The influence of social demography on sex change schedules in protogynous reef fishes is well established, yet effects across spatial scales (in particular, the magnitude of natural variation relative to size-selective fishing effects) are poorly understood. Here, I examine variation in timing of sex change for exploited parrotfishes across a range of environmental, anthropogenic and geographical factors. Results were highly dependent on spatial scale. Fishing pressure was the most influential factor determining length at sex change at the within-island scale where a wide range of anthropogenic pressure existed. Sex transition occurred at smaller sizes where fishing pressure was high. Among islands, however, differences were overwhelmingly predicted by reefal-scale structural features, a pattern evident for all species examined. For the most abundant species, Chlorurus spilurus, length at sex change increased at higher overall densities and greater female-to-male sex ratios at all islands except where targeted by fishermen; here the trend was reversed. This implies differing selective pressures on adult individuals can significantly alter sex change dynamics, highlighting the importance of social structure, demography and the selective forces structuring populations. Considerable life-history responses to exploitation were observed, but results suggest potential fishing effects on demography may be obscured by natural variation at biogeographic scales.
The increasing rate of fisheries exploitation, both commercial and artisanal, on coral reef populations is a matter of concern on a global basis . Largely restricted to shallow water environments, coral reef fishes are considered to be especially vulnerable to expanded fishing activities. In addition, many targeted species are protogynous hermaphrodites (female-to-male sex transition), which is widely considered a defining feature of the vulnerability of reef fishes to size-selective exploitation . However, the demographic consequences of exploitation on protogynous populations and the extent to which anthropogenic influences on sex change schedules may be obscured by natural variation have yet to be clarified .
The mechanisms involved in protogynous sex change dynamics have been studied extensively for decades [4,5]. Much of this work demonstrated the importance of social systems in determining when and why to change sex and this is largely dependent on or facilitated by local demography, mortality schedules of populations or even habitat effects on various spatial scales [6–8]. Ultimately, what dictates the timing of sex change is the goal of maximizing the reproductive value of individuals, and thereby the reproductive output of populations .
Protogynous hermaphroditism often results in the restriction of males to the larger size classes of a population. Hence, the impact of size-selective fishing on protogynous species is considered especially detrimental as selective removal of large individuals may deplete males to the extent that reproductive outputs of fished populations are compromised [10,11]. This argument generally assumes the length and age of sexual transition is fixed with the corollary that gonochoristic species have a greater capacity to respond to fishing pressure than protogynous species [12,13]. However, both experimental and observational studies have demonstrated a high adaptive capacity, especially in the labrids, to alter the timing of sex change in response to selective removal of individuals [7,14,15]. Although, this response may differ across species  and its use as a compensatory mechanism when faced with heavy fishing pressure will be dependent on the intensity and extent of extraction on the population.
Another factor influencing the timing of sex change is the surrounding environment. Environmental effects on sex change and sexual demography in reef fishes have received little attention [3,16], despite substantial evidence of considerable intraspecific natural variation [6,17–19]. Many environmental factors affect life-history traits either directly or indirectly by processes occurring across various spatial scales, such as differences in temperature (e.g. latitudinal), quantity and distribution of habitats or resources and density-dependent processes. However, much of this work has focused on smaller spatial scales (among habitats), whereas potential regional or larger patterns in widely distributed species are necessary information for fisheries management .
In this study, I address three major questions: (i) what is the hierarchical nature of drivers structuring length at sex change at different spatial scales? (ii) how does selective pressure (e.g. fishery selection) influence sex change dynamics? and (iii) are general patterns from the above questions consistent across related species or are responses species-specific? This was done using high-resolution length estimates obtained from stereo–video surveys of parrotfishes at two spatial scales across oceanic islands. Islands ranged considerably both in reef habitat configuration and anthropogenic pressure. Responses to factors at differing scales will importantly influence the ability to predict or assess future changes, especially in data-poor tropical regions.
2. Material and methods
(a) Study species and location
This study was conducted across seven islands in the biogeographic region of Micronesia. These included two high fringing island systems (Guam and Kosrae) with limited lagoon or backreef habitat, two high island barrier reef systems (Pohnpei and Yap) with extensive lagoon, reef flat, mangrove and backreef habitat and three atoll island systems (Sorol, Ifalik and Lamotrek; electronic supplementary material, figure S1). The bullethead parrotfish Chlorurus spilurus is the most numerically common parrotfish species throughout Micronesia and many regions of the Indo-pacific. It has two colour phases: females and primary males exhibit a brown initial-phase (IP) coloration that is highly distinct from that of large terminal-phase (TP) males which are bright blue-green to yellow. Most TP individuals are secondary males (have previously functioned as mature females) and all have undergone physical metamorphosis from IP coloration . The confounding factor of IP primary males represents only a small proportion of IP individuals across Micronesian populations and does not affect estimates of sex change schedules (see the electronic supplementary material, figure S2). Bullethead parrotfish are heavily targeted by fishermen on Guam where they comprise approximately 15% of the parrotfish landings in the recent decade (Guam Department of Agriculture, unpublished data). However, they are virtually unfished or lightly harvested at the other islands [22,23].
(b) Population surveys
Bullethead parrotfish populations were surveyed at 50 sites on the outer reef slopes across the seven islands (see the electronic supplementary material, figure S1). At each site, individuals were surveyed using diver-operated stereo–video  along 16 replicate transects (5 m wide by 3 min long, averaging 315 m2) stratified at two depths (6–10 m and 18–22 m). Individuals were quantified, measured (millimetre fork length) and categorized by colour phase using the EventMeasure software (www.seagis.com.au). These surveys also provided estimates of operational density and sex ratio (density and sex ratio of individuals above mean length at female maturity; hereinafter ‘density’ and ‘sex ratio’). Length at 50% sex change (X50) was estimated at each site by pooling the individual observations across transects and depths and fitting a logistic curve to the proportion of TP individuals in 10 mm length classes (average precision for length measurements was less than 10 mm). DeMartini et al.  discussed the use of using colour phases from visual surveys with centimetre-scale accuracy to estimate X50. Stereo–video technology improves the resolution of length estimates by an order of magnitude compared with standard visual surveys (precision in millimetres versus centimetres).
For Guam, a site-specific fishing pressure index for parrotfishes was derived from historical creel survey data (survey protocols following ) collected since January 2001, when marine reserves were enforced, because the spatial distribution of fishing effort changed with the enforcement of five marine reserves. The index summed up the total number of fishing trips targeting parrotfishes by sector, combining the shore-based  with the boat-based  surveys, with shore-based values multiplied by a scaling factor (0.68) to account for differences in relative catch efficiency between the two fisheries. Site-specific values were divided by respective reef areas (km2) and natural log-transformed. Additional explanatory variables for each site included wave exposure (following ), reef slope (0–90°; mean of three measurements per transect), coral cover and rugosity (1–5 scale; mean of five estimates per transect), distance to reef pass (kilometres), predator biomass, latitude and longitude. Predator biomass represented the mean biomass density of all species known or highly suspected to prey upon adults of smaller bodied parrotfish species based on an exhaustive literature search regarding diets and length–weight ratios of all piscivores recorded during surveys.
Two complementary statistical techniques  were employed to model the response of length at sex change across sites. First, univariate regression trees  were used to analyse the hierarchical structuring of variation in the response variable (X50; square root-transformed) relative to the explanatory variables. This was done at the within-Guam level and the Micronesia level (among sites across islands). Fishing pressure was excluded from explanatory variables in the Micronesia-scale analysis because Guam is the only island where C. spilurus is heavily targeted, whereas island type (fringing, barrier or atoll reef system) was included as a factor. Cross-validation of relative error (CVRE) was used to prune the trees.
I also fitted models to determine the combination of factors that best predicts X50 patterns for C. spilurus at the two scales. Across sites at Guam, where human extraction from fishing may have a strong effect on demographic processes, I fitted generalized linear models (GLM) with Gaussian errors and an identity link . Explanatory variables included the fishing pressure index and all environmental variables described above. At the Micronesia scale, the nested structure called for linear mixed effects model (LME) analysis, where island (random factor) was nested within-island type (fringing, atoll or lagoon reef system; fixed factor) and all environmental variables were included as fixed factors. Fishing pressure was again dropped from the explanatory variables at this scale, but special consideration regarding its effects is included in the interpretation of the results. Prior to fitting models, explanatory variables were tested for collinearity by using variance inflation factors  and variables with values over 3 were removed from the analysis. Coral cover and rugosity (highly collinear factors) were combined by using values of the first principal component (explaining 94 and 89% of variance among them at the Guam and Micronesia scales, respectively) from a principal components analysis on the covariance matrix. The mixed model was fitted using the lme function of the nlme package in the R statistical computing language . Model selection for the LME was done via multi-model averaging  based on minimization of corrected Akaike's Information Criterion (AICc) using the dredge function in the MuMIn package in R. The 10 LME models with the lowest AICc values were refitted with restricted maximum-likelihood estimation and model validation was carried out by plotting standardized residuals against fitted values to identify violation of homogeneity. Residuals were also plotted against explanatory variables to check for potential trends.
Finally, to determine whether potential trends across island types were specific to C. spilurus or were common among related species, I pooled observations of three other sexually dimorphic parrotfish species for which there was great enough resolution (Scarus forsteni, Scarus schlegeli, and Scarus rubroviolaceus) from video surveys at the island type scale and compared trends in X50 using boxplots based on 50 bootstrapped re-estimates from the sampled distributions. I also present X50 estimates for fringing (Guam) and barrier reef systems (Yap and Pohnpei pooled) from biological samples based on histological examination of gonads for a fourth species, Chlorurus microrhinos, which does not show strong phenotypic evidence of sexual transition.
On Guam, X50 ranged considerably across sites from 185 to 244 mm fork length. The regression tree model indicates fishing pressure was the most important variable explaining this variation (r2 = 0.42; figure 1a). Sites with higher fishing pressure had significantly smaller X50 values, suggesting that selective removal of TP males prompts sex change in smaller individuals. Results from the GLM differed slightly in that density was the best predictor of X50 across sites (p = 0.0017), whereas fishing pressure was next (p = 0.0229), followed by the coral-rugosity metric (p = 0.0529; table 1). However, density and fishing pressure were considerably correlated (r2 = 0.32) on Guam despite variance inflation factors indicating low collinearity.
Among islands, X50 values ranged from 154 to 244 mm fork length (figure 2a). Three sites at Lamotrek contained C. spilurus abundances too low to estimate X50. At this scale, island configuration appeared to be driving most of the variation in the response variable (figure 2b), and a nested analysis of variance demonstrated significant differences in X50 among island types (F2,40 = 36.27, p < 0.0001) but not among islands nested within-island types (F4,40 = 1.96, p = 0.119). Regression tree analysis confirms the hierarchical structuring of variation based on island type (figure 1b), in which the optimal model based on CVRE was the split between fringing reef systems and atoll and barrier reef systems. I display additional splits, which further separated the data by island type (atolls versus barrier reef systems) and by density within fringing reef systems.
Of the 10 optimal LME models, four were eliminated based on outliers or low homogeneity in residual plots. The remaining six models shared a common feature in that density, island type and the interaction between these were always included (table 2). Island type was significant in every case, indicating that the response of X50 is dependent on island configuration (see the electronic supplementary material, table S1). While density was never significant, the interaction between density and island type always was, suggesting that the relationship between X50 and density varies by island type, thus necessitating further examination (see the electronic supplementary material, table S1). Other factors emerging in models included coral-rugosity, reef slope and wave exposure (table 2). These factors were important at the within-island scale and varied in importance among islands.
There was an overall positive relationship between density and sex ratio across sites and this strengthened within-island types (figure 3a), where fringing reef systems (particularly Guam sites) stood apart from atoll and barrier reef systems. Plots of X50 by density reiterate that the response of X50 is dependent on island type, but also highlights the interaction between density and island type (figure 3b). The relationship (expressed by the slope) between density and X50 is positive for atoll and barrier reef systems as well as for Kosrae (fringing reef system, although variation among the six Kosrae sites is too high for definitive conclusion), whereas on Guam, where C. spilurus is heavily targeted, this relationship was reversed. The same patterns were observed for relationships between sex ratio and X50, where Guam not only stands apart in terms of a negative correlation, but also has a much higher range of sex ratios indicating greater proportions of IP individuals.
Data for other species pooled at the island-type scale demonstrates that the major pattern observed for C. spilurus (X50 at fringing reef systems X50 at atoll reef systems > X50 at barrier reef systems) is consistent across related species (figure 4). One minor exception was S. forsteni, for which X50 was greater at barrier than at atoll reef systems. Overall, the response of X50 appears to be dependent on island configuration in a very similar manner across all parrotfishes evaluated.
A substantial number of studies have demonstrated latitudinal and temperature effects on patterns of life-history variation [35–37]. On smaller spatial scales, sexual demography and the timing of sex change have been shown to vary in coral reef fishes across different reef configurations and reef sizes [4,6,18]. Here, results demonstrate a clear effect of reef configuration among oceanic islands on the timing of sex change in parrotfishes over broad spatial scales (1000s of kilometres) that overrides the effects of latitude, fishing pressure and local-scale environmental factors. This highlights the inherent intraspecific plasticity in life-history traits and challenges the overlooked assumption that these traits are equivalent across oceanic islands within a biogeographic region. There are strong implications for assessments of fishing effects, given that potential impacts of fishing on demographic processes are obscured by the strong effect of island type, leaving limited potential for inference across islands. Thus, comparisons of traits among populations with respect to potential fishing effects are precluded unless these differences in reef structure and configuration are adequately controlled by survey design  or statistical analysis.
The magnitude of the perceived island-effect on the timing of sex change was unexpected. Given the distribution of islands surveyed, the observed patterns reflect locality-specific adaptive responses to local environments and differential selective pressures unique to each island type. However, the proximal mechanism driving the observed differences will be difficult to isolate. The island types considered here can be qualitatively defined by their diversity and relative proportions of various habitats present. The presence, scale and proximity of particular habitats have a strong influence on the diversity and carrying capacity of reef fish assemblages [39,40], and analysis of fish communities on oceanic islands throughout the Pacific reveals predictable structuring of trophic groups based on the island types described here . In particular, Pinca et al.  found that, after partialling out the effects of fishing, parrotfish density and biomass was positively related to barrier reef systems with lagoons and negatively related to atolls and low complexity fringing reef systems, whereas fringing reefs supported greater mean body sizes. Specific habitats prevalent in wave-protected environments like lagoons provide major nursery habitat for many parrotfishes, including C. spilurus [41,42], and therefore may boost abundance potential for adults. Interspecific density-dependent processes, such as competition and predation, have considerable effects on population demography and probably influence the social mating systems distinctively at each island type . Furthermore, the high correlation between length at sex change and other length-based life-history metrics in Micronesian parrotfishes  suggests that similar patterns would be observed for other variables including length at maturity and mean maximum length.
While potential demographic responses to heavy fishing were subordinate to island-effects, responses to fishing pressure were clearly evident at the within-island scale. On Guam, where C. spilurus has been increasingly targeted over the past decade, there was a significant decline in length at sex change at sites of heavy fishing pressure, indicating a compensatory response to the disproportionate targeting of larger TP males . The level of the response suggests a high adaptive capacity to fishery extraction at this scale, although the differences remain less pronounced than those among island types. The timing of sex change is heavily influenced by the relative size of individuals, the sex ratio of social groups and local density , and the plots in figure 3 illustrate several interesting effects of fishery extraction on these relationships. The fringing reef systems Guam and Kosrae stand apart with the highest values of X50, but Guam differs from all other islands with respect to range of sex ratios and the relationships between X50, density and sex ratio, which implies that differing selective pressures (i.e. fishery selection versus natural selection) can significantly alter the dynamics of sex change and possibly those of other life-history traits. Hence, the significant interaction identified in LME models was driven by differences in selection pressures among island types, where Guam dominated the response in fringing reef systems.
The high abundance and almost ubiquitous distribution of C. spilurus across sites facilitated high-resolution analysis of sex change parameters at two very different spatial scales. Obviously, this is difficult to achieve for less abundant species without greatly increasing the sampling effort within each site (surveys already covered over 5000 m2 of reef area on average). However, pooled data for other parrotfish species strongly suggest that the patterns identified among island types are consistent across related species, especially in that length at sex change is greatest at fringing reef systems where adult densities in part may be relatively low because of reduced replenishment resulting from less local nursery habitat. These species have a range of life histories  and, across all the islands, are subjected to variable exploitation levels from light (C. spilurus and S. schlegeli) to intense (S. rubroviolaceus and C. microrhinos) fishing pressures. As sexually dimorphic protogynes, there is a consistent fishery targeting of large males. For the most heavily targeted species C. microrhinos, the length distribution from the heavily fished Guam population was severely truncated compared with Yap and Pohnpei, whereas a greater length at sex change was still observed in Guam (see the electronic supplementary material, figure S3). This again highlights that observed island-effects override potential fishing effects but further analysis of the plasticity among species is warranted. Peterson & Warner  stressed that our knowledge of flexibility in sex change stems mainly from higher density small reef fishes, whereas the capacity for adaptation in life-history traits is of prime importance in larger fishery species. Experimental work on small reef fishes [7,14] and within-island demographic data from this study demonstrate a high capacity for altering the timing of sex change following selective male removal, but examples from aggregating groupers [10,44] suggest that the capacity for sex-ratio compensation after heavy fishing may be low. A plausible explanation is that potential flexibility in traits is dependent on the life history of the species considered. Longer lived, later maturing species, as many groupers, have a lower turnover rate and will therefore respond to changes in population structure at a slower pace. However, responses will always be dependent on the scale, intensity and selectivity of the fishery extraction on a given population, which may hinder generalizations from the limited existing examples.
The LME models identified several other factors influencing variation in X50 including coral-rugosity, reef slope and wave exposure. The effects of these factors were apparent at the within-island scale and were not consistent across islands or island types, but rather varied among them. For example, exposure and slope had a strong correlative relationship with X50 on Pohnpei, whereas that with the coral-rugosity metric was weak. In Yap, these relationships were reversed. In Guam, potential effects of environmental variables were obscured by fishing effects. Interestingly, predator biomass did not emerge as an important predictor as seen in other studies [19,38], although the estimated range of predator biomass values in those studies were one to two orders of magnitude greater than that observed here, suggesting that top-down influence on demography of prey may be weak across many sites in this study. Sample replication was not high enough on other islands to make similar conclusions but the results at this scale demonstrate that environmental factors can have variable effects on population demography among islands and this probably reflects inherent differences in habitat structure (specifically, the value range of each factor) at each island.
Understanding the variation in demographic traits across multiple spatial scales is recognized as a management priority . This study demonstrates that while selective fishing can have a profound effect on traits, inherent differences based on reefal-scale environmental properties may obscure these effects at broader spatial scales. However, this represents only a piece of a much larger puzzle which extends beyond oligotrophic oceanic island systems to include continental reef systems as well as intra-island habitat effects and an underlying effect of historical biogeography that influences the distribution, ecology and demography of coral reef fishes. Studies of both environmental and fishery-induced effects on populations understandably avoid large-spatial-scale surveys because of logistical concerns and low statistical power . But multi-scale processes in coral reef ecology remain an important gap in our knowledge base that has been acknowledged for over a decade [3,9,47].
Research was carried out under James Cook University ethics permit number A1674 and University of Guam Institutional Animal Care and Use Committee permit number UOG1202.
The study was funded in part by NOAA Pacific Islands Fisheries Science Center and the author was supported by an International Postgraduate Research Scholarship from James Cook University.
I thank N. Pioppi, S. Lindfield, A. Halford, P. Houk, K. Rhodes, A. Marshell, J. McIlwain, A. Simeon, the University of Guam Marine Laboratory, Yap Marine Resources Management Division, Conservation Society of Pohnpei, Kosrae Conservation and Safety Organization, Kosrae Village Resort, and the S. S. Thorfinn for field assistance or logistical support. Discussions with J. H. Choat, P. Munday, R. Jones and S. Brandl and comments from two anonymous reviewers helped to improve the manuscript.
- Received September 16, 2013.
- Accepted November 11, 2013.
- © 2013 The Author(s) Published by the Royal Society. All rights reserved.