INTRODUCTION
Toxoplasma gondii (Coccidia, Apicomplexa) is a highly successful protozoan parasite infecting about one third of humans worldwide. Felines, especially cats, are the definitive hosts of Toxoplasma, while any warm-blooded animal can be an intermediate host and many vertebrates as well as invertebrates can serve as the paratenic host of this parasite (Dubey, Reference Dubey1998). While congenital infection can lead to an abortion or serious health consequences for infected children, postnatal infection of immunocompetent humans with a common non-virulent strain of Toxoplasma usually takes a subacute course that spontaneously converts into latent toxoplasmosis.
Until the beginning of the 21st century, latent toxoplasmosis, the lifelong presence of slowly dividing bradyzoites encysted in various tissues of infected hosts, had mostly been considered asymptomatic and harmless in immunocompetent subjects. However, within the past 20 years several independent studies have shown that latent toxoplasmosis could produce a plethora of consequences in humans. Most strikingly, latent toxoplasmosis is associated with an increased risk of various psychiatric and neurological disorders, such as schizophrenia, bipolar disorder, personality disorder, Parkinson disease, Alzheimer disease, obsessive-compulsive disorder, cryptic epilepsy, recurrent migraines, autism, suicides, homicides and even brain tumours, for a recent review see Flegr (Reference Flegr2013a ). Further, latent toxoplasmosis enhances the occurrence of chronic heart failure, myocarditis, arrhythmia (Paspalaki et al. Reference Paspalaki, Mihailidou, Bitsori, Tsagkaraki and Mantzouranis2001; Yazar et al. Reference Yazar, Gur, Ozdogru, Yaman, Oguzhan and Sahin2006), inflammatory bowel disease (Prandota, Reference Prandota2012), liver cirrhosis (Ustun et al. Reference Ustun, Aksoy, Dagci and Ersoz2004) and diabetes mellitus types 1 and 2 (Gokce et al. Reference Gokce, Yazar, Bayram and Gundogan2008; Krause et al. Reference Krause, Anaya, Fraser, Barzilai, Ram, Abad, Arango, Garcia and Shoenfeld2009) in the infected individuals. Incidence of many diseases positively correlates with the prevalence of latent toxoplasmosis in different countries (Flegr et al. Reference Flegr, Prandota, Sovickova and Israili2014).
A series of interesting reproductive phenotypes are seen in women with latent toxoplasmosis, like slower prenatal (Flegr et al. Reference Flegr, Hrdá and Kodym2005; Kaňková and Flegr, Reference Kaňková and Flegr2007) and postnatal (Kaňkova et al. Reference Kaňkova, Šulc, Křivohlavá, Kuběna and Flegr2012) development of offspring, increased length of pregnancy with significantly higher foetal weight gain post 16th week, especially in RhD negative women (Kaňková et al. Reference Kaňková, Šulc and Flegr2010), and most conspicuously, the duration-specific effect on offspring sex ratio at birth (SRB) (defined as the ratio of male births to female births in a population). Women with a relatively high concentration of anamnestic antibodies, i.e. the women with already latent but still relatively recent Toxoplasma infection, bear a higher proportion of sons while women with a lower concentration of anamnestic antibodies, i.e. women with older Toxoplasma infections, tend to produce a significantly higher proportion of daughters (Kaňková et al. Reference Kaňková, Šulc, Nouzová, Fajfrlik, Frynta and Flegr2007b ). These two phenomena have been confirmed in experimentally infected mice, which delivered 59% male pups 86–120 days post-infection, which was reduced to 40% by 121–222 days (Kaňková et al. Reference Kaňková, Kodym, Frynta, Vavřinová, Kuběna and Flegr2007a ).
Prevalence of toxoplasmosis in different countries depends on local environmental factors, especially on temperature and moisture, kitchen habits and hygienic standards. Toxoplasma seroprevalence shows a drastic global variation, with as low as 4% in Korea to a very high of 78% in Nigeria. Even within Europe it varies from 11% in Norway to 63% in Germany (Table 1). While the global presence of Toxoplasma is bound to affect all the populations, the striking variation in prevalence across populations is likely to produce seroprevalence dependence in these outcomes. Building upon the results from human and laboratory mice studies, we tested whether the variation of offspring sex ratio across countries is influenced by the prevalence of Toxoplasma in humans. We studied the offspring sex predictive power of Toxoplasma seroprevalence by accounting for the known confounding factors across the world, and show that Toxoplasma could be an important mediator of global variation in the offspring sex ratio.
Table 1. Prevalence of latent toxoplasmosis in women of childbearing age in various countries. Third column shows prevalence (%) adjusted to a standard age of 22 years to account for variation in childbearing age across countries using the formula Prevalenceadj = 1 – (1 – Prevalence)^(22/childbearing age) (Lafferty, Reference Lafferty2006). Year in which the given study has been carried out is shown in the fourth column, the fifth column states the number of women in the sample and the last one gives sex ratio at birth (SRB). Data for 88 countries published in Flegr et al. (Reference Flegr, Prandota, Sovickova and Israili2014) have been supplemented with six other countries: Botswana (Joubert and Evans, Reference Joubert and Evans1997), Kenya (Kamau et al. Reference Kamau, Jaoko and Gontier2012), Lebanon (Usta et al. Reference Usta, Seoud, Maarouf, Hobeika and Nassar2006), Namibia (Joubert and Evans, Reference Kamau, Jaoko and Gontier1997), Uganda (Lindstrom et al. Reference Lindstrom, Kaddu-Mulindwa, Kironde and Lindh2006), and Zambia (Kistiah et al. Reference Kistiah, Barragan, Winiecka-Krusnell, Karstaedt and Frean2011)
MATERIAL AND METHODS
Dependent variable
SRB for the period of 2002–2007 was taken from the United States Central Intelligence Agency (CIA) (CIA, 2011) and averaged. A ratio above or below 1 means there are more males or females, respectively, whereas a ratio of 1 indicates equality of both the sexes at birth. These datasets have been criticized for not matching census data of some countries such as Switzerland, Sweden, Norway, Ireland, India and Japan. However, these differences are minor and the CIA data are accepted and widely used by cross-cultural researchers (Navara, Reference Navara2009; Dama, Reference Dama2012). There were three cases of a very high (China, South Korea) and low (Grenada) sex ratio in the sample. In the former two cases, this was due to a strong son preference and female feticide. The reason for the low ratio in Grenada is not known. However, having rerun the analyses excluding these three cases yielded results similar to those reported below.
Independent variables
Prevalence of toxoplasmosis: Data for 88 countries (Flegr et al. Reference Flegr, Prandota, Sovickova and Israili2014) were supplemented with six other countries: Botswana (Joubert and Evans, Reference Joubert and Evans1997), Kenya (Kamau et al. Reference Kamau, Jaoko and Gontier2012), Lebanon (Usta et al. Reference Usta, Seoud, Maarouf, Hobeika and Nassar2006), Namibia (Joubert and Evans, Reference Joubert and Evans1997), Uganda (Lindstrom et al. Reference Lindstrom, Kaddu-Mulindwa, Kironde and Lindh2006), and Zambia (Kistiah et al. Reference Kistiah, Barragan, Winiecka-Krusnell, Karstaedt and Frean2011). The final list contains data on prevalence of toxoplasmosis (seroprevalence) in women of childbearing age published mostly between 1995 and 2008 for 94 countries, 30 European; see Table 1. It is known that prevalence of toxoplasmosis varies between different regions of the same country and also between different age cohorts (Liu et al. Reference Liu, Wei, Gao, Jiang, Lian, Yuan, Yuan, Xia, Liu and Xu2009; Gao et al. Reference Gao, Zhao, He, Wang, Yang, Chen, Shen, Wang, Lv, Hide and Lun2012). Therefore, many of the data points are most probably imprecise, which can highly increase the risk of a Type II error.
The obtained data were adjusted to a standard age of 22 years to eliminate differences in prevalence caused by a different childbearing age in the particular countries using the formula Prevalenceadj = 1 – (1 – Prevalence)^(22/childbearing age) (Lafferty, Reference Lafferty2006). As pointed out by an anonymous referee, this formula was based on some non-realistic assumptions, for example on an assumption of a constant infection rate, and could therefore provide biased results; however, it is widely used in toxoplasmosis research. Also, results obtained with adjusted and unadjusted data in the present study are virtually identical (see Table 2).
Table 2. Side-by-side comparison of categorical regressions of sex ratio at birth on known confounding factors and toxoplasmosis seroprevalence (both adjusted and unadjusted for mother age), respectively
ΔR 2 in the two models in which toxoplasmosis seroprevalence has been included refers to a change in R 2 relative to the model in which the variable has not been included. *Denotes P < 0·05, **P < 0·01 and ***P < 0·001.
a Denotes a trend at P < 0·1.
Toxoplasma prevalence as well as SRB are believed to be influenced by various socioeconomic and environmental factors. To deal with the confounding effect of these factors on the correlation of SRB with toxoplasmosis prevalence, we controlled for the effects of all known factors in regression modelling. These factors are fertility, maternal age, polygyny intensity, wealth, son preference, latitude, parasite stress, nutritional stress, contraceptive use and health status. The rationale behind inclusion of each control variable is presented below.
Coital rate hypothesis of James (Reference James1971), that predicts more female births from intercourse around the time of ovulation, was tested by Barber (Reference Barber2004) for 148 countries revealing a significant correlation of SRB with the intensity of polygyny (r = −0·41) and total fertility (r = −0·60). In the same analysis, SRB was found to be significantly positively correlated with wealth, the proportion of women using any form of contraception and maternal age. While maternal age had been known to influence offspring sex long ago (Lowe and McKeown, Reference Lowe and McKeown1950), influence of parental wealth on offspring sex gained further support by recent studies (Cameron and Dalerum, Reference Cameron and Dalerum2009). Indicator of polygyny for the year of 2009 was obtained from Gender, Institutions and Development Database (OECD, 2009). Polygyny is defined as men having multiple wives simultaneously. Countries were coded as 1 = generally not accepted/polygyny is not legal in a country, 2 = accepted by part of the population/polygyny is only legal for some people, or 3 = generally accepted/polygyny is legal in a country. Contraceptive use, which indicates the proportion of women of reproductive age who are using (or whose partner is using) a contraceptive method for the period of 2005–2009, was obtained from the World Bank (2011). Total fertility estimates for the year of 2008 were taken from the World Bank (2011). Total fertility rate represents the number of children who would have been born to a woman if she were to live to the end of her childbearing years and bear children in accordance with the current age-specific fertility rates. The estimate includes all the children born dead or alive. Gross national income per capita based on purchasing power parity (GNI) was used as a measure of wealth. GNI is the sum of value added by all resident producers plus any product taxes (less subsidies) not included in the valuation of output plus net receipts of primary income (compensation of employees and property income) from abroad. GNI, calculated in national currency, is usually converted to US dollars according to official exchange rates for comparisons across economies. GNI data were taken from the World Bank (2011). Values were log transformed for normality. Mother age was calculated as a mode of the 5-year age block with the highest fertility in a country as explained by Barber (Reference Barber2004).
Navara proposed that latitudinal variation of climatic factors, by its ability to create cross-cultural variation in resource availability and consumer density, could influence offspring sex ratio (Navara, Reference Navara2009). She examined SRB in relation to latitude and associated climatic variables across 202 countries, and showed that SRB was positively correlated with latitude such that tropical populations produce more daughters compared with temperate and subarctic populations. Latitude values, namely rounded latitude for the centroid or centre point of a country expressed in degrees and minutes, for nations were obtained from the CIA World Factbook (CIA, 2011) and numerical values were used irrespective of direction.
Testing the relation of high resource availability and male-biased sex ratios frequently found in many animals, Mathews et al. (Reference Mathews, Johnson and Neil2008) studied a large cohort of British Women and found that 56% of women in the highest third of preconception energy intake bore boys, compared with 45% in the lowest third. Disability-adjusted life years (DALY) lost due to protein energy malnutrition, iodine deficiency, vitamin A deficiency, and iron deficiency were used as an independent measure of nutrition stress. These variables were obtained from the World Health Organization (WHO, 2008) and log transformed for normality. DALY has also been used by sex ratio researchers before (Dama, Reference Dama2011). Henceforth, DALY owing to nutrition will be referred to as nutritional stress.
Some of the cultures, especially from Asian countries, show strong preference for sons. These populations frequently terminate female pregnancies leading to a significant increase in population SRB values (Hesketh and Xing, Reference Hesketh and Xing2006). Son preference (missing women) describes the difference between the number of women that should be alive (assuming no son preference) and the actual number of women in a country. Values for prevalence of son preference for the year of 2009 were obtained from Gender, Institutions and Development Database (OECD, 2009). As son preference results in female selective abortions (Hesketh and Xing, Reference Hesketh and Xing2006), it is necessary to adjust for the level of son preference in the statistical analysis. The variable was treated as ordinal and coded in the following fashion: 1 = male babies generally not preferred over female ones, 2 = male babies preferred over female ones by part of the population, or 3 = male babies generally preferred over female ones.
Dama (Reference Dama2011) tested the relation of measures of parental condition with SRB and found that SRB shows a strong positive correlation with health indicators and a strong negative correlation with mortality rates across the world. As a measure of mortality level for each nation, health adjusted life expectancy at birth (HALE for the year of 2007), estimated by WHO (2009), was used. While life expectancy at birth summarizes the mortality pattern that prevails across all age groups, HALE adds up expectation of life for different health states and measures the average number of years that a person can expect to live in ‘full health’ by taking into account the years lived in less than full health due to disease and/or injury. These two measures reflect the age-standardized summary of mortality in a population; however, only HALE will be used in the present analysis as it is a more complete estimate of health than standard life-expectancy rates. Mortality rates at different stages of life were: infant mortality rate (2009), under-five mortality rate (2009), maternal mortality rate (2008), and adult mortality rate (2008), obtained from the World Bank (2011). While infant mortality rate and maternal mortality ratio are the actual numbers of deaths of infants (during the first year of life per 1000 live births in a given year) and mothers (per 100 000 live births in a given year), under 5 mortality rate and adult mortality rate are the probabilities of dying before reaching the age of five and between the age of 15–65, respectively.
In a follow-up study, Dama (Reference Dama2012) showed that the relation of SRB with latitude and health variables is most likely to be driven by the cross-national variation of parasite stress level as all the statistical analyses pointed towards a strong negative association of parasite intensity and possibility of son births (Dama, Reference Dama2012). Parasite stress across nations was measured by DALY lost due to parasitic diseases (WHO, 2008). One DALY owing to parasites equals one healthy year of life lost per one million people. This measure covers disability due to 28 important parasites across the world. The variable was log transformed for normality. Henceforth, DALY owing to parasites will be referred to as parasite stress.
Control variables
Survival of oocysts and therefore the effectiveness of transmission of parasites from definitive to intermediate host depend on moisture of the soil (Ruiz et al. Reference Ruiz, Frenkel and Cerdas1973), consumption of meat, and number of cats. To control for these factors, we included the average relative humidity in each country (data taken from http://www.climatemps.com/, accessed 2. 4. 2013), sanitation rate, yearly per capita consumption of meat and a number of cats per capita.
Descriptive statistics of the raw variables and their inter-correlation characteristics are summarized in online Supplementary Table 1 and 2, respectively. It must be noted that the datasets used above, especially the control variables, are for different years. However, we have tried to use the data for the dependent and main predictor variable for the same period of time. This discrepancy is owing to a longer sampling frequency for most of the social variables. These differences are unlikely to affect the statistical outcomes, as changes, if any, in social factors are produced very gradually. Hence, most of the cross-cultural studies use data from the nearest sampling year when data for the desired year are not available (Ember and Ember, Reference Ember and Ember2001; Mace et al. Reference Mace, Jordan and Holden2003; Barber, Reference Barber2004).
Estimation of missing data
The following five variables had high counts of missing values: son preference (50·5%), polygyny prevalence (42·1%), contraceptive use (6·3%), maternal mortality rate (2·1%) and adult mortality rate (1·1%). To deal with these, we used the missForest package (Stekhoven, Reference Stekhoven2013a ), available from the Comprehensive R Archive Network (CRAN) and run in the R (R Development Core Team, 2008). Recommended for conducting multiple imputation of mixed data (numeric and factor variables in one data frame) (Starkweather, Reference Starkweather2014), it has been compared with other imputation methods and found to have the least imputation error for both continuous and categorical variables and the smallest prediction difference (error) (Waljee et al. Reference Waljee, Mukherjee, Singal, Zhang, Warren, Balis, Marrero, Zhu and Higgins2013). Default settings were used (Stekhoven, Reference Stekhoven2013b ).
Data reduction
To reduce the number of variables, principal component analysis (PCA) using the IBM SPSS (IBM Corp., 2012) categorical PCA (CATPCA) Optimal Scaling option was performed on the 5 mortality variables (Health adjusted life expectancy, Adult mortality rate, Maternal mortality ratio, Under-five mortality rate, and Infant mortality rate), after confirming that the data were suitable for reduction. The CATPCA settings involved discretizing the numeric variables by means of ranking and selecting variable principal as the normalization method. Dimensions in solution were determined upon several trials in order to obtain the most interpretable structure of loadings. The recommendation of Stevens (Reference Stevens1992) on factor loadings with respect to sample size was followed, and loadings >0·512 were considered significant (all >0·9). The CATPCA of the five mortality variables revealed one factor with eigenvalue greater than one (4·69), accounting for 93·82% of the total variance. This factor was labelled ‘health factor’. Cronbach's alpha reliability coefficient (=0·710) for the health factor indicated satisfactory internal consistency of the computed variable.
Statistical analysis
Analyses were carried out with IBM SPSS 21.0 (IBM Corp., 2012). Normality of the raw data was checked, firstly, by producing skewness and kurtosis values and their respective s.e., from which z-scores were computed and compared with the value of 1·96, as suggested by Field (Reference Field2013). Secondly, we have visually examined individual histograms of all relevant variables. Finally, Shapiro–Wilk's W tests were run. Given the violation of the normality assumption in all analysed variables non-parametric tests were preferred. To regress the SRB on the above-specified variables, a categorical regression analysis was run using the SPSS Optimal Scaling (CATREG) feature. The assumptions of the test were met since the number of valid cases exceeded the number of predictor variables plus one. Scale and ordinal variables were treated as numeric and ordinal, respectively, and all were discretized by multiplying. A numerical initial configuration was selected, as recommended when no variables are treated as nominal (IBM Corp., 2011). Perfect multicollinearity (intercorrelations >0·9) did not appear a serious problem, which was further supported by reviewing the variance inflation factors (VIF), which were nowhere near the value of 10, and the average VIF was not greater than 1, as recommended by Field (Reference Field2013). Moreover, parallel analyses with multiple linear regression (not reported here in detail) showed comparable results. In both the cases, we nevertheless decided to err on the side of caution and employ the ridge regression option using default settings and 0·632 bootstrap with 50 samples for resampling. Ridge regression artificially reduces correlation coefficient of each pair of variables by incorporating a ridge parameter to the diagonal of a correlation matrix of highly collinear independent variables, leading to reduced error variance of estimators. Based on this principle, ridge regression copes with the collinearity problem (Fox, Reference Fox1991).
RESULTS
Regression of SRB on the above-specified independent variables revealed that SRB was negatively predicted by Toxoplasma prevalence, fertility, parasite stress, and polygyny intensity, and positively by the health factor, and son preference. This means that a greater number of male offspring was produced in populations with lower Toxoplasma prevalence, fertility, less intense parasite stress and polygyny, better health status, and a tendency to prefer sons (Table 2; Fig. 1). Importantly, side-by-side comparisons of regression models including Toxoplasma prevalence (adjusted and unadjusted) and not, respectively, revealed a significant change in R 2 following the inclusion of the variable in the model (Table 2 and online Supplementary Table 4 and 5), showing it to be an important predictor of SRB.
Fig. 1. Relation of toxoplasmosis seroprevalence and population secondary sex ratio across 94 populations. The standard residuals were computed for the model containing all 10 covariates but not the toxoplasmosis prevalence.
When countries were divided into European and non-European ones, the results were different. While in the European countries the model was not significant, outside Europe SRB was predicted in the same direction by the same variables as in the whole set of countries excluding polygyny intensity but including contraception, with countries in which more males are born exhibiting greater contraceptive use (online Supplementary Table 3 and 4).
Four additional regression models were fitted to investigate the influence of control variables on the prediction of SRB by Toxoplasma prevalence (Table 3). Results show that all the models were significant at least at P < 0·01 and the βs of Toxoplasma prevalence were not much affected by the sequential addition of humidity, sanitation rate, cat ownership, and meat consumption. This further strengthens the hypothesis that Toxoplasma is likely to influence the offspring sex ratio in humans.
Table 3. Sequential addition of control variables did not change the significant prediction of sex ratio at birth by toxoplasmosis prevalence in the categorical regression models
*Denotes a model/predictor significant P < 0·05, **<0·01, and ***<0·001 # Standard independent variables included are Contraceptive use, Health factor, Latitude, log Wealth, Mother age, Nutrition stress, Parasite stress, Polygyny intensity, Son preference, Total fertility and Toxoplasmosis prevalence. The predictor-related values represent βs.
Furthermore, as suggested by an anonymous referee, SRB may explain a large part of son preference, and thus the latter may not be suitable for inclusion in the regression models. Although this was clearly not the case in the present study (Kendall's Tau = 0·226; P < 0·01), we ran the same regression models as those reported in Table 2 with the only difference that the son preference variable has not been included. These analyses yielded results virtually identical to those reported in Table 2 (see online Supplementary Table 5 for comparison).
DISCUSSION
Our analysis showed that prevalence of toxoplasmosis is negatively correlated with offspring sex ratio in 94 countries. The correlation can be detected in the whole data set and in the subset of non-European countries, but could not be detected in European countries.
The interpretation of results of ecological regression studies is sometimes complicated, especially if aggregated data are used for the estimation of the strength and direction of the influence of particular factors within a population (Guthrie and Sheppard, Reference Guthrie and Sheppard2001; Wakefield and Salway, Reference Wakefield and Salway2001). Therefore, the existence of strong effects of prevalence of toxoplasmosis on SRB, suggested by results of an ecological study, should be confirmed by independent case controls or cohort studies. The results of a cohort study performed on a population of European women as well as a study performed on experimentally infected female mice suggest that Toxoplasma infection increases SRB in the first phases of latent toxoplasmosis while decreases the offspring sex ratio in the latter phases. Women with high titres of specific anti-Toxoplasma IgG antibodies (i.e. women probably infected with Toxoplasma less than 2 years before pregnancy) had a SRB of 0·72, while women with low titres of specific anamnestic antibodies had a SRB of 0·45 (Kaňková et al. Reference Kaňková, Šulc, Nouzová, Fajfrlik, Frynta and Flegr2007b ). Similarly, female mice that delivered 86–120 days after Toxoplasma infection had a SRB of 0·59 while those that delivered at 121–222 days had a SRB of 0·40 (Kaňková et al. Reference Kaňková, Kodym, Frynta, Vavřinová, Kuběna and Flegr2007a ). The increased sex ratio is speculated to be caused either by a higher probability of survival of more immunogenic male embryos (Kaňková and Flegr, Reference Kaňková and Flegr2007) or by an increased level of testosterone around the time of fertilization (James, Reference James2010). The decreased sex ratio in women and female mice infected with Toxoplasma for a long time is considered to be a manifestation of the Trivers–Willard effect, i.e. the increased probability of birth of female offspring in females in poor physical conditions (Flegr, Reference Flegr2010). It is not important in the present context whether the decreased sex ratio is just a side-effect of the impaired health of infected females or if it is the evolutionary adaptation that results in increased or, rather, not-so-impaired fitness of the infected females. On the population level, the direction of Toxoplasma infection-associated effect on sex ratio will depend on the fraction of recently infected women in a birth giving age in a particular population. In the high prevalence countries, most women are probably infected by contact with contaminated food and water in the early childhood, i.e. long before their birth giving age. In such countries, prevalence of toxoplasmosis should correlate negatively with sex ratio, which is the pattern observed in non-European countries. The situation will be more complicated in the low prevalence, low fertility countries. For example in the Czech Republic, an increased rate of incidence of toxoplasmosis occurs in women in a birth giving age; in men and women the seroprevalence of toxoplasmosis increases between the ages of 19–39 from 25·5 to 27 and 31·1 to 46·3%, respectively (Kodym et al. Reference Kodym, Malý, Švandová, Lekatková, Badoutová, Vlková, Beneš and Zástěra2000). Such an increase is hypothesized to occur either due to manipulation with undercooked/raw meat when the young women start cooking in their own households or by transmission during unprotected intercourse with infected men (Dass et al. Reference Dass, Vasudevan, Dutta, Soh, Sapolsky and Vyas2011; Flegr, Reference Flegr2013a ; Holub et al. Reference Holub, Flegr, Dragomirecka, Rodriguez, Preiss, Novak, Cermak, Horacek, Kodym, Libiger, Höschl and Motlova2013). Therefore, a higher fraction of Toxoplasma-infected women in low prevalence countries could acquire their infection a short time before pregnancy and increased offspring sex ratio of recently-infected women could neutralize or even reverse the negative correlation between prevalence of toxoplasmosis and the offspring sex ratio. Such an effect in recently infected women could be especially strong in low-fertility countries because the probable duration of Toxoplasma infection positively correlates with the age of women and therefore also with parity. Therefore, a stronger positive effect of toxoplasmosis on sex ratio can be expected to occur rather in primiparous women than in multiparous women in low prevalence countries.
Of course, the absence of a significant effect of prevalence of toxoplasmosis on SRB in the European countries subset could also have another explanation. It has been observed, for example, that Rhesus factor (Rh) phenotype modifies the effects of toxoplasmosis on the human organism (Kaňková et al. Reference Kaňková, Šulc and Flegr2010) and positive heterozygotes are strongly protected against these effects (Novotna et al. Reference Novotna, Havlicek, Smith, Kolbekova, Skallova, Klose, Gasova, Pisacka, Sechovska and Flegr2008; Flegr, Reference Flegr2016). The frequency of Rh heterozygotes is much higher in the Caucasian population of European countries than populations in other parts of the globe. It is, however, also possible that the observed association between prevalence of toxoplasmosis and SRB is caused by an unknown factor that is not present in Europe and that independently correlates both with prevalence of toxoplasmosis and SRB.
In our opinion, it is probable that each set of the countries used in the previous nine ecological regression studies dealing with toxoplasmosis published by other authors in the past 10 years would give a slightly different result of the analysis. To maximally decrease the risk of subjectivity in the selection of countries, we included data for all the countries (n = 94) available as of 15th August 2013; which represented the largest ever data set analysed for biological influence of Toxoplasma.
It is also highly probable that at least some of the Toxoplasma prevalence data are incorrect. As far as we know, national surveys for latent toxoplasmosis are not systematically performed in any country, or at least, no results of such studies have been published. Our experience with such a survey performed in a relatively small and rather ethnically and sociologically homogeneous Czech Republic showed that the prevalence varies drastically among different parts of a country (Kodym et al. Reference Kodym, Malý, Švandová, Lekatková, Badoutová, Vlková, Beneš and Zástěra2000). Therefore, it is difficult to estimate the average prevalence of toxoplasmosis in women of birth giving age in a particular country in a single study. Hence, we always searched for all available data concerning the time period of 1990–2008 and, whenever possible, we also took into consideration the results on the prevalence published earlier. It should be noted that a lack of precision in the prevalence data increases the risk of false negative but not of false positive results of statistical tests. Lack of precision (stochastic noise) in the prevalence data could bias the estimate of regression parameters (e.g. betas) and increase the risk of false negative results of statistical tests. Stochastic noise alone cannot increase the risk of false positive results of studies. However, it is possible that the precision of the prevalence data (or the SRB data) is somehow related to the prevalence of toxoplasmosis. It is, for example, known that toxoplasmosis influences some personality traits of infected humans, including conscientiousness (Lindová et al. Reference Lindová, Příplatová and Flegr2012) and neuroticism (Flegr et al. Reference Flegr, Preiss and Klose2013), and such effects explain a statistically significant portion of the variance in aggregate neuroticism among populations (Lafferty, Reference Lafferty2006). It is therefore possible that such a behavioural effect of toxoplasmosis could result in a systematic error in the available prevalence (or other) data, which could result in false positive results of an ecological study. Therefore, even the positive results of our study should be approached with caution and should only be considered preliminary until confirmed in other cross-sectional or longitudinal studies. For example, our study can be repeated on regional data from a large country, e.g. the USA, Mexico, or France, for which the sex ratio as well as toxoplasmosis prevalence data are probably available. It is also necessary to confirm the results of an earlier cohort study (Kaňková et al. Reference Kaňková, Šulc, Nouzová, Fajfrlik, Frynta and Flegr2007b ), showing decreased probability of birth of male offspring in women with ‘old’ Toxoplasma infections, on another population. This study has been supported by the results of the latter mice-infection study (Kaňková et al. Reference Kaňková, Kodym, Frynta, Vavřinová, Kuběna and Flegr2007a ); however, it has not been repeated on any other human population.
Another possible source of error are temporal changes in the incidence of toxoplasmosis. In many parts of world, the prevalence of toxoplasmosis is changing, mostly having decreased in the past two decades. It is highly probable that the offspring sex ratio reflects the situation in an (unknown) past, rather than the current prevalence of toxoplasmosis. Again, the existence of such a delay could increase the risk of false negative, rather than false positive results of studies.
The observed correlation can be most parsimoniously explained by the effect of latent toxoplasmosis on secondary sex ratio, i.e. by the effect demonstrated in earlier studies including the experimental infection study (Kaňková et al. Reference Kaňková, Kodym, Frynta, Vavřinová, Kuběna and Flegr2007a ). The opposite causality, i.e. influence of the offspring sex ratio on prevalence of toxoplasmosis, is rather improbable. However, we also cannot exclude a possibility that some unknown factor can correlate both with the offspring sex ratio and the prevalence of toxoplasmosis. Such a factor could be either a particular disease, particular epidemiological factor for toxoplasmosis, e.g. rural vs urban style of life, or even some cultural habit. The difficulty with studying causality relationships on the Toxoplasma–human model are discussed in detail in Flegr (Reference Flegr2013b ).
The increased probability of producing a baby girl by a Toxoplasma-infected woman could be the result of impaired health. The incidence of many diseases, including cardiovascular diseases and certain forms of cancer, positively correlates with the prevalence of toxoplasmosis in different countries (Flegr et al. Reference Flegr, Prandota, Sovickova and Israili2014). The possible physiological basis for the correlation of SRB and Toxoplasma prevalence could be derived from the sex differences in the costs of offspring production. Male foetus grows faster (Marsal et al. Reference Marsal, Persson, Larsen, Lilja, Selbing and Sultan1996) and requires significantly higher parental investment during gestation (Tamimi et al. Reference Tamimi, Lagiou, Mucci, Hsieh, Adami and Trichopoulos2003), which means that, to produce sons, women should be in superior body condition to meet higher physiological costs required. Indeed, male foetuses are more often aborted spontaneously than a female fetus due to various stressors (Mace et al. Reference Mace, Jordan and Holden2003; Boklage, Reference Boklage2005). It is, therefore, possible that various factors that influence maternal investment ability (health impairment caused by toxoplasmosis in the present analysis) are more likely than genetic and geographical factors to form the basis for the striking cross-cultural variation in SRB.
Of course, this explanation of the phenomenon is not the only explanation available. Alternatively, the decreased offspring sex ratio can be the result of a decreased level of testosterone (James, Reference James2010, Reference James2012) observed in infected women (Flegr et al. Reference Flegr, Lindová and Kodym2008a , Reference Flegr, Lindová, Pivoňková and Havlíček b ) and mice (Kaňková et al. Reference Kaňková, Kodym and Flegr2011). However, the Trivers–Willard effect-related hypothesis of the observed effect should be preferentially tested because if it is correct, then the effect of latent toxoplasmosis on public health could be much more serious than it is usually supposed to be. Toxoplasma gondii infects most species of homoeothermic animals. This suggests that abundance of this parasite could be one of the most important ecological factors influencing the global variation of offspring sex ratio not only in humans, but possibly also in many other animal species.
SUPPLEMENTARY MATERIAL
The supplementary material for this article can be found at http://dx.doi.org/10.1017/S0031182016000597.
ACKNOWLEDGEMENT
We thank L. Lanchava for improving the manuscript.
FINANCIAL SUPPORT
J.F. and L.M.N.'s work was supported by the Charles University Research Centre (UNCE 204004). J.F. was further supported by the Grant Agency of the Czech Republic (Grant No. P303/16/20958). L.M.N.'s work was further funded by the project ‘National Institute of Mental Health (NIMH-CZ)’, under grant number ED2.1.00/03.0078, and the European Regional Development Fund, and the Ministry of Education, Youth and Sports – Institutional Support for Longterm Development of Research Organizations – Charles University, Faculty of Humanities (project PRVOUK P20).
