Introduction

The secondary sex ratios (SSRs) of newborn mammals have attracted enormous research effort since Darwin (1871). There are several theoretical backgrounds for maternal manipulation that would result in a biased sex ratio of the progeny. These include the Fisherian theory, which assumes the allocation of equal investment to male and female progeny (Fisher 1930), the model of local resource competition, which expects SSRs to be biased in favour of the dispersing sex (Clark 1978) and the maternal condition hypothesis, which predicts a higher proportion of males under favourable conditions and, conversely, a higher proportion of females under unfavourable conditions (Trivers and Willard 1973; Carranza 2002; Cameron and Linklater 2002; for a review, see Cameron 2004). Current theoretical studies combine the above models as well as other selective forces into complex mathematical models for predicting the sex ratios (e.g. Wade et al. 2003; Wild and West 2007).

Empirical results support the view that SSR is usually close to equality; nevertheless, significant deviations from the 1:1 ratio have been repeatedly reported (Austad and Sunquist 1986; Gosling 1986; Labov et al. 1986; Huck et al. 1990; Perret and Colas 1997; Creel et al. 1998; Côté and Festa-Bianchet 2001; Cameron 2004; Sheldon and West 2004; Kaňková et al. 2006). These deviations have been attributed either to the above-mentioned ultimate causes or to proximate mechanisms that are by definition not mutually exclusive. Of the proximate mechanisms, the most promising are (1) the level of circulating steroid hormones (James 1996, 1998, 2004, 2006; Grant 2007), (2) developmental asynchrony of sexes in blastocyst growth (Krackow 1995, 1997; Krackow and Burgoyne 1998; Krackow et al. 2003) and (3) circulating glucose level (Gutiérrez-Adán et al. 2001; Cameron 2004; Kimura et al. 2005; Cameron et al. 2008; Helle et al. 2008).

In recent years, the evidence for the adaptive maternal SSR adjustment model has been questioned both on empirical (see below) and theoretical grounds (Krackow 2002). Festa-Bianchet (1996) accentuated the high frequency of contradictory results and also the selective publication success of papers presenting positive results. In this respect, he shared the scepticism of an earlier paper of Clutton-Brock and Iason (1986). A similar conclusion is also supported by a thorough review by Cockburn et al. (2002). Interestingly, an empirical study analysing extensive datasets from wild savannah baboons (Papio cynocephalus) has clearly demonstrated that significant results are a product of stochastic biases that arise in small samples (Silk et al. 2005). The only effects on mammalian SSR clearly supported by recent reviews are those of maternal nutritional status around the time of conception (Cameron 2004; Rosenfeld and Roberts 2004; Sheldon and West 2004).

Among mammals, rodents are a suitable model for studying sex ratios and influencing factors, since they may be easily kept under laboratory conditions, thereby providing an opportunity to gather sufficient sample sizes required for relevant statistical analysis (for review, see Sikes 2007). Classical laboratory rodents, such as mice (Mus musculus; Krackow and Hoeck 1989; Krackow and Gruber 1990; Kaňková et al. 2007), rats (Rattus norvegicus; Bird and Contreras 1986), golden hamsters (Mesocricetus auratus; Labov et al. 1986, Huck et al. 1990) and guinea pigs (Cavia porcellus; Peaker and Taylor 1996) are traditional models for empirical studies. Nevertheless, SSRs have been studied in many other species as well, such as coypus (Myocastor coypus; Gosling 1986), Mongolian gerbils (Meriones unguiculatus; Clark et al. 1991), wood rats (Neotoma floridana; McClure 1981), wood mice (Apodemus sylvaticus; Frynta and Žižková 1994) and voles (Microtus agrestis, M. oeconomus; Hansson 1987; Ims 1994). Unfortunately, most of the published studies rely solely on correlations between sex ratio and parameters of maternal life history. Experimental studies have mostly focussed on the manipulation of food intake and food quality (e.g. Huck et al. 1986; Wright et al. 1988; Koskela et al. 2004; Rosenfeld et al. 2003; Cameron et al. 2008; Fountain et al. 2008). The effects of true social factors, such as dominance (golden hamsters: Pratt and Lisk 1989), population density (voles: Microtus townsendii, Lambin 1994; M. oeconomus, Aars et al. 1995), group size (house mice: Wright et al. 1988) and/or group composition (marmots (Marmota flaviventris): Armitage 1987, Mongolian gerbils: Scheibler et al. 2005), on sex ratios have been addressed less frequently (but see Cameron 2004 for review in other mammals). Although social factors have complex consequences that are usually difficult to interpret on a proximate level, they cannot be ignored. The considerable variation in densities and social circumstances that affects breeding females of most rodent species even under natural conditions has to be taken into account.

In the study reported here, we focussed on the sex ratio in spiny mice of the genus Acomys and its relationships to social and life history variables. Spiny mice are social animals, and under laboratory conditions they should be kept in families consisting of an adult male, multiple females and their descendants (Young 1976), which mimics their wild social system. Their societies are not anonymous, and individual recognition has been demonstrated unequivocally (Porter et al. 1986). In groups consisting of related individuals, communal care for the young comprising allosuckling is frequent; however, mothers are able to recognize their own offspring (Porter et al. 1980). In addition to the ease of keeping and breeding spiny mice under laboratory conditions, their newborns, unlike those of many other rodents, can be easily and reliably sexed according to external characters. Moreover, spiny mice are likely to be predisposed to maternal manipulation of the sex ratio for the following three reasons.

  1. (1)

    In contrast to many other muroid rodents, after an extended gestation period (36–42 days, which is nearly twice as long as that for laboratory mice), spiny mice produce only small litters (most frequently consisting of two or three newborns, range1–7; Frynta et al. unpublished results) consisting of relatively large and well-developed newborns (Brunjes 1990). Maternal effort is therefore high in the prenatal period compared with that in many other muroid rodents. Thus, we consider that maternal manipulation of the sex ratio prior to parturition is more advantageous than sex-selective parental infanticide of newborns.

  2. (2)

    The large body size of newborns relative to maternal body size (the mean litter weight at birth represents 20–25% of maternal weight; Dieterlen 1961) facilitates the potential effect of litter size on offspring quality. Resources available per individual offspring are considerably reduced even by the increment of a single newborn. Small litters thus may be predisposed for the production of the sex exhibiting a more closer relationship between consumed maternal resources and fitness prospects (Trivers and Willard 1973). Nevertheless, the assumptions of the Trivers–Willard hypothesis (Hewison and Gaillard 1999; Carranza 2002; Cameron and Linklater 2002; Blanchard et al. 2005) predicting an association between parameters related to maternal condition and allocation of maternal investment to the sexes have not been empirically tested in spiny mice yet. Two of these three principal assumptions (mothers in better condition can produce weanlings in better condition; adult males benefit more from a good condition than adult females) are most probably satisfied in spiny mice due to high maternal investment and strong male-male competition, while the validity of the third one (correlation between weanling and adult condition) is likely, but remains unproved. Spiny mice live in environments with high spatiotemporal variation in resource availability (semideserts, seasonal savannas and woodlands). Under natural conditions, the reproductive value of the offspring produced in good and bad conditions may differ considerably, and thus alternative reproductive strategies that switch according to actual ecological and nutritional conditions may evolve.

  3. (3)

    Spiny mice are social rodents in which social circumstances may also contribute to fitness. The reproductive fate of rodent female may be determined by the presence or absence of maternal kin in the neighbourhood (e.g. Lambin and Yoccoz 1998). Both sexes, but especially males, may be limited by the presence of older cohort of the same sex. Our data from laboratory colonies suggest that adolescent males are driven away by the territorial male while females are tolerated. Local resource competition between the mother and female offspring can, therefore, be a reasonable expectation. The number of females within a family may indicate the expected cost of bearing additional daughters. Consequently, we hypothesized positive association between the number of females in the family group and the sex ratio. Conversely, an excess of juvenile and/or adolescent males in the environment may reduce the reproductive prospect of additional male offspring. Thus, a negative association between the number of immature males and sex ratio should be expected.

Spiny mice are, therefore, a suitable model for critically testing the predictions of the sex ratio theory. We have analysed the data on SSR collected in captive colonies of spiny mice belonging to four closely related species. The aim of our study was to assess (1) deviations from the one to one and/or Fisherian ratios, (2) effects of life history variables and (3) social composition of the breeding groups on SSR.

Material and methods

Spiny mice of the genus Acomys are small rock-dwelling rodents inhabiting Africa and Middle East. Despite their earlier systematic placement, spiny mice are more related to gerbils than to true murids belonging to the subfamily Murinae, which are represented by rats and house mice (Steppan et al. 2004).

Our laboratory colonies of spiny mice were of the following origin: Acomys cahirinus, Abu Simbel archaeological site, southern Egypt (22º 22′ N, 31º 38′ E); A. cilicicus, east of Silifke, southern Turkey (36º 26′ N, 34º 06′ E); Acomys. sp., Zagros, southwestern Iran (28º 56′ N, 52º 32′ E); A. dimidiatus, laboratory strain, Prague zoo, probably from Israel or Sinai. Phylogenetic analysis of mitochondrial control region sequences in these colonies revealed that all of the studied populations/species belong to the cahirinus/dimidiatus group. The former two populations/species belong to the clade of A. cahirinus sensu lato inhabiting North Africa and the eastern Mediterranean region (Crete, Cyprus, Kilikian coast in Anatolia), while the latter two belong to the A. dimidiatus sensu lato clade ranging from Sinai, throughout the Arabian Peninsula and along the coast of the Gulf of Oman from Iran to Pakistan. The specific/subspecific status of the Iranian population, referred to here as Acomys sp., needs further clarification (Frynta and Průšová unpublished results).

The animals were kept in terrariums (60 × 50 × 40 cm or 70 × 60 × 40 cm) or in rodent standard plastic cages (VELAZ T4 , 55 × 32 × 18 cm) under standard laboratory conditions. Wood shavings were used as bedding material, a clay flowerpot with a lateral opening served as a shelter and tree branches for climbing and gnawing were provided as environmental enrichment. Food (standard diet for rats and mice ST1; VELAZ, Czech Republic, supplemented with a mixture of grains, dry bread, apples and herb leaves) and water were available ad libitum.

The spiny mice were kept in family groups consisting of two closely related females (full sisters or uterine sisters), one non-relative male and their descendants. The groups were established from founding animals about 3 months of age (maturing age) and then allowed to breed freely for several months. Manipulation of group structure only occurred in the case of male-to-male aggression between the male founder and his mature male offspring (if so, young males were removed). The experimental groups were regularly checked (either daily or every other day), and each litter was sexed immediately after its detection. Compared to many other rodents, sexing in the spiny mice does not rely solely on ambiguous anogenital distances, as female nipples are clearly visible, even in newborns. Thus, the sexing error approaches zero in these animals. The putative mother was identified, and other circumstances (see below) were thoroughly recorded.

The recorded data included the sex ratio itself (expressed as the proportion of males in the litter), maternal life history variables, such as (1) parity, (2) age in days, (3) postpartum estrus (the litter was considered as conceived postpartum when delivered before day 41 after the previous one) and (4) litter size, and social variables, such as (5) time from the founding of the group (in months), (6) litter order (from the group perspective), (7) number of adult males actually present in the group (aged ≥90 days), (8) number of adult females (see 7), (9) number of breeding females (i.e. those that already gave birth) in the group—coded as the presence of either one or more than one breeding female in the group and further referred to as breeding females, (10) maternal status (first breeding founder, second breeding founder, their daughters), (11) number of immature males and (12) number of immature females (aged ≤90 days for both sexes). It should be noted that maternal body weight was assessed but not included because it was missing in an additional 342 cases. Nevertheless, the statistical models referred below were not substantially affected by the inclusion of maternal body weight, and this factor remained non-significant.

Statistical analysis

We estimated generalized linear mixed models (GLMM) and/or generalized linear models (GLM) in which litter sex ratio was treated as a dependent variable with a binomial distribution and the logit link function was adopted. As the models require complete sets of explanatory variables, all litters with at least one missing value were excluded (n = 212). Therefore, the numbers of litters and newborns used for the computation of the overall sex ratios exceed those included in GLMM analyses.

We first computed GLMMs in which maternal identity was included as a random factor to avoid pseudoreplications (Krackow and Tkadlec 2001). The significance of this random effect was tested using the log-likelihood ratio test based on χ2 distribution, and the effect was found to be non-significant. This allowed us to further use the simpler GLMs instead of GLMMs.

As no effect of any factor associated with maternal identity (i.e. maternal parity, age, postpartum estrus and status) was found to be significant in either the complete GLMM analysis or in any separate GLMM analysis of data concerning individual species, we removed these variables from subsequent GLMs. We then incorporated all remaining explanatory variables into main effects GLMs computed separately for each species and pooled the data. Finally, we computed GLM allowing interactions between species and explanatory variables. The above full models were then reduced to variables with P < 0.1 (see Results).

The size of the effects is presented either graphically and/or as the percentage point difference in sex ratio from the nominal value of 50% due to a unit change of the predictor (CPU). The calculations were performed using the R statistical package ® development core team 2005). Traditional chi-square tests were also conducted to test deviations of the observed sex ratios from the expected equality. Although this approach is theoretically less appropriate than the above-mentioned one and may inflate the significance, it is more intuitive and allows for the inclusion of all records.

Results

We recorded the sex of 4048 newborns of which 1995 were males and 2053 were females. Thus, the overall sex ratio was very close to 1:1 (49.2%, χ2 = 0.831, P = 0.36). Among the studied species, only Acomys sp. from Iran exhibited a significant deviation from the balanced sex ratio (42.5% males; Table 1).

Table 1 Secondary sex ratio in study population categorized according to species
Full size table

Generalized linear mixed models

The initial full GLMM evaluating the effect of all examined factors—i.e. four maternal life history (maternal parity, age, postpartum estrus and litter size) and eight social variables (time from the founding of the group, litter order, number of adult males, number of adult females, number of breeding females, maternal status, number of immature males, number of immature females)—revealed significant effects of species (A. cilicicus CPU = 9.02, A. cahirinus CPU = 7.89, A. dimidiatus CPU = 9.55) and litter size (CPU = -1.89) on the sex ratio. Table 2 presents details on the statistics. The effect of the number of immature males approached the chosen α level of 0.05 (CPU = 0.75; F (1,963) = 3.21, P = 0.0734).

Table 2 Results of analyses using the GLMM and GLM models for testing the secondary sex ratio in four species/populations of the genus Acomys
Full size table

Main-effects GLM

We then excluded the factors associated with maternal identity and performed the GLM analysis. The effect of the number of immature males reached significance (F (1,1661) = 6.55, P = 0.0105), while the significance of effects of the species (F (3,1658) = 2.56, P = 0.0530) and litter size (F (1,1655) = 3.23, P = 0.0722) decreased. When the model was further reduced to include only these three factors, the analysis revealed significant effects of the number of immature males (CPU = 0.81) and litter size (CPU = -1.61), but not of the species.

Generalized linear model allowing interactions with species

When the interactions between the factors included in the original GLM and the species identity were allowed, three of the factors appeared to be significant. The final model (including effects with P < 0.1; see Table 2) revealed significant effects of the number of immature females × species interaction (Fig. 1), the number of immature males × species interaction (Fig. 2), the number of immature males (CPU = −1.19) and the number of breeding females × species interaction. The effects of the species (F (3,1658) = 2.56, P = 0.0530) and litter size (F (1,1657) = 3.27, P = 0.0707) dropped below the significance level.

Fig. 1
figure 1

The predictions of the generalized linear model (GLM) for the number of immature females and species interaction

Full size image
Fig. 2
figure 2

The predictions of the GLM for the number of immature males and species interaction

Full size image

Separate GLMs for each species

We also performed separate analyses for each species. Only the results of the final reduced models that revealed the following significant factors are provided: in A. cahirinus the number of breeding females (CPU = −10.21); in A. cilicicus, the number of immature males (CPU = 7.46) and the number of immature females (CPU = -4−93); in A. dimidiatus, the number of breeding females (CPU = 12.19) and the number of immature females (CPU = −1.89); in Acomys sp., the number of immature females (CPU = 2.70). For detailed statistics see Table 2.

In A. cahirinus, the species represented by the largest dataset, we also calculated partial GLMs, including those for only litters consisting of two and three newborns (i.e. most common litter sizes), respectively. This analysis was performed to avoid the possible interaction of the effects of differential cost by sex and by litter size. No significant effect was revealed by these partial analyses.

Discussion

The SSRs found in our dataset were very close to parity in three of the four populations/species studied. Interestingly, the only population exhibiting a slightly female-biased sex ratio was represented by the smallest sample size. Overall, these results were unsurprising, as balanced ratios at birth in other mammals have been frequently reported (Clutton-Brock and Iason 1986). Nevertheless, this phenomenon cannot be viewed as an unavoidable consequence of chromosomal sex determination. Although mammalian sex ratios are primarily determined by the sperm carrying the sex chromosomes and, consequently, are not biased to any large extent at the time of conception, episodes of sex-specific selective mortality of zygotes and embryos could lead to considerable deviations from parity, especially during early ontogeny (Kirby et al. 1967; Milki et al. 2003). These processes provide a good chance for maternal manipulation of the SSR (compare Grant and Irvin 2009). Therefore, the evolutionary maintenance of unbiased sex ratios requires the presence of a specific stabilizing mechanism, as assumed, for example, by Fisher (1930) who expected natural selection to favour an equal allocation of investment in male and female progeny. In any case, the demonstration of unbiased sex ratios does not mean a falsification of the hypothesis of maternal sex ratio adjustment (Wild and West 2007).

Despite our extensive datasets, we detected only a few significant and consistent effects on sex ratios in spiny mice. Surprisingly, the final models included only three variables, all of which were associated with the composition of the group: the number of breeding females, the number of immature males and the number of immature females. Thus, if any factor plays a role in sex ratio adjustment in these species, it has probably something to do with the actual social environment. This hypothesis corresponds quite well with the recent finding that individual families of spiny mice differ in levels of faecal cortisol metabolites (Nováková et al. 2008). Conversely, a significant interaction with species is revealed in all of the three variables included in our final models. The same factors (e.g. number of immature females; Fig. 1) showed even opposite effects in separate analyses of particular species/populations of spiny mice. In other words, the detected effects are not consistent across the studied species. Although information on the ecology of spiny mice under natural conditions is scarce, these species have fairly comparable requirements, and we can provide no straightforward explanation for the observed inconsistency. Therefore, these results need to be interpreted with caution.

None of the remaining nine explanatory variables appeared to be significant. Our results provide some support for the assumption that the effects of most factors of mammalian SSRs tend to be only small and biologically unimportant, while significant effects are mostly associated with small sample sizes and publication bias (e.g. Festa-Bianchet 1996; Cockburn et al. 2002; Silk et al. 2005, but see Cameron 2004).

Factors such as glucose level in the circulating blood around the time of conception (Cameron et al. 2008) and its correlates (such as fat content in diet; Rosenfeld et al. 2003; Rosenfeld and Roberts 2004) are currently considered to be promising proximate mechanisms of maternal SR manipulation. Not one of our variables provides a direct measure of maternal condition or metabolic status. It can be reasonably expected that some of the examined factors, such as postpartum conception, age and, possibly, social status, are related to maternal condition; however, we have no data on maternal glucose or fat levels, which are difficult to collect in large samples. Thus, covert effects of these variables cannot be excluded.

As there are both good theoretical reasons for maternal manipulation of the offspring sex ratios and data from reliable studies demonstrating the influence of various factors on sex ratios (see Introduction), we avoid drawing over-generalized conclusions from our particular study and instead focus on the peculiarities of the biology of the studied species.

True laboratory animals were selected in order to maximize the reproductive efforts for many generations. However, spiny mice are originally savanna and/or desert dwellers living in unpredictable or seasonal environments. They are therefore likely to be able to regulate their reproduction in response to actual resource availability (e.g. rainfall: Sicard and Fuminier 1996) and the corresponding prospect of the reproductive event. We may only speculate that since spiny mice as wild and more K-selected species (remember their large-sized precocial newborns) strictly avoid breeding whenever they perceive the environmental or social conditions as not fully favourable, there may be reduced variance in body condition and, consequently, no reason for maternal manipulation and/or any other maternal effect on the sex ratio of the progeny.

In conclusion, we found no consistent effects of the studied factors on the sex ratio in spiny mice. Although our correlation approach to the sex ratios has many inherent limitations, it still represents the only easy approach to obtain sufficient datasets from non-domestic mammals. Our results are not interpretable in terms of the most popular sex ratio theories (e.g. the Trivers–Willard hypothesis and/or local resource competition hypothesis). We found fairly balanced SSRs, and suspect that only some factors associated with group composition affect this trait in spiny mice.