducibil- et al. (2011) show that strong phylogenetic inertia
ity (I) via bath treatments, DIC favours the full is present in their dataset, with regard to suscepti-
model that fits all fixed effects plus the three random bility to toxins. However, it is not surprising that
effects in the intercept models (ModelBFULL, see Guénard et al. (2011) detected phylogenetic signal
Table 1). However, none of the fixed effects we because their analysis looked at species from a wider
investigated in the bath dataset are statistically taxonomic background than our study (e.g. their

 2011 Blackwell Publishing Ltd, F I S H and F I S H E R I E S , 14, 60–76 69


A comparative study of sex reversal A M Senior and S Nakagawa

analysis includes amphibians and crustaceans effects present independent of phylogeny (i.e.
alongside fish). When species from very high-level between species, I varies highly regardless of phy-
taxonomic groups are included in comparative logenetic relationship) or (ii) when there is little
analysis, phylogenetic signal is very likely to be variation among species (Losos 2011). There is a
detected because of the clear biological differences possibility that our analysis suffers from the problem
between the different groups, even though there is of publication bias (Rothstein et al. 2005), which
very little variation within those groups. For exam- results in between-species level effects being over-
ple, including amphibian species in the current looked. That is, our dataset may not include species
analyses would have resulted in detecting strong where ESR is difficult or even impossible because
phylogenetic signal due to differences between the such results are unlikely to be published or high
inducibility of amphibians and fish, rather than mortality causes the sample size of such studies to
differences between the inducibility of fish groups. fall below our cut-off (see Methods section above).
Given that we did not detect any phylogenetic In response to this argument, we would like to
signal in our dataset, the question becomes ‘why make three comments. Firstly, we acknowledge that
not?’ There are two potential reasons: (i) it is it is possible to find examples where sex reversal is
possible that phylogenetic signal does exist in the impossible or difficult [e.g. masculinization in the
dataset, but our statistical procedures did not detect black molly (Poecilia sphenops, Poeciliidae) (George
it owing to inaccuracy or poor sensitivity or and Pandian 1998)].
(ii) phylogenetic signal does not exist in susceptibil- However, given the relatively uniform estimates
ity to induced ESR in teleosts in general. With of I that our analyses report, there is no reason to
regard to the first point, observing a number of believe that such between-species level effects are
expected results indicates to us that our analytical widespread. Secondly, even if between-species
procedures were probably sound. Firstly, our anal- effects are present, it is still unlikely that they are
ysis (PMMs) detected the difference in I between related to higher-level phylogenetic effects (e.g.
populations of a single genotypic sex vs. those of family-level effects) because if this was the case,
mixed genotypic sex (see Fig. 4). Populations of our analysis would have detected some phylogenetic
mixed genotypic sexes have (on average) an equal signal. Finally, we have sufficient evidence to assess
sex ratio (1:1) at dosage zero (e.g. Komatsu et al. the theory proposed by Pandian and Sheela (1995),
2006), meaning the intercept on the natural scale is who suggest the relationships between the relative
0.5, whereas those of single genotypic sex have only inducibility of six teleost families from four orders.
one sex at dosage zero (1:0) (e.g. Kang et al. 2008), Of those six families, we have representatives from
meaning the intercept on the natural scale is 0. four (and all orders), including the families that
Hence, the slopes produced by our initial binomial Pandian and Sheela (1995) suggest are the most
models should be shallower when based on data inducible (Cichlidae) and the least inducible (Cyp-
from mixed-sex populations, than mono-sex popu- rinidae), and we find no differences between the
lations (see Box). Secondly, the same can be said for inducibility of these families.
studies that induce sex reversal using bath treat- In all, Teleostei totals forty orders (Helfman et al.
ments as opposed to those that use dietary treat- 2009) of which we have representatives from
ments. Realistically, bath treatments expose fish to eleven, and we would argue that representation of
much higher dosages of sex-reversing chemical over one quarter of teleost orders is reasonable.
than dietary treatments (see Methods section Further, we believe that our analysis is as thorough
above). Therefore, the estimate of I produced from an analysis as is possible given the data currently
the bath dataset should be larger than that from the available. Based on our results and the reasonable
diet dataset, and we show that it is (see Fig. 2). representation of numerous taxa, we believe it is
Given that our procedures detected these most likely that little variation in I exists between
expected differences between ESR-inducing tech- teleosts. If this was the case, why would it have been
niques, we suggest that our methodology would suggested that such taxonomic differences are
have detected any phylogenetic signal if at all present (Pandian and Sheela 1995)? Our analysis
present. Hence, it seems unlikely that phylogenetic confirms that considerable variability in estimates of
signal in I exists in the current dataset. Phylogenetic I within the published literature exists, and it seems
signal will typically not be detected under two likely that phylogenetic signal has been inferred
circumstances: (i) when there are between-species from this observed variation. We believe that this

70  2011 Blackwell Publishing Ltd, F I S H and F I S H E R I E S , 14, 60–76


A comparative study of sex reversal A M Senior and S Nakagawa

inference was made because, when variation is Not only is it possible that sex reversal events
observed in numerous studies across taxa, biological could induce a transition between male and female
variability is probably the most ‘satisfying’ explana- heterogamety, but it has also been suggested that
tion available. In other words, biologists tend to sex-reversed individuals play a role in maintaining
ascribe observed variation to biological rather than the structure of sex chromosomes in their species
statistical phenomena. (Perrin 2009; Stelkens and Wedekind 2010; Wede-
However, in the case of ESR, it seems more likely kind 2010). Perrin (2009) hypothesized that sex
that sampling errors (i.e. variation owing to limited reversal would allow sex chromosomes to recom-
sample size and study designs) drive a large portion bine in the heterogametic sex (XY or ZW). When
of variation in published estimates of inducibility. recombinations of this type occur, they allow the
A further possibility is that methodological differ- chromosome unique to that genotypic sex (Y or W)
ences account for the variation in published data. to be purged of deleterious mutations (Perrin 2009).
While we were able to control for many of the Perrin (2009) argues that ultimately such events
differences between methodologies (e.g. type of lead to the homomorphic sex chromosomes present
chemical used), it was not possible to account for in fish and amphibian species.
unstated aspects of the methods researchers used. To date, there has been a large body of evidence
For example, some studies do not directly report the suggesting that recombination rates between sex
quantity of food experimental fish are fed, reporting chromosomes are dependent on phenotypic sex
that fish were fed ad libitum or to satiation (Arslan (Perrin 2009; Wedekind 2010; Matthias et al.
and Phelps 2004), whereas other studies report 2011), the first requirement for Perrin’s (2009)
feeding fish at a fixed percentage of body weight hypothesis. However, the prevalence of mass sex
(Fujioka 2002). This inconsistency would have a reversal events in wild populations is largely
particularly strong bearing on the estimates of I unknown (Wedekind 2010), the second require-
reported by the large number of studies that ment for Perrin’s (2009) hypothesis. Alho et al.
administer hormones via dietary supplementation. (2010) demonstrated that in the common frog
(Rana temporaria, Ranidae), sex reversals do occur in
wild populations of amphibians. Given how uni-
Wider evolutionary insights
formly plastic the mechanisms of sex determination
Comparative analyses such as the current study can appear to be in fish according to our results, we
provide accurate characterizations of general bio- would suggest that numerous species of fish also
logical and taxonomic trends (Losos 2011). Mank have the potential to undergo regular sex reversal.
et al. (2006) used comparative methods to demon- Over evolutionary time, such events are likely to be
strate the wide diversity of sex-determining mech- common in populations of most teleosts, ultimately
anisms in fish. In particular, Mank et al. (2006) playing a role in the maintenance of homomorphic
have suggested that the source of this diversity is sex chromosomes in fish lineages.
switched between somatic mechanisms of sex The view that biologists hold of sex as a trait is
determination (i.e. male heterogamety, XY, and currently evolving and it is no longer accepted that
female heterogamety, ZW), which occur fairly one of two distinct phenotypes (male or female) is
readily over evolutionary time. Further, it has been determined by a single factor. It is clear that even
suggested that sex reversal events, which lead to when sex appears to be determined by one gene on
individuals in the population that bear an opposing a sex chromosome, a complex interplay between
sexual genotype and phenotype, can drive the genome and environment determines sexual char-
evolution of a change in sex-determining mecha- acteristics (Mittwoch 2006). Ultimately, sex should
nism (Mank and Avise 2006a; Grossen et al. 2011). be viewed as a reaction norm like many other
Assuming the absence of widespread between-spe- phenotypic traits, where the resulting phenotype is
cies level effects, our results suggest that the dependent on the reaction of genes to environment
mechanisms of sex determination in teleosts are (Ah-King and Nylin 2010). The concept of sex as a
relatively plastic and open to environmental inter- reaction norm is not immediately obvious in some
ferences, which may ultimately lead to sex reversals. taxa such as mammals, where sex chromosomes
We propose that their potential ubiquity explains play a very stable role in the sex-determining
the diversity of sex-determining mechanisms oper- process and are unperturbed by most environmental
ating in teleosts. factors (Ah-King and Nylin 2010).

 2011 Blackwell Publishing Ltd, F I S H and F I S H E R I E S , 14, 60–76 71


A comparative study of sex reversal A M Senior and S Nakagawa

However, sex determination by a gene–environ- References


ment interaction is obvious in fish, where sexual
Ah-King, M. and Nylin, S. (2010) Sex in an evolutionary
development is plastic and environment plays a perspective: just another reaction norm. Evolutionary
strong role in determining sexual phenotype, as also Biology 37, 234–246.
demonstrated by our analyses. We have not dis- Alho, J.S., Matsuba, C. and Merila, J. (2010) Sex reversal
cussed the adaptive or evolutionary advantage that and primary sex ratios in the common frog (Rana
plasticity in sex determination offers. Nonetheless, temporaria). Molecular Ecology 19, 1763–1773.
given the apparent prevalence of plasticity in sexual Aoyama, J., Nishida, M. and Tsukamoto, K. (2001)
development among teleosts, it would seem that fish Molecular phylogeny and evolution of the freshwater
are an ideal taxa in which to explore the evolution- eel, genus Anguilla. Molecular Phylogenetics and Evolu-
ary significance of plastic sex determination. tion 20, 450–459.
Arslan, T. and Phelps, R.P. (2004) Production of monosex
male black crappie, Pomoxis nigromaculatus, populations
by multiple androgen immersion. Aquaculture 234,
Ecological insights 561–573.
Numerous studies have reported ESR in wild fish Baroiller, J.F., Chourrout, D., Fostier, A. and Jalabert, B.
(1995) Temperature and sex-chromosomes govern
populations as a result of exposure to effluents
sex-ratios of the mouthbrooding cichlid fish Oreochr-
containing EDCs (Hotchkiss et al. 2008). Such
omis niloticus. Journal of Experimental Zoology 273,
effects that pollutants could have on fish population 216–223.
dynamics were modelled by Hurley et al. (2004). Baroiller, J.F., Guiguen, Y. and Fostier, A. (1999) Endo-
They found that in systems of male heterogametic crine and environmental aspects of sex differentiation in
sex determination (XY), environmental masculini- fish. Cellular and Molecular Life Sciences 55, 910–931.
zation could lead to population collapse, with the Birstein, V.J. and DeSalle, R. (1998) Molecular phylogeny
population becoming dependent on the masculiniz- of Acipenserinae. Molecular Phylogenetics and Evolution
ing chemical to maintain phenotypic males. 9, 141–155.
In contrast, environmental feminization in male Borsa, P., Lemer, S. and Aurelle, D. (2007) Patterns of
heterogametic systems was not quite so perilous lineage diversification in rabbitfishes. Molecular Phylog-
enetics and Evolution 44, 427–435.
(the converse is true for systems of female heterog-
Chen, D.L., Guo, X.G. and Nie, P. (2007a) Non-monophyly
ametic sex determination – i.e. ZW). In the light of
of fish in the Sinipercidae (Perciformes) as inferred from
our results, researchers investigating the effects of cytochrome b gene. Hydrobiologia 583, 77–89.
EDCs in aquatic ecosystems should not assume that Chen, W.J., Ruiz-Carus, R. and Orti, G. (2007b) Relation-
the effects, which they observe, are limited to just ships among four genera of mojarras (Teleostei :
one or two fish species. Given that we observe Perciformes : Gerreidae) from the western Atlantic and
uniform inducibility across all teleost families their tentative placement among percomorph fishes.
(included in our analyses), if ecologists detect Journal of Fish Biology 70, 202–218.
chemically induced ESR in one species, it seems Cnaani, A. and Levavi-Sivan, B. (2009) Sexual develop-
likely that other fish species are also affected. ment in fish, practical applications for aquaculture.
Sexual Development 3, 164–175.
Craig, M.T. and Hastings, P.A. (2007) A molecular
phylogeny of the groupers of the subfamily Epinepheli-
nae (Serranidae) with a revised classification of the
Acknowledgements
Epinephelini. Ichthyological Research 54, 1–17.
We would like to thank Jiahui Nat Lim for his help in Crespi, B.J. and Fulton, M.J. (2004) Molecular systematics
constructing our phylogeny, as well as Eduardo of Salmonidae: combined nuclear data yields a robust
Santos, Manna Warburton and two anonymous phylogeny. Molecular Phylogenetics and Evolution 31,
658–679.
reviewers for their comments and thoughts on this
Dettai, A. and Lecointre, G. (2005) Further support for the
research. We would also like to thank the University
clades obtained by multiple molecular phylogenies in the
of Otago, Zoology Department and the Marsden
acanthomorph bush (vol 328, pg 674, 2005). Comptes
Fund (UOO0812), New Zealand, for funding this Rendus Biologies 328, 1031–1032.
research. Additionally, we would like to thank Dunja Devlin, R.H. and Nagahama, Y. (2002) Sex determination
Lamatsch, Mark Lokman and Gerry Closs for helping and sex differentiation in fish: an overview of genetic,
to write a funding proposal, without which none of physiological, and environmental influences. Aquacul-
this research would have been possible. ture 208, 191–364.

72  2011 Blackwell Publishing Ltd, F I S H and F I S H E R I E S , 14, 60–76


A comparative study of sex reversal A M Senior and S Nakagawa

Egger, M., Smith, G.D., Schneider, M. and Minder, C. Kang, I.J., Yokota, H., Oshima, Y., Tsuruda, Y., Shimasaki,
(1997) Bias in meta-analysis detected by a simple, Y. and Honjo, T. (2008) The effects of methyltestoster-
graphical test. BMJ 315, 629–634. one on the sexual development and reproduction of
Fraga, E., Schneider, H., Nirchio, M., Santa-Brigida, E., adult medaka (Oryzias latipes). Aquatic Toxicology 87,
Rodrigues, L.F. and Sampaio, I. (2007) Molecular 37–46.
phylogenetic analyses of mullets (Mugilidae, Mugilifor- Koedprang, W., Na-Nakorn, U., Nakajima, M. and Tanig-
mes) based on two mitochondrial genes. Journal of uchi, N. (2007) Evaluation of genetic diversity of eight
Applied Ichthyology 23, 598–604. grouper species Epinephelus spp. based on microsatellite
Fujioka, Y. (2002) Effects of hormone treatments and variations. Fisheries Science 73, 227–236.
temperature on sex-reversal of Nigorobuna Carassius Komatsu, T., Nakamura, S. and Nakamura, M. (2006)
carassius grandoculis. Fisheries Science 68, 889–893. Masculinization of female golden rabbitfish Siganus
George, T. and Pandian, T.J. (1998) Dietary administration guttatus using an aromatase inhibitor treatment
of androgens induces sterility in the female-heterogametic during sex differentiation. Comparative Biochemistry
black molly, Poecilia sphenops (Cuvier & Valenciennes, and Physiology C-Toxicology & Pharmacology 143, 402–
1846). Aquaculture Research 29, 167–175. 409.
Grossen, C., Neuenschwander, S. and Perrin, N. (2011) Kosmidis, I. (2007) Bias reduction in exponential family
Temperature-dependant turnovers in sex-determination nonlinear models. PhD, The University of Warwick,
mechanisms: a quantitative model. Evolution 65, 64–78. Warwick, 145 pp.
Guénard, G., von der Ohe, P., de Zwart, D., Legendre, P. Ku, X.Y., Peng, Z.G., Diogo, R. and He, S.P. (2007) MtDNA
and Lek, S. (2011) Using phylogenetic information to phylogeny provides evidence of generic polyphyleticism
predict species tolerances to toxic chemicals. Ecological for East Asian bagrid catfishes. Hydrobiologia 579, 147–
Applications, doi: 10.1890/10-2242.1. 159.
Hadfield, J.D. (2010) MCMC methods for multi-response Li, C.H., Lu, G.Q. and Orti, G. (2008) Optimal data
generalized linear mixed models: the MCMCglmm R partitioning and a test case for ray-finned fishes (Actin-
package. Journal of Statistical Software 33, 1–22. opterygii) based on ten nuclear loci. Systematic Biology
Hadfield, J.D. and Nakagawa, S. (2010) General quantita- 57, 519–539.
tive genetic methods for comparative biology: phyloge- Liu, Z.H., Zhang, Y.G. and Wang, D.S. (2008) Studies on
nies, taxonomies and multi-trait models for continuous feminization, sex determination, and differentiation of
and categorical characters. Journal of Evolutionary Biol- the Southern catfish, Silurus meridionalis: a review. Fish
ogy 23, 494–508. Physiology and Biochemistry 36, 223–235.
He, A., Luo, Y., Yang, H., Liu, L., Li, S. and Wang, C. Losos, J.B. (2011) Seeing the forest for the trees: the
(2011) Complete mitochondrial DNA sequences of the limitations of phylogenies in comparative biology. The
Nile tilapia (Oreochromis niloticus) and Blue tilapia American Naturalist 177, 709–727.
(Oreochromis aureus): genome characterization and phy- Lunn, D.J., Thomas, A., Best, N. and Spiegelhalter, D.
logeny applications. Molecular Biology Reports 38, (2000) WinBUGS – a Bayesian modelling framework:
2015–2021. concepts, structure, and extensibility. Statistics and
Helfman, G.S., Collette, B.B., Facey, D.E. and Bowen, B.W. Computing 10, 325–337.
(2009) The Diversity of Fishes; Biology, Evolution and Lydeard, C., Wooten, M.C. and Meyer, A. (1995) Mole-
Ecology, 2nd edn. Wiley and Sons Ltd, Chichester. pp. cules, morphology, and area-cladograms – a cladistic
263–301. and biogeographic analysis of Gambusia (Teleostei,
Holcroft, N.I. and Wiley, E.O. (2008) Acanthuroid rela- Poeciliidae). Systematic Biology 44, 221–236.
tionships revisited: a new nuclear gene-based analysis Lynch, M. (1991) Methods for the analysis of comparative
that incorporates tetraodontiform representatives. Ich- data in evolutionary biology. Evolution 45, 1065–1080.
thyological Research 55, 274–283. Mair, G.C., Penman, D.J., Scott, A., Skibinski, D.O.F. and
Hotchkiss, A.K., Rider, C.V., Blystone, C.R. et al. (2008) Beardmore, J.A. (1987) Hormonal sex-reversal and the
Fifteen years of ‘‘wingspread’’ – environmental endo- mechanisms of sex determination in Oreochromis.
crine disrupters and human and wildlife health: where Proceedings of The World Symposium on Selection, Hybrid-
we are today and where we need to go. Toxicological ization and Genetic Engineering in Aquaculture, Bordeux 2,
Sciences 105, 235–239. 27–30.
Hurley, M.A., Matthiessen, P. and Pickering, A.D. (2004) Mank, J.E. and Avise, J.C. (2006a) The evolution of
A model for environmental sex reversal in fish. Journal of reproductive and genomic diversity in ray-finned fishes:
Theoretical Biology 227, 159–165. insights from phylogeny and comparative analysis.
Jondeung, A., Sangthong, P. and Zardoya, R. (2007) The Journal of Fish Biology 69, 1–27.
complete mitochondrial DNA sequence of the Mekong Mank, J.E. and Avise, J.C. (2006b) Comparative phyloge-
giant catfish (Pangasianodon gigas), and the phylogenetic netic analysis of male alternative reproductive tactics in
relationships among Siluriformes. Gene 387, 49–57. ray-finned fishes. Evolution 60, 1311–1316.

 2011 Blackwell Publishing Ltd, F I S H and F I S H E R I E S , 14, 60–76 73


A comparative study of sex reversal A M Senior and S Nakagawa

Mank, J.E., Promislow, D.E.L. and Avise, J.C. (2005) Ruber, L., Britz, R., Tan, H.H., Ng, P.K.L. and Zardoya, R.
Phylogenetic perspectives in the evolution of parental (2004) Evolution of mouthbrooding and life-history
care in ray-finned fishes. Evolution 59, 1570–1578. correlates in the fighting fish genus Betta. Evolution 58,
Mank, J.E., Promislow, D.E.L. and Avise, J.C. (2006) 799–813.
Evolution of alternative sex-determining mechanisms Saitoh, K., Sado, T., Mayden, R.L. et al. (2006) Mitoge-
in teleost fishes. Biological Journal of the Linnean Society nomic evolution and interrelationships of the cyprini-
87, 83–93. formes (Actinopterygii : Ostariophysi): the first evidence
Matthias, S., Horn, A., Grossen, C. et al. (2011) Ever- toward resolution of higher-level relationships of the
young sex chromosomes in European tree frogs. PLOS world’s largest freshwater fish clade based on 59 whole
Biology 9, e1001062. mitogenome sequences. Journal of Molecular Evolution
Mayden, R.L., Chen, W.J., Bart, H.L. et al. (2009) Recon- 63, 826–841.
structing the phylogenetic relationships of the earth’s Sloss, B.L., Billington, N. and Burr, B.M. (2004) A molec-
most diverse clade of freshwater fishes-order Cyprini- ular phylogeny of the Percidae (Teleostei, Perciformes)
formes (Actinopterygii: Ostariophysi): a case study based on mitochondrial DNA sequence. Molecular Phy-
using multiple nuclear loci and the mitochondrial logenetics and Evolution 32, 545–562.
genome. Molecular Phylogenetics and Evolution 51, Sparks, J.S. and Smith, W.L. (2004) Phylogeny and
500–514. biogeography of cichlid fishes (Teleostei : Perciformes :
Mittwoch, U. (2006) Sex is a threshold dichotomy mim- Cichlidae). Cladistics 20, 501–517.
icking a single gene effect. Trends in Genetics 22, 96– Stelkens, R.B. and Wedekind, C. (2010) Environmental sex
100. reversal, Trojan sex genes, and sex ratio adjustment:
Near, T.J., Bolnick, D.I. and Wainwright, P.C. (2004) conditions and population consequences. Molecular
Investigating phylogenetic relationships of sunfishes Ecology 19, 627–646.
and black basses (Actinopterygii : Centrarchidae) Sullivan, J.P., Lundberg, J.G. and Hardman, M. (2006)
using DNA sequences from mitochondrial and nuclear A phylogenetic analysis of the major groups of catfishes
genes. Molecular Phylogenetics and Evolution 32, 344– (Teleostei : Siluriformes) using rag1 and rag2 nuclear
357. gene sequences. Molecular Phylogenetics and Evolution
Near, T.J., Bolnick, D.I. and Wainwright, P.C. (2005) Fossil 41, 636–662.
calibrations and molecular divergence time estimates in Takehana, Y., Naruse, K. and Sakaizumi, M. (2005)
centrarchid fishes (Teleostei : Centrarchidae). Evolution Molecular phylogeny of the medaka fishes genus Oryzias
59, 1768–1782. (Beloniformes : Adrianichthyidae) based on nuclear and
Orban, L., Sreenivasan, R. and Olsson, P.E. (2009) Long mitochondrial DNA sequences. Molecular Phylogenetics
and winding roads: testis differentiation in zebrafish. and Evolution 36, 417–428.
Molecular and Cellular Endocrinology 312, 35–41. Wedekind, C. (2010) Searching for sex-reversals to explain
Orrell, T.M. and Carpenter, K.E. (2004) A phylogeny of the population demography and the evolution of sex chro-
fish family Sparidae (porgies) inferred from mitochon- mosomes. Molecular Ecology 19, 1760–1762.
drial sequence data. Molecular Phylogenetics and Evolu- Yamamoto, T.O. and Kajishima, T. (1968) Sex hormone
tion 32, 425–434. induction of sex reversal in the goldfish and evidence for
Pagel, M. (1999) Inferring the historical patterns of male heterogamity. Journal of Experimental Zoology 168,
biological evolution. Nature 401, 877–884. 215–222.
Pandian, T.J. and Kiran Kumar, S. (2002) Hormonal sex
reversal in fish. Journal of Endocrinology and Reproduction
6, 2–3.
Supporting Information
Pandian, T.J. and Sheela, S.G. (1995) Hormonal induction
of sex reversal in fish. Aquaculture 138, 1–22. Additional Supporting Information may be found in
Perrin, N. (2009) Sex reversal: a fountain of youth for sex the online version of this article:
chromosomes? Evolution 63, 3043–3049. Data S1. Data bibliography.
Piferrer, F. and Lim, L.C. (1997) Application of sex reversal
Please note: Wiley-Blackwell are not responsible
technology in ornamental fish culture. Aquarium Science
for the content or functionality of any supporting
and Conservation 1, 113–118.
R Development Core Team (2011) R: A language and
materials supplied by the authors. Any queries
environment for statistical computing, Version 2.13.0. (other than missing material) should be directed to
Available at http://www.r-project.org. the corresponding author for the article.
Rothstein, H.R., Sutton, A.J. and Borenstein, M. (2005)
Publication Bias in Meta-Analysis. Wiley, Chichester.

74  2011 Blackwell Publishing Ltd, F I S H and F I S H E R I E S , 14, 60–76


A comparative study of sex reversal A M Senior and S Nakagawa

Appendix
Table A1 The modal estimates of inducibility (I) and 95% credible intervals (CI) of contrasts for factorial variables,
the modal estimates of I and 95% CI for numeric variables and the models that those estimates are calculated from.
Estimates with CIs that do not cross zero are considered statistically significant. Inhibitor chemical types were not
present in the bath data set.

Dietary data Bath data

Variable/contrast Mode I CI Model Mode I CI Model

Mono sex – mixed sex (genotypic sex) 1.567 0.860–2.579 ModelDDIC 3.406 )4.775–12.839 ModelBFULL
Hormone – inhibitor (chemical type) 0.451 )0.605–2.148 ModelDFULL NA NA NA
Hormone – ‘other’ (chemical type) 0.806 )2.418–4.912 ModelDFULL 3.592 )13.585–21.362 ModelBFULL
Inhibitor – ‘other’ (chemical type) 0.162 )3.806–4.080 ModelDFULL NA NA NA
Feminization – masculinization )0.883 )4.004–1.174 ModelDFULL 5.467 )0.698–10.691 ModelBFULL
(direction of sex change)
Age 0.188 )11.474–9.864 ModelDFULL 8.412 )15.966–24.475 ModelBFULL
Frequency 0.015 )4.098–13.382 ModelDFULL 2.164 )12.149–9.165 ModelBFULL
Length of treatment days 2.632 )16.264–24.342 ModelDFULL )0.078 )0.339–0.214 ModelBFULL

Table A2 The model mean and mode estimates of I for fixed and random effects ± 95% confidence intervals (CI), as
given by intercept, DIC favoured and Full PMMs for both dietary and bath data.

Fixed effects Mean I Mode I CI

ModelDDIC
Intercept (mono-genotypic sex groups) 3.121 2.960 2.224–3.953
Mixed genotypic sex 1.429 1.385 0.994–1.853

ModelDFULL
Intercept (hormone, mono-genotypic sex and 2.828 3.079 1.260–4.307
feminization studies)
Inhibitor 2.068 2.146 0.529–3.545
‘Other’ chemicals 1.706 1.333 )1.771–6.041
Masculinization (diff. between masculinization 0.127 0.176 )0.510–0.835
and feminization)
Mixed genotypic sex (diff. between mixed and )1.907 )1.800 )3.053–)0.967
mono-genotypic sex groups)
Length of treatment 0.002 0.003 )0.004–0.007
Age 0.598 0.460 )1.109–2.384
Frequency of exposure 0.057 )0.006 )0.126–0.259

ModelBFULL
Intercept (hormone, mono-genotypic sex and 15.52 17.82 1.503–29.09
feminization studies)
Other chemicals 12.66 9.53 )9.949–36.97
Masculinization (diff. between masculinization )5.278 )5.467 )10.69–0.698
and feminization)

 2011 Blackwell Publishing Ltd, F I S H and F I S H E R I E S , 14, 60–76 75


A comparative study of sex reversal A M Senior and S Nakagawa

Table A2 (Continued)

Fixed effects Mean I Mode I CI

Mixed genotypic sex (diff. between mixed )5.255 )3.406 )12.84–4.775


and mono-genotypic sex groups)
Length of treatment )0.056 )0.078 )0.339–0.214
Age 6.053 8.412 )15.97–24.48
Frequency of exposure )0.966 2.164 )12.15–9.165

Random effects

ModelD0
Article 0.309 0.008 3 · 10)03–1.83
Species 0.065 0.002 2 · 10)03–0.287
Phylogeny 0.076 0.004 1 · 10)03–0.367
Residual 3.445 3.357 2.063–4.471

ModelB0
Article 12.04 )0.04 2 · 10)03–53.60
Species 5.223 0.204 2 · 10)03–26.35
Phylogeny 17.22 )0.168 2 · 10)03–106.5
Residual 56.00 54.34 18.40–94.21

ModelDDIC
Article 0.178 0.006 3 · 10)03–0.698
Species 0.035 0.001 2 · 10)03–0.159
Phylogeny 0.046 0.002 3 · 10)03–0.216
Residual 3.160 3.411 1.997–4.214

ModelDFULL
Article 0.178 0.005 2 · 10)03–0.792
Species 0.040 0.002 2 · 10)03–0.170
Phylogeny 0.053 0.002 3 · 10)03–0.240
Residual 3.374 3.409 2.205–4.768

ModelBFULL
Article 37.40 0.057 3 · 10)03–98.95
Species 4.889 0.357 3 · 10)03–27.58
Phylogeny 9.854 )0.153 2 · 10)03–51.64
Residual 38.95 16.79 7.334–84.38

76  2011 Blackwell Publishing Ltd, F I S H and F I S H E R I E S , 14, 60–76


Copyright of Fish & Fisheries is the property of Wiley-Blackwell and its content may not be copied or emailed
to multiple sites or posted to a listserv without the copyright holder's express written permission. However,
users may print, download, or email articles for individual use.

You might also like