Ontogenic differences in sexual size dimorphism across four plover populations

Sexual size dimorphism (SSD) among adults is commonly observed in animals and is considered to be adaptive. However, the ontogenic emergence of SSD, i.e. the timing of divergence in body size between males and females, has only recently received attention. It is widely acknowledged that the ontogeny of SSD may differ between species, but it remains unclear how variable the ontogeny of SSD is within species. Kentish Plovers Charadrius alexandrinus and Snowy Plovers C. nivosus are closely related wader species that exhibit similar, moderate (c. 4%), male‐biased adult SSD. To assess when SSD emerges we recorded tarsus length variation among 759 offspring in four populations of these species. Tarsus length of chicks was measured on the day of hatching and up to three times on recapture before fledging. In one population (Mexico, Snowy Plovers), males and females differed in size from the day of hatching, whereas growth rates differed between the sexes in two populations (Turkey and United Arab Emirates, both Kentish Plovers). In contrast, a fourth population (Cape Verde, Kentish Plovers) showed no significant SSD in juveniles. Our results suggest that adult SSD can emerge at different stages of development (prenatal, postnatal and post‐juvenile) in different populations of the same species. We discuss the proximate mechanisms that may underlie these developmental differences.

Sex differences in the body size of males and females are prevalent across the animal kingdom and their adaptive significance has been under investigation for over a century (Darwin 1871). Across taxa, the evolution of sexual size dimorphism (SSD) has been associated both with sexual selection for larger male body size (Sz ekely et al. 2000, Bertin & C ezilly 2003, Lislevand et al. 2009) and with natural selection for larger female body size linked to increased fecundity (Blanckenhorn 2000, Blondel et al. 2002, Fairbairn 2007. However, among birds, body size is not always strongly related to fecundity (Serrano-Meneses & Sz ekely 2006, Lislevand et al. 2009) and the magnitude of SSD varies across bird species. To fully understand the evolution of SSD in birds, it is necessary to determine both how and when SSD develops among individuals at the proximate level (Price 1984).
Despite numerous studies focusing on patterns of adult SSD, research has only recently focused on the ontogenic development of SSD (Badyaev et al. 2001a, Blanckenhorn et al. 2007, Dietrich-Bischoff *Corresponding author. Email: n.dos-remedios@sheffield.ac.uk et al. 2008, Hegyi et al. 2011, Klenov sek & Kry stufek 2013. Identifying exactly when morphological divergence between males and females starts to develop is of central importance in determining the underlying mechanisms (Cox & John-Alder 2007, Stillwell et al. 2014. SSD may emerge during three main stages: (i) prenatally, due to sex differences in embryonic growth rates resulting either from intrinsic genetic differences (Godfrey & Farnsworth 1952, Sellier 2000 or from differences in maternal investment in male and female embryos (M€ uller et al. 2012(M€ uller et al. , Helle et al. 2013); (ii) postnatally, due to sex differences in the rate or duration of growth of offspring (Leigh & Shea 1995, Blanckenhorn 2005, Hasumi 2010, Zhang & Liu 2013, which may be hormone-mediated (e.g. linked to differential testosterone and activity levels; Klukowski et al. 2004, Cox et al. 2009); or (iii) post fledging, due either to sex differences in continued growth, or to mortality differences between juvenile or adult males and females in relation to body size (Badyaev et al. 2001a, Kersten & Brenninkmeijer 2008. Comparative studies have revealed that variation in SSD often appears between closely related species. Even within a single species, variation in the extent of SSD can occur between different ecomorphs (Blanckenhorn et al. 2006, Dunham et al. 2013, Laiolo et al. 2013. For example, Blondel et al. (2002) found greater SSD in a mainland population of Blue Tits Cyanistes caeruleus than in an island population, where environmental stress was suggested to limit the size of the larger sex (males). Ontogenic factors are important in generating such population differences (Badyaev et al. 2001b). However, it remains unclear whether population differences in the ontogenic growth patterns of males and females might occur in species where adult SSD is consistent across populations.
Here, we investigate sex differences in body size during development in four plover populations: three populations of Kentish Plover Charadrius alexandrinus (Fig. 1) and one population of Snowy Plover Charadrius nivosus. The Snowy Plover and Kentish Plover are phenotypically similar and were long considered to be the same species (Hayman et al. 1988). Snowy Plovers are on average smaller than Kentish Plovers (tarsus length c. 25 and 29 mm, respectively; K€ upper et al. 2009) but adults in these four populations exhibit very similar, moderate, male-biased SSD (c. 1 mm, or 4%, difference in tarsus length, K€ upper et al. 2009, T. Sz ekely unpubl. data). Plover chicks are preco-cial, leaving the nest scrape a few hours after hatching, and feed for themselves throughout the postnatal growth period (Warriner et al. 1986. In waders (notably plovers and sandpipers), sexual selection appears to play a major role in the evolution of sexual size dimorphism and is associated with male display behaviour during courtship (Sz ekely et al. 2000(Sz ekely et al. , 2004. However, the proximate mechanisms responsible for the emergence of SSD in this group are still unclear. We aim to assess when SSD develops ontogentically among plovers (prenatally, postnatally or post-fledging), and whether SSD emerges during the same ontogenic period in different populations. We use tarsus length as a proxy for structural body size (Rising & Somers 1989, Senar & Pascual 1997, and compare hatchling size and growth up to fledging age (25 days) for males and females in each population.

Data collection
We collected hatchling size and growth data from four plover populations (Table 1). At Bah ıa de Ceuta, Mexico, we measured juvenile Snowy Plovers from May to July 2006-2009. Kentish Plover populations were investigated at Tuzla Lake, southern Turkey, from April to July 1996July -1999July and in 2004at Al Wathba Wetland Reserve, United Arab Emirates, between March andJuly 2005-2006; and on Maio Island, Cape Verde, from September to November 2008-2010.
We followed standard protocols (Sz ekely et al. 2008) to search for nests by observation from a mobile hide, from a car or on foot. We predicted hatch dates based on the floating stage of eggs in lukewarm water, and visited nests daily as the expected hatch dates approached. On the day of hatching, chicks were captured and marked with a single metal ring, and in most cases (70%) with an additional colour ring. The length of the right tarsus (tarsometatarsal bone) was measured to the nearest 0.1 mm. Structural measurements such as bill or wing length were not available for all populations as field workers aimed to minimize handling time for each chick and keep the stress for families at a minimum.
Measurement errors (% ME) were estimated for three of the four populations by comparison of variation within and between individuals. Withinindividual differences were based on comparison of right tarsus with left tarsus measurements using the formula (Bailey & Byrnes 1990): % ME ¼ ½s 2 within =ðs 2 among þ s 2 within Þ Ã 100 where s 2 within = MS within , and s 2 among = (MS among -MS within )/2. MS within and MS among are the withinindividual and between-individual mean sum of squares of a Model II ANOVA. Measurement errors were estimated at 0.83% in Ceuta, 1.49% in Tuzla and 0.38% in Al Wathba, indicating that approximately 99% of the total variation was due to differences in tarsus length between birds rather than measurement error.
A droplet of blood (25 lL) was taken for molecular sex-typing, by puncturing the leg vein using a hypodermic needle (Oring et al. 1988). Blood was stored in Eppendorf tubes containing 1 mL of Queen's lysis buffer (Seutin et al. 1991). In total, 103 chicks (29 Ceuta, 33 Tuzla, 18 Al Wathba, 23 Maio) with known hatch date and age were not captured on the day of hatching, but were captured, measured and blood sampled subsequently. These chicks were not included in the hatchling size dataset but were included in growth calculations. We monitored broods up to the age of 25 days, as most chicks fledge at (or shortly after) this point (Sz ekely & Cuthill 1999). We attempted to measure the tarsus length of chicks multiple times prior to fledging; chicks were recaptured during opportunistic encounters in the field and their tarsus length was recorded (Table 1). Where any pair of parents produced multiple broods within or between breeding seasons, we included chicks from only one brood to avoid pseudoreplication. The included brood was chosen to be the one with most chicks at hatching, or the first brood for which data were collected.

Molecular sex-typing
DNA was extracted from blood samples using an ammonium acetate extraction method (Nicholls et al. 2000, Richardson et al. 2001. For molecular sex-typing, two markers were amplified via  Years of study 19962005-2010  . PCR of Calex-31 amplified a Wchromosome fragment in females, whereas in males no product was amplified. Using two different marker systems overcomes the problem of mis-typing due to allelic dropout (Toouli et al. 2000) or Z polymorphism (Dawson et al. 2001, Dos Remedios et al. 2010. Alleles were scored using GENEMAPPER software version 4.1 (Applied Biosystems, MA, USA).

Statistical analysis
Hatchling size Because broods usually contained more than one chick (modal clutch size is three in Kentish and Snowy Plovers; Sz ekely et al. 1994), we assessed variation in hatchling size using linear mixed models (LMMs) implemented in R version 3.0.2 (R Core Team 2013) using the package 'lme4' (version 1.0-5; Bates et al. 2013). Brood identity was included as a random grouping structure to analyse sex differences in tarsus length of chicks on the day of hatching. We ran separate analyses for each population, rather than a single crosspopulation model, as our aim was to monitor within-population SSD development across two species that differed in body size (K€ upper et al. 2009). We used standardized tarsus length (zvalues) as the response variable in LMMs for each population separately to provide a comparable measure of SSD across populations. Initial models included chick sex (two-level factor), year (fixed factor) and hatch date (fixed covariate) together with their pairwise interactions. Because the timing and length of the breeding season differed between populations we also standardized hatch dates for each population using z-values. Chick sex was retained in all models. For all other terms we applied stepwise model simplification, removing non-significant terms one by one, until the 'minimum model' with lowest Akaike information criterion (AIC) value was reached for each population. Likelihood ratio tests were then carried out to estimate the significance of terms in the minimum models.

Chick growth
Early-phase tarsus growth in plovers (aged 0-25 days) is well described by a linear equation (Sz ekely & Cuthill 1999): where T is tarsus length (in mm) at age D (in days), and a and b are estimated parameters (size at hatching and daily growth rate respectively; Table 1). Linear regression models were fitted for each population separately, including all tarsus length measurements of known-age chicks captured at least once after the day of hatching (see Supporting Information Fig. S1 for plots of fitted lines).
To analyse the deviation of each data point (tarsus measurement) from the fitted linear regression models for chicks captured at any age (0-25 days old) in each population, standardized residuals were used as the response variable in LMMs, with chick identity nested within brood identity as a random grouping structure. Residuals were largely consistent for individuals measured multiple times independent of age (i.e. chicks remained larger or smaller than average with age, in 69-81% of cases in the four populations). In Ceuta, no chicks were recaptured in 2008, and therefore this year was excluded from the growth analyses for this population. Initial models included chick sex, year (fixed factors) and standardized hatch date (fixed covariate). We also included pairwise interactions between sex, year and hatch date and ran model simplifications as for hatchling size. Likelihood ratio tests were carried out to estimate the significance of terms in the minimum models. In two populations (Ceuta and Tuzla) several chicks took part in experiments in which eggs or chicks (shortly after hatching) were moved between nesting pairs. Therefore we ran separate model sets with either social or genetic brood ID as random factors. As the two model sets gave qualitatively the same results, we report only the results of models with social brood ID.

Cross-population analyses
Sex differences in hatchling size and chick growth were tested independently for each study population (Methods above). However, the results of independent tests may not be directly comparable if datasets differ in sample size or if there is additional structure caused by other explanatory variables. Therefore, to statistically combine measures of effect for each independent test, to determine the presence and magnitude of an overall effect, and to measure the degree of heterogeneity in the data between locations, we implemented meta-analytic methods in METAWIN version 2.0 (Rosenberg et al. 1999). The effect size estimates for sex differences in the final model in each population were calculated from likelihood ratio test Chi-squared statistics using Fisher's z-transformation. Negative values of Fisher's z-transformation represent a negative effect, positive values represent a positive effect, and a zero value represents no effect. A cumulative effect size (grand mean) weighted by sample sizes was then calculated to represent the overall magnitude of the effect across populations. Heterogeneity across sample sets was estimated (Q T values) to evaluate the likelihood that variance among effect sizes was greater than expected by sampling error. For both hatchling tarsus length and tarsus growth models, the ratio of the square root pooled variance to mean study variance (hatching ratio = 0.00; growth ratio = 0.81) indicated that cross-population analyses were appropriate without the need for further grouping.

Hatchling size
In Ceuta, male chicks had significantly larger tarsi than females on the day of hatching (b = 0.263 AE 0.099 (se); Table 2). In the other three populations, no significant sex differences in hatchling size were identified, although in all cases a non-significant trend emerged for larger male than female hatchlings (Fig. 2a). In Ceuta and Al Wathba, the tarsus length of hatchlings varied over time: in Ceuta, chicks hatching in 2008 had shorter tarsi than those hatching in other years of the study and in Al Wathba, tarsus length was shorter among those hatching later in the season (b = À0.631 AE 0.120; Fig. 2a).
Meta-analytic results supported the existence of population differences in SSD at hatching. For Ceuta, the effect size for tarsus length exceeded the grand mean (Zr = 0.164, var(Zr) = 0.004; grand mean 0.095; Fig. 3) in contrast to the relatively lower effect sizes of the other three populations. Furthermore, no significant heterogeneity among datasets emerged (Q T = 1.874, 3 df, P = 0.599).

Chick growth
In Tuzla, male chicks grew faster than females (b = 0.386 AE 0.146; Table 2). In contrast, no significant sex differences were identified in Ceuta, Al Wathba or Maio, based on independent testing. In Tuzla, chicks that hatched later in the season grew faster than those hatching earlier on (b = 0.192 AE 0.089; Table 2) and there was also a trend for faster growth among later chicks in Ceuta in two of three study years (2006 and 2009; year 9 hatch date interaction; Table 2, Fig. 2b). After correcting for sample size, meta-analyses indicated that the magnitude of the effect size in Al Wathba (Zr = 0.159, var(Zr) = 0.016) was similar to that of Tuzla (Zr = 0.159, var(Zr) = 0.004; Fig. 3), suggesting a trend towards faster male than female growth in both populations, in contrast to Ceuta and Maio, where effect sizes were below the grand mean (0.096). No significant heterogeneity among datasets emerged (Q T = 3.027, 3 df, P = 0.388), suggesting that the variance among effect sizes was within the range expected due to sampling error alone.

DISCUSSION
Our results suggest that despite very similar SSD in adults (4%, K€ upper et al. 2009, T. Sz ekely unpubl. data), the ontogeny of SSD differs among populations in two closely related plover species (Fig. 2). In Ceuta, male Snowy Plover chicks were larger than females on the day of hatching, but subsequent growth rates until fledging (age 25 days) did not differ between the sexes. This suggests that SSD in the Ceuta population can largely be attributed to prenatal development. Conversely, in Tuzla, male and female Kentish Plover hatchlings did not differ significantly in size but male chicks grew faster than female chicks up to fledging, suggesting that SSD in this population developed largely postnatally. A similar effect was identified at Al Wathba (Kentish Plover) in cross-population meta-analyses (though non-significant based on LMMs). Lastly, in the Maio Kentish Plover population, no significant sex differences in body size were identified among chicks at any stage, therefore the SSD previously observed among the adults of this population is more likely to have emerged after fledging due to either sex differences in growth rate, growth duration, or differential survival relative to size.
Although minor differences in sampling were present among populations, these sampling biases cannot explain the observed results. Despite sample sizes being lower for two of the four populations (Al Wathba, Maio; Table 1), meta-analytic comparison of effect sizes enabled the identification of population differences in sex-bias, and no  Table 2). Separate plots are provided to visualize the effects of significant environmental variables: 'year', 'hatch date' and their interactions. Note that hatch date was a continuous variable in the analyses; however, it has been converted to a binary variable (early, late) to visualize interactions here. Males are represented by filled symbols, females by open symbols. Circles represent means for the entire season, squares for the early season and triangles for the late season.
significant heterogeneity emerged between datasets. Additionally, a larger proportion of recapture measurements was available for younger than older chicks (pre-rather than post-12.5 days); however, the relative distribution of measurements by age was similar across populations (mean age at capture: Ceuta 5 days, Tuzla 7 days, Al Wathba 6 days, Maio 4 days).
In contrast to previous studies (Badyaev et al. 2001b, Blondel et al. 2002, Blanckenhorn et al. 2006, Laiolo et al. 2013, very similar levels of SSD were identified across populations of breeding adult plovers, but our results suggest this SSD emerges at different ontogenic stages in different populations. We suggest that multiple proximate mechanisms may therefore be involved, enabling divergence in body size at different stages even within a single species. Badyaev (2002) suggested that at any given age, the level of SSD would depend on the relative influence of age-specific genetic effects, environmental effects, maternal effects and age-specific epigenetic interactions. Whether SSD develops may depend on sex-specific responses to the external environment as well as responses to internal selection pressures, for example differences in the functional coordination of developmental processes (Gebhardt-Henrich & Richner 1998, Badyaev 2002, Klenov sek & Kry stufek 2013. The three Kentish Plover populations studied are genetically similar and distinct from the Snowy Plovers (K€ upper et al. 2009, 2012). Patterns of SSD ontogeny were consistent with genetic differences, as meta-analyses suggested the two most closely related populations, Tuzla and Al Wathba, exhibited similar postnatal SSD development between 0 and 25 days of age. Environmental differences between populations may also affect sexspecific development. Previous studies have reported context-dependent SSD in relation to environmental variation. For example, variation in climate, food resources or parasite abundance can influence patterns of growth differently for males and females, leading to differences in the extent of adult SSD in closely related populations (Richner 1989, Cooch et al. 1996, Sheldon et al. 1998, Badyaev et al. 2001a, Blondel et al. 2002, Blanckenhorn et al. 2006, Hegyi et al. 2011, Stillwell et al. 2014. Whether sex differences emerge prenatally, postnatally or after fledging may depend on whether sex differences in growth are   Vincze et al. 2013). However, these differences are not consistent with the observed population differences in development of SSD among plovers, as Snowy Plovers from Ceuta and Kentish Plover chicks from Al Wathba hatched at very similar temperatures but showed different ontogenies for SSD. Furthermore, although overall tarsus length at hatching or tarsus growth varied significantly with hatch date and/or year in three of the four populations, no significant interactions emerged between seasonal variables and the sex of offspring, suggesting that within-population environmental variation influenced both sexes equally.
One environmental factor that may be correlated with the observed population differences is locality: Maio (where no sex differences were identified among offspring) is an island location (Cape Verde, 625 km from mainland Africa), whereas the other three populations are in mainland regions. Offspring development in island populations is often slower than on the mainland (Andrews 1976, Higuchi & Momose 1981. One possibility is that males in Maio continue to grow beyond 25 days, causing SSD to develop later on. Our monitoring did not continue beyond 25 days due to the difficulties of recapturing fledged chicks. Although the tarsus length of fledglings was approaching that of adults across study populations (K€ upper et al. 2009; T. Sz ekely unpubl. data), to understand more fully how SSD develops it would be necessary to recapture and monitor growth of juvenile plovers beyond 25 days of age.
Alternatively, adult SSD may emerge through sex-biased differences in migration patterns, dispersal or behavioural exclusion of larger or smaller individuals (Stamps 1993, Watkins 1996, Haenel & John-Alder 2002, Cox & John-Alder 2007. The reported levels of SSD were based upon breeding adults captured either on the nest or with chicks (K€ upper et al. 2009, T. Sz ekely unpubl. data). This may not provide a true representation of the body size distribution for the entire population. For example, it may be that smaller males do not breed in some populations such as Maio, and SSD may be reduced or absent in the population as a whole.
The patterns of sex-biased offspring development identified within this study may have implications for population-level evolutionary processes. Sex-biases in body size can influence the relative mortality rates of males and females, and among bird species with only moderate SSD such as in plovers, the larger sex usually has the survival advantage (Bortolotti 1986, Oddie 2000, Hipkiss et al. 2002, R aberg et al. 2005, Rowland et al. 2007). Sex-biased mortality can lead to biased adult sex ratios (ratio of males to females). There is increasing theoretical and empirical evidence for a role of adult sex ratio in the evolution of parental care and mating systems (Sz ekely et al. 1999, Kokko & Jennions 2008, Jennions & Kokko 2010, Kosztol anyi et al. 2011, Liker et al. 2013. This is of particular relevance among Kentish and Snowy Plovers, as extremely high variation has been reported in parental care strategies at the population level , Arg€ uelles-Tico 2011. In summary, we present evidence that moderate male-biased adult SSD may emerge at different stages of development (prenatal or postnatal) across closely related species and also across genetically similar populations of the same species. Further comparative cross-population studies are needed to address how the ontogeny of SSD varies in relation to particular environmental conditions, and to determine the proximate mechanisms involved in driving variation in SSD ontogeny among populations.
We thank Istvan Szentirmai, Adam Lendvai, Thijs van Overveld, Medardo Cruz-Lopez, Lydia Lozano-Angulo, Iwan Fletcher and Araceli Arg€ uelles-Tico for helping with fieldwork. We also thank the staff at the wetland reserves at Al Wathba, Tuzla Lake, the Sea Turtle Program (PROTOMAR), Universidad Aut onoma de Sinaloa and Salina Porto Ingles, Maio, for logistical support. We are grateful to Dan Ruthrauff, Jeroen Reneerkens and an anonymous reviewer for providing constructive comments that helped to improve the manuscript. This study was funded by an NERC studentship (No. NE/ H525011/11) to T.S., and formed part of the PhD dissertation of N.d.R. Molecular sexing was conducted at NERC-Biomolecular Analysis Facility at the University of Sheffield. We thank Deborah Dawson for supplying aliquots of the unpublished Z-002B sex-typing marker. C.K. was supported by a Marie Curie IEF postdoctoral fellowship. A.K. was supported by the Hungarian Scientific Research Fund (OTKA K81953) and the Hungarian Academy of Sciences ('Lend€ ulet' LP2012-37/2012). Permits for fieldwork were provided by the national agencies in each country.