Rensch’s rule inverted – female-driven gigantism in nine-spined stickleback Pungitius pungitius


Correspondence author. E-mail:


1. Allometric scaling of sexual size dimorphism (SSD) with body size is a commonplace occurrence in intraspecific or interspecific comparisons. Typically, SSD increases with body size when males, and decreases when females are the larger sex – a pattern known as Rensch’s rule. Intraspecific studies of Rensch’s rule in vertebrates are extremely scarce.

2. In an allometric SSD–body size relationship, the sex with the larger body size variation is the driver of size divergence whereas the other sex is following it owing to correlational selection. Hence, one can test which sex is responsible for the observed body size divergence within this framework.

3. Nine-spined stickleback (Pungitius pungitius) provides an excellent model to study intraspecific variation in SSD owing to the large interpopulation variation in mean body size. Using data on body size variation in 11 nine-spined stickleback populations covering the full known size range of the species, we investigated: (i) whether variation in SSD scales allometrically with mean body size across the populations; (ii) which sex is driving the allometric relationship and (iii) whether the observed pattern is likely to have a genetic component. In addition, we analysed the size dependency of female reproductive output.

4. We found strong support for an inverse of Rensch’s rule: level of female-biased SSD increased with increasing mean size while females were the more variable sex. Results from a common garden experiment supported the pattern found in the wild. Females from giant populations had 2–3 times larger reproductive output than normal-sized females.

5. The fact that females were the more variable sex indicates that the evolution of gigantism in nine-spined sticklebacks is driven by females, and the 2–3 times larger reproductive output per clutch of giant vs. normal-sized females suggests fecundity selection to have an important role in it. Our results oppose the commonly held view that males drive the evolution of SSD as a result of sexual selection favouring larger males.


Body size is an ecologically and evolutionary important trait, which influences and often correlates with a number of physiological and fitness traits both within and among populations and species (Peters 1983; Roff 1992; Stearns 1992). Interpopulation variation in body size can result from several co-occurring evolutionary and ecological factors. Apart from random events (Wasserzug et al. 1979), prey size distribution, predation risk, resource levels and the degree of interspecific and intraspecific competition can all contribute to body size evolution (e.g. Wilson 1975; Case 1978, 1979; Gittleman 1985; Blanckenhorn 2000; Simberloff et al. 2000; Clegg & Owens 2002; Boback 2003; Wu, Li & Murray 2006).

Body size can also differ considerably between sexes of the same species (or population), a phenomenon known as sexual size dimorphism (SSD). SSD is common among both plants and animals (Fairbairn 1997), but whether females or males are the larger sex varies in taxon specific manner. In invertebrates and ectothermic vertebrates females are typically the larger sex, whereas in endothermic animals, males are generally larger than females (Fairbairn & Preziosi 1994; Fairbairn 1997; Blanckenhorn 2005; and references therein). Furthermore, the degree of SSD appears to correlate with the mean body size of the species (or population) displaying an allometric relationship (e.g. Fairbairn 1997). In species where females are the larger sex, increasing size is coupled with decreasing SSD (hypo-allometry), whereas in species where males are the larger sex, SSD increases with increasing mean size (hyper-allometry). This relationship is known as the Rensch’s rule (e.g. Rensch 1950, 1959; Fairbairn 1997). Note that hypo- and hyper-allometry represent the two ‘ends’ of one relationship where males are the more variable sex (Δ male size > Δ female size; see Fig. 1 in Fairbairn & Preziosi 1994 or Fairbairn 1997). In general, this common allometry implies that: (i) males display larger evolutionary size divergence than females, and that (ii) there is a strong covariance between male and female size. Even though allometry in male-biased SSD follows Rensch’s rule in the majority of the studied taxa (e.g. Fairbairn 1997; Colwell 2000; Kratochwíl & Frynta 2002; Székely, Freckleton & Reynolds 2004; Johansson, Crowley & Brodin 2005; Fairbairn, Blanckenhorn & Székely 2007), there are some rare exceptions showing an inverse of Rensch’s rule (cf. Fairbairn 1997) and the trend is questionable in taxa with female-biased SSD (Webb & Freckleton 2007; Stephens & Wiens 2009). However, the correlational selection hypothesis evoked by Fairbairn (1997) accommodates both trends, stating that “…allometry for SSD evolves as a consequence of directional selection acting primarily on one sex (e.g. sexual selection on males or fecundity selection on females) combined with correlational selection on the other sex”. Hence, by studying allometry in SSD, one should be able to test which of the sexes is driving body size evolution in a given study system/organism. In other words, given the correlational evolution of body size between sexes, the sex showing greater variation among species/populations should be the driver. Rensch’s rule was originally formulated at the interspecific level and most of its large scale tests are performed between species and within lineages (Abouheif & Fairbairn 1997; Székely et al. 2004; Johansson et al. 2005; Serrano-Meneses et al. 2009; Stephens & Wiens 2009) whereas intraspecific studies in vertebrates are scarce.

Figure 1.

 Map of Fennoscandia showing the location of the sampling sites. Open circles denote small isolated ponds and filled circles marine populations. The underlined populations are the ones represented in the common garden experiment. For population abbreviations, see Table 1.

Gigantism has been observed in a few cases in the three-spined stickleback (Gasterosteus aculeatus Linneaus) and explained as an evolutionary response to the presence of gape-limited predators (Moodie 1972a, b; Moodie & Reimchen 1976; Bell 1984; Reimchen 1988, 1991). Gigantism has also been found in the nine-spined stickleback, Pungitius pungitius Linneaus (Kuusela 2006; Merilä 2006). We have demonstrated elsewhere that P. pungitius living in geographically and genetically isolated small ponds, being released from the constraints of predation and interspecific competition, are larger – sometimes reaching giant sizes – than their conspecifics in lake or sea populations where they coexist with several predatory and competing fish (Herczeg, Gonda & Merilä 2009a).

In this study, we provide one of the first (also see Lengkeek et al. 2008) intraspecific tests of Rensch’s rule in fish, using data on SSD from 11 (Fig. 1 and Table 1) nine-spined stickleback populations differing markedly in their mean body size and covering the whole known size range of the species. Data from a common garden experiment using a subset of the populations were utilized to demonstrate that the observed patterns were not attributable to only phenotypic plasticity, but had a genetic component. The Rensch’s rule framework was also used to test which sex is driving gigantism in P. pungitius. By analysing data on female reproductive output in different populations, we also evaluated the plausibility of the hypothesis that fecundity selection acting on females might be an important explanatory factor for gigantism, and/or the observed patterns population divergence in SSD.

Table 1.   Sampling sites, their abbreviations, coordinates, sample sizes (males/females) and country. Coordinates of the Russian sites (LEV, BOL, MAS) are approximate (based on Ziuganov & Zotin 1995)
Sampling siteAbbrevationsSample sizeCoordinatesCountry
Marine (coastal) populations
 Helsinki, Baltic SeaHEL34/3760°13′ N; 25°11′ EFinland
 Levin Navolok, White SeaLEV26/4966°18′ N; 33°25′ ERussia
 BolotnojeBOL18/2566°18′ N; 33°25′ ERussia
 MashinnojeMAS23/4066°18′ N; 33°25′ ERussia
 PyöreälampiPYÖ36/8166°15′ N; 29°26′ EFinland
 RytilampiRYT19/6766°23′ N; 29°19′ EFinland
 KirkasvetinenlampiKIR29/4466°26′ N; 29°08′ EFinland
 AbbortjärnABB40/4064°29′ N; 19°26′ ESweden
 BynästjärnenBYN45/4664°27′ N; 19°26′ ESweden
 HansmyrtjärnHAN30/3164°33′ N; 19°10′ ESweden
 Lill-NavartjärnNAV32/3264°33′ N; 19°11′ ESweden

Materials and methods

Sampling sites and sampling

Adult fish (Nmale = 332; Nfemale = 492) were collected from 11 Fennoscandian populations (two marine and nine pond populations; Fig. 1 and Table 1) during the breeding seasons (late May–early July) of 2006–2007. Only samples with enough fish from both sexes in a given year were used (mean Nsex per population = 37·45, 95% CI = 31–44, min–max = 18–81). As P. pungitius starts reproduction after its first wintering, and most of its growth takes place before that (Jones & Hynes 1950; Bănărescu & Paepke 2001), distinguishing adults and juveniles is easy. While the Russian populations (Bolotnoje, Mashinnoje and White Sea at Levin Navolok Bay) where the isolation of the ponds happened only recently from the White Sea (Ziuganov & Zotin 1995; information from the White Sea Biological Station) form a genetically indistinguishable group, all other populations were found to be genetically (highly) isolated from each other and from the Russian populations based on analysis of variability of highly polymorphic microsatellite loci (Shikano et al. 2010).

Collected fish were killed with an overdose of MS 222 (tricaine methanesulphonate) at the site of capture and stored in 96% ethanol for c. 2 months. After this, all individuals were moved to 4% formalin. Standard length (from the tip of the mouth to the end of the tail base) was measured at this time with a digital calliper (to an accuracy of 0·01 mm). Gender was identified by eye: male P. pungitius develop conspicuous black nuptial colouration during the reproductive season, which becomes even more apparent during the application of MS 222. We note that storage in alcohol can cause some shrinkage. However, length in alcohol is known to become stable in less than 2 months (e.g. Fox 1996; Kristoffersen & Salvanes 1998), and the change found in fish species comparable with P. pungitius in size is minor (less than 3%; Kristoffersen & Salvanes 1998).

Common garden experiment

To assess whether patterns observed in the wild have a genetic component, individuals from four populations (Fig. 1) differing in mean size in the wild (Herczeg, Gonda & Merilä 2009c; Herczeg et al. 2009a) were raised in a common garden experiment. We used two geographically isolated ponds (Bynästjärnen and Pyöreälampi, separated by >500 km) and two marine populations (Baltic Sea near Helsinki and White Sea at Levin Navolok Bay, separated by several thousands of kilometres by coastline distance). Detailed descriptions of the common garden procedures are available from Herczeg, Gonda & Merilä (2009b), Herczeg et al. (2009c) and Gonda, Herczeg & Merilä (2009a, b). In short, for each population, five full sib families were created in vitro and 10 randomly chosen individuals from each family (4 populations × 5 families × 10 individuals = 200 fish) were transported individually to 1·4-L tanks of two Allentown Zebrafish Rack Systems (Aquaneering Inc., San Diego, USA). The racks were equipped with physical, chemical, biological and UV filters. Temperature was set to 17 °C throughout the experiment. Fish were fed two times a day in excess, first with brine shrimp (Artemia salina) nauplii, then with frozen copepods (Cyclops sp.) and frozen bloodworms (Chironomidae larvae). As a result of the latitudinal differences between the source populations we did not aim to mimic natural photoperiod changes precisely: we started with a 24 h light period (representative of high latitudes at summer) and changed it to a 12:12 h light–dark periodism gradually during 1 week after 12 weeks.

We aimed to quantify female reproductive output at two stages. First, during the in vitro fertilizations we took c. 15–20 eggs from each female to 4% formalin. Later five eggs were chosen (randomly from the round-shaped ones) and photographed with a digital camera through a connected dissecting microscope (Wild M5A, Heerbrugg, Switzerland). A millimetre scale was positioned in each photograph for the later measurements. We also counted the total number of eggs in each clutch. Second, the 10 randomly chosen fish used in the common garden experiment from each family were photographed 4 days after hatching when their yolk sac had been absorbed and fish started to swim freely. There was no external feeding done until this stage. Ten individuals had to be discarded either because of early mortality or deformations. Photographs were taken under standardized conditions with a millimetre scale placed in each photograph for scaling. Diameter of the eggs and total length of the fry was measured using tpsDig 1·37 (Rohlf 2002) software.

Finally, common garden fish were measured at the age of 36 weeks [the growth curves of every population and both sexes (i.e. all individuals) had approached their asymptotes at this time; Herczeg, Gonda & Merilä, unpublished data]. Standard length was measured with tpsDig 1·37 (Rohlf 2002) software on digital photographs made from a standard angle from all individuals (a millimetre scale was placed in all photographs as a reference). Body weight was measured with a digital balance to the nearest 0·01 g. Genders were identified via dissection. Individuals from Bynästjärnen could not be analysed as our samples included only two males, so only the remaining three populations were retained for analyses. Due to mortality and other scientific purposes (Gonda et al. 2009a, b, unpublished data), we could only use 66 individuals: 21 from the Baltic Sea (family representations: 6, 5, 4, 3, 3), 23 from the White Sea (family representations: 7, 6, 5, 3, 2) and 22 from Pyöreälampi (family representations: 6, 5, 5, 5, 1). We assumed that these individuals are representative for their source populations.

Statistical analyses

First, to analyse the data collected from wild-caught fish, we applied a General Linear Model (GLM) with standard length as dependent variable, and population and sex as fixed factors to test for sex differences in mean standard length among the 11 populations. To test for allometry vs. isometry, the log size of one sex was regressed against the log size of the other using the population means. As in this case neither males’ or females’ size measurements are fixed (and neither of them is dependent nor independent), and both are measured with error, the standard ordinary least squares (Model I) regression would be statistically incorrect, and major axis regression (Model II) is regarded as more appropriate (e.g. Fairbairn 1997). We run both models for comparative reasons. Following Fairbairn (1997), we placed female size on the x-axis. We note that some authors recommend using the maximal values instead of population means (e.g. Stamps & Andrews 1992; Kratochwíl & Frynta 2002). We rerun our GLM using the five largest individuals from each sex in every population and also the regression using the maximal value from each sex in every population, and the results were qualitatively the same as in the analyses based on means (data not shown).

Size of the 36-week-old common garden fish was analysed by General Linear Mixed Models (GLMMs), with body weight or standard length as dependent variables, population and sex as fixed factors and family nested within population as random factor. Size of the females used in the common garden fertilizations was compared with a GLM with population as fixed factor. To test for maternal size dependence in egg volume (calculated from the diameter), egg number and fry size we run GLMMs with mean egg volume (per mother), egg number and mean fry length (per mother) as dependent variables, standard length of the mother as continuous predictor and population as random factor.

Population × sex interactions were included in the models wherever applicable. For the pairwise comparisons following GLMMs, we applied Tukey’s post hoc tests. Model II regression was done manually following Sokal & Rohlf (1981), while the other analyses were performed with STATISTICA 8·0 (StatSoft. Inc., Tulsa, Oklahoma, USA) and SAS 9·1 (SAS Institute Inc., Cary, NC, USA) for Windows.


SSD in the wild

The GLM revealed that the degree of SSD differed among populations (population: F10, 802 = 196·72, P < 0·0001; sex: F1, 802 = 67·90, P < 0·0001; population × sex: F10, 802 = 8·00, P < 0·0001). Tukey’s post hoc tests revealed that female-biased SSD was significant in the four populations with the largest mean size (all P < 0·004; Fig. 2a) whereas in the others no statistical support for the existence of SSD was found (all P > 0·49; Fig. 2a). Model I regression revealed a significant relationship between the mean size of the sexes across populations and showed that the relationship was significantly different from that assumed under isometry (i.e. both β = 0 and β = 1 are rejected; R2 = 0·97, β = 0·738, SE(β) = 0·040, P < 0·0001; Fig. 2b). Model II regression retrieved virtually the same slope (β = 0·745, SE(β) = 0·037). Hence, the results provide strong support for an inverse of Rensch’s rule: the detected trend represents hyper-allometry with female-biased SSD.

Figure 2.

 (a) Sexual size dimorphism across populations measured in wild Pungitius pungitius (means + 95% confidence intervals). Asterisks denote populations where post hoc Tukey’s tests revealed significant sexual differences. (b) Allometry of sexual size dimorphism. Linear (Model I) regression line (β = 0·738) with 95% confidence interval (dotted line) is shown. The thick grey line represents isometry, i.e. β = 1. For the population abbreviations, see Table 1.

SSD in common garden

The GLMMs on the 36-week-old common garden fish revealed that sex differences were population dependent both in standard length (population: F2, 11·7 = 187·44, P < 0·0001; sex: F1, 63·9 = 127·16, P < 0·0001; population × sex: F2, 64 = 12·98, P < 0·0001; family [population]: Z = 1·68, P = 0·05) and body weight (population: F2, 9·95 = 259·60, P < 0·0001; sex: F1, 55·4 = 66·35, P < 0·0001; population × sex: F2, 55·9 = 19·62, P < 0·0001, family[population]: Z = 0·83, P = 0·20). Tukey’s post hoc tests revealed that SSD in standard length was significant in every population (P < 0·02; Fig. 3a). However, SSD in body weight was only significant in the population with the largest mean size (P < 0·0001; Fig. 3b), but not in the two smaller populations (all P > 0·14; Fig. 3b).

Figure 3.

 Sexual size dimorphism (a, standard length; b, body weight) across populations measured in common garden-reared Pungitius pungitius (means + 95% confidence intervals are shown). Asterisks denote populations where post hoc Tukey’s tests revealed significant sexual differences. HEL denotes Baltic Sea near Helsinki; LEV denotes White Sea at Levin Navolok Bay; and PYÖ denotes Pyöreälampi (Finnish pond).

Female reproductive output

Standard length of females used for the common garden experiment differed significantly between populations (population: F3, 16 = 72·33, P < 0·0001; mean ± SD; Baltic Sea = 43·39 ± 2·48; White Sea = 44·03 ± 1·85; Bynästjärnen = 62·97 ± 0·75; Pyöreälampi = 74·31 ± 7·29). The number of eggs in a clutch was positively related to the size of the female (F1, 4·3 = 21·54, P = 0·008; Fig. 4a). Mean egg volume was unrelated to the size of the mother (F1, 6·35 = 3·15, P = 0·12; Fig. 4b). Mean fry size was positively related to the mothers’ size (F1, 13·35 = 13·35, P = 0·04, Fig. 4c). The population effect was always non-significant (Z < 0·91, P > 0·18).

Figure 4.

 Size dependence of female reproductive output (a, number of eggs; b, egg volume; c, length of fry) in Pungitius pungitius. HEL denotes Baltic Sea near Helsinki; LEV denotes White Sea at Levin Navolok Bay; BYN denotes Bynästjärnen (Swedish pond); and PYÖ denotes Pyöreälampi (Finnish pond). Regression lines are shown in the cases of significant relationships, only for illustrative purposes. Egg volume and length of fry are mean (per female) values.


The most salient finding of this study is the demonstration of a clear pattern showing the inverse of Rensch’s rule in P. pungitius. In other words, female-biased SSD displays a hyper-allometric relationship with the mean body size such that SSD increases with increasing mean body size. Several large meta-analyses have found strong support for Rensch’s rule whereas only a very few studies have demonstrated an inverse of Rensch’s rule (Abouheif & Fairbairn 1997; Fairbairn 1997; Colwell 2000; Lindenfors, Székely & Reynolds 2003; Székely et al. 2004; Johansson et al. 2005). The majority of studies so far found that body size/SSD evolution is mainly driven by males, resulting from sexual selection favouring larger males (e.g. Berry & Shine 1980; Fairbairn & Preziosi 1994; Abouheif & Fairbairn 1997; Cox, Kelly & John-Adler 2003; Székely et al. 2004; Dale et al. 2007; Lislevand, Figuerola & Székely 2009; Serrano-Meneses et al. 2009), whereas female body size evolves as correlated response to selection on males (Fairbairn 1997).

Studies on systems with primarily female-biased SSD are rare, but the reported patterns still follow Rensch’s rule in many cases (cf. Fairbairn 1997; Stephens & Wiens 2009; Stuart-Fox 2009). Inversed pattern of Rensch’s rule have been found exclusively in taxa with female-biased SSD (cf. Fairbairn 1997), and the results of our intraspecific study are consistent with this pattern. This relationship in P. pungitius appears to emerge owing to the female-biased SSD in populations characterized by extraordinary large individuals. Because (i) larger body size variation is observed among females than among males and (ii) there is no SSD in ‘normal-sized’ populations our results indicates that selection on female body size is driving body size differences in our system. This is expected to take effect through fecundity selection, or in other words, increased reproductive output of larger females, a phenomenon proven both within (Heins, Johnson & Baker 2003; Heins et al. 2005) and among P. pungitius populations (present study).

It is noteworthy that even though Rensch’s rule is originally framed for interspecific comparisons, its predictions can be tested also at the within species level where detailed knowledge about the species’ biology can help to give cues to the mechanisms behind SSD allometry (e.g. Fairbairn & Preziosi 1994; Fairbairn 2005; Teder & Tammaru 2005; Blanckenhorn et al. 2007). However, whereas reporting phenotypic body size/SSD trends from the wild can be interesting, such studies cannot distinguish between the alternative causes (i.e. local adaptation vs. phenotypic plasticity) behind the observed patterns. This can be extremely important considering the fact that environmental conditions can markedly influence the expression of SSD (Fairbairn 2005; Teder & Tammaru 2005). In contrast, results of our common garden experiment concurred with the pattern found in the wild. Although SSD in length was significant in all populations, SSD in body weight was only significant in the giant population. Therefore, the pattern in body weight SSD was as expected under an inverse of Rensch’s rule. Even though the low number of common garden populations did not allow us to formally test for allometry in SSD, the results suggest that the presence/absence of SSD has a genetic component in P. pungitius.

A relevant and obvious question in this context arises: how are we to explain the differences in the degree of SSD among the genetically isolated P. pungitius populations? Considering that most components of selection for increased reproductive success (i.e fecundity selection, intrasexual competition, mate choice) favour larger body size (Wooton 1979; Clutton-Brock, Guiness & Albon 1982; Shine 1988, 1989; Andersson 1994), studies reporting selection for larger body size within species (e.g. Kingsolver & Pfennig 2004) or evolution of larger body size within lineage over geological time (e.g. Cope 1887; Alroy 1998) are not surprising. Still, a permanent increase in size is not something that characterizes the majority of natural populations, most probably because of counterbalancing evolutionary forces originating from ecological constraints and biotic interactions (e.g. Wilson 1975; Blanckenhorn 2000; Boucher et al. 2004). Therefore, whenever a certain population is released from these constraints, evolution towards some ‘optimal’ body size (certainly not a single value for diverse taxa: Meiri, Simberloff & Dayan 2005) is to be expected (Damuth 1993; Boback & Guyer 2003).

When P. pungitius is released from biotic environmental constraints (interspecific competition and predation) imposed by sympatric fish species (i.e. when inhabiting isolated ponds), local populations can evolve into aggressive, bold, quickly feeding giants (Herczeg et al. 2009a, c) facing a developmental deficit when kept in groups whereas marine (representing ‘normal’ populations constrained by interspecific competition and predation in nature) fish do not (Gonda et al. 2009a; Herczeg et al. 2009b). Interestingly, the size-wise ‘normal’P. pungitius populations (standard length = 40–55 mm; e.g. Bănărescu & Paepke 2001) are not sexually dimorphic; the female-biased SSD indeed appears exclusively in populations where the mean body size is outstandingly large. The largest individuals we caught were females and exceeded 110 mm in total length (Rytilampi and Pyöreälampi populations). The observed inverse of Rensch’s rule suggests that the increase in body size is driven by females as expected if driven by fecundity selection (e.g. Fairbairn 1997). Heins et al. (2003, 2005) have shown that clutch size is strongly depends on the mothers’ body size in P. pungitius. In line with this, our results show that populations differing in the mean size of mothers differ also in various measures of female reproductive output. Unfortunately we lack measures of the size dependence of male reproductive success (note that in P. pungitius males defend territories and fight with each other), but it seems feasible to suggest that the two- to threefold increase in reproductive output per clutch giant females can achieve was the main driver of body size evolution in the pond populations of P. pungitius.

In summary, our results demonstrate hyper-allometry in SSD in a species where the sexes are either equally sized or females are larger than males. This translates to a pattern following an inverse of Rensch’s rule. Results of the common garden experiment suggest that the presence/absence of female-biased SSD has a genetic component, and hence, representing evolutionary shifts most likely caused by selection. Significant SSD was only observed in the populations consisting of giant individuals, a fact that together with the inverse of Rensch’s rule indicate that the evolution of gigantism is female-driven in P. pungitius. The two- to threefold increase of reproductive output per clutch in giant as compared with ‘normal’ females suggest that one selective agent responsible for female-driven gigantism is likely to be fecundity selection.


Victor Berger, Göran Englund, Tuomas Leinonen, Daniel Lussetti and Pirkko Siikamäki provided invaluable help in obtaining the samples. Access to the facilities of the Oulanka Research Station (University of Oulu) and the White Sea Biological Station (Russian Academy of Sciences) made the collections possible. Financial support was provided by the Academy of Finland, Ministry of Education and Centre for International Mobility (CIMO). The authors are highly indebted to Lukas Kratochvil and Wolf Blanckenhorn for their comments leading to improvements of the manuscript and John Loehr for correcting the English. The experiment was conducted under the licence of the Helsinki University Animal Experimentation Committee.