Understanding childlessness
Verweij, Renske
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from
it. Please check the document version below.
Document Version
Publisher's PDF, also known as Version of record
Publication date:
2019
Link to publication in University of Groningen/UMCG research database
Citation for published version (APA):
Verweij, R. (2019). Understanding childlessness: Unravelling the link with genes and socio-environment.
Rijksuniversiteit Groningen.
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.
CHAPTER
*Currently under review by an international peer-reviewed journal
Renske M. Verweij, Melinda C. Mills, Gert Stulp, Ilja M Nolte,
Nicola Barban, Felix C. Tropf, Douglas T Carrell, Kenneth
I Aston, Krina T Zondervan, Nilufer Rahmioglu, Marlene
Dalgaard, Carina Skaarup, M. Geoffrey Hayes, Andrea
Dunaif, Guang Guo & Harold Snieder
A sociogenomic approach to
childlessness: Using multiple polygenic
risk scores and socio-demographic
factors to explain childlessness
ABSTRACT
Biological, genetic, and socio-demographic factors are all important in explaining reproductive behavior, yet these factors are typically studied in isolation. In this study we explore an innovative sociogenomic approach, which entails including key socio-demographic (marriage, education, occupation, religion, cohort) and genetic factors related to both behavioral (age at first birth (AFB), number of children ever born (NEB)) and biological fecundity-related outcomes (endometriosis, age at menopause and menarche, polycystic ovary syndrome, azoospermia, testicular dysgenesis syndrome) to explain childlessness. We examine the predictive power of both sets of factors as well as the interplay between them. We derive polygenic risk scores (PGS) from recent genome-wide association studies (GWAS) and apply these in the Health and Retirement Study (N=10,686) and Wisconsin Longitudinal Study (N=8,284). Both socio-demographic and genetic factors were associated with childlessness. Where socio-demographic factors explain 19-46% in childlessness the current PGS explain less than 1% of the variance, and only PGSs from large GWASs have predictive power. Our findings also indicate that genetic and socio-demographic factors are not independent, with higher PGSs for AFB related to higher education and age at marriage. The predictive power of polygenic risk scores in explaining childlessness is currently limited, but is expected to increase in the future due to rapidly increasing sample sizes of GWASs.
5.1 INTRODUCTION
Childlessness has increased in many Western countries, from 10% in the 1970s to currently 15% in the US (Frejka, 2017). Childlessness can have far reaching consequences, including changing the age composition of the population and lower well-being among the involuntary childless (Hansen et al., 2009; Sleebos, 2003).
Three parallel strands of research have examined changes in contemporary fertility patterns. Firstly, the social sciences examined socio-demographic factors such as educational attainment, occupational behavior, religiosity, marital status and birth cohort (Balbo et al., 2013). Secondly, medical research has focused on fecundity, infertility or the biological ability to conceive such as sperm defects and ovulatory, cervical, fallopian tube and uterine problems (Blundell, 2007). Thirdly, a growing body of research focuses on the genetics of fertility outcomes, with twin and family studies showing that genetics may explain up to 50% (Mills & Tropf, 2015; Tropf, Barban, et al., 2015; Verweij et al., 2017). Recent Genome-wide Association Study (GWAS) discoveries have isolated genetic markers for reproductive behavior such as the timing and number of children (Barban et al., 2016) and more biologically based infertility traits related to sperm defects or the timing of menopause (e.g., Day et al., 2015; Painter et al., 2011), allowing us for the first time to include an individual’s genetic architecture as predictors in our statistical models.
Until now, these three strands of research have existed in isolation (Mills & Tropf, 2015), largely due to lack of data, training or realization of the importance of adopting a combined sociogenomic approach. The result is a lack of understanding of the relative contribution of biological and genetic versus socio-demographic factors – and their interaction – associated with childlessness. We also do not know whether estimates based solely on socio-demographic factors are biased due to their correlation with an individual’s genetic makeup (Tropf & Mandemakers, 2017) or if genetic propensities are more influential at certain life phases or interact with socio-demographic factors to be more influential in particular groups. In this study we explore an innovative sociogenomic approach. Using known socio-demographic measures and results from recent GWAS discoveries, we apply a novel design in which polygenic risk scores (PGSs) are created for a variety of behavioral and infertility-related reproductive outcomes. Due to the novelty of our design, we use two independent datasets to replicate our results, namely the two US-based Health and Retirement Study (HRS, N=10,686) and the Wisconsin Longitudinal Study (WLS, N=8,284). Both include individuals born between 1920-1960, where childlessness rose from 6% among women born in 1935 to 16% around 1950 (Human Fertility Database, 2017a) (see Supplementary Material Figure 1). We first introduce our conceptual model, followed by an explanation of the data and methods, main results and implications for future research related to childlessness and beyond.
ABSTRACT
Biological, genetic, and socio-demographic factors are all important in explaining reproductive behavior, yet these factors are typically studied in isolation. In this study we explore an innovative sociogenomic approach, which entails including key socio-demographic (marriage, education, occupation, religion, cohort) and genetic factors related to both behavioral (age at first birth (AFB), number of children ever born (NEB)) and biological fecundity-related outcomes (endometriosis, age at menopause and menarche, polycystic ovary syndrome, azoospermia, testicular dysgenesis syndrome) to explain childlessness. We examine the predictive power of both sets of factors as well as the interplay between them. We derive polygenic risk scores (PGS) from recent genome-wide association studies (GWAS) and apply these in the Health and Retirement Study (N=10,686) and Wisconsin Longitudinal Study (N=8,284). Both socio-demographic and genetic factors were associated with childlessness. Where socio-demographic factors explain 19-46% in childlessness the current PGS explain less than 1% of the variance, and only PGSs from large GWASs have predictive power. Our findings also indicate that genetic and socio-demographic factors are not independent, with higher PGSs for AFB related to higher education and age at marriage. The predictive power of polygenic risk scores in explaining childlessness is currently limited, but is expected to increase in the future due to rapidly increasing sample sizes of GWASs.
5.1 INTRODUCTION
Childlessness has increased in many Western countries, from 10% in the 1970s to currently 15% in the US (Frejka, 2017). Childlessness can have far reaching consequences, including changing the age composition of the population and lower well-being among the involuntary childless (Hansen et al., 2009; Sleebos, 2003).
Three parallel strands of research have examined changes in contemporary fertility patterns. Firstly, the social sciences examined socio-demographic factors such as educational attainment, occupational behavior, religiosity, marital status and birth cohort (Balbo et al., 2013). Secondly, medical research has focused on fecundity, infertility or the biological ability to conceive such as sperm defects and ovulatory, cervical, fallopian tube and uterine problems (Blundell, 2007). Thirdly, a growing body of research focuses on the genetics of fertility outcomes, with twin and family studies showing that genetics may explain up to 50% (Mills & Tropf, 2015; Tropf, Barban, et al., 2015; Verweij et al., 2017). Recent Genome-wide Association Study (GWAS) discoveries have isolated genetic markers for reproductive behavior such as the timing and number of children (Barban et al., 2016) and more biologically based infertility traits related to sperm defects or the timing of menopause (e.g., Day et al., 2015; Painter et al., 2011), allowing us for the first time to include an individual’s genetic architecture as predictors in our statistical models.
Until now, these three strands of research have existed in isolation (Mills & Tropf, 2015), largely due to lack of data, training or realization of the importance of adopting a combined sociogenomic approach. The result is a lack of understanding of the relative contribution of biological and genetic versus socio-demographic factors – and their interaction – associated with childlessness. We also do not know whether estimates based solely on socio-demographic factors are biased due to their correlation with an individual’s genetic makeup (Tropf & Mandemakers, 2017) or if genetic propensities are more influential at certain life phases or interact with socio-demographic factors to be more influential in particular groups. In this study we explore an innovative sociogenomic approach. Using known socio-demographic measures and results from recent GWAS discoveries, we apply a novel design in which polygenic risk scores (PGSs) are created for a variety of behavioral and infertility-related reproductive outcomes. Due to the novelty of our design, we use two independent datasets to replicate our results, namely the two US-based Health and Retirement Study (HRS, N=10,686) and the Wisconsin Longitudinal Study (WLS, N=8,284). Both include individuals born between 1920-1960, where childlessness rose from 6% among women born in 1935 to 16% around 1950 (Human Fertility Database, 2017a) (see Supplementary Material Figure 1). We first introduce our conceptual model, followed by an explanation of the data and methods, main results and implications for future research related to childlessness and beyond.
5.2 CONCEPTUAL MODEL AND EXPECTATIONS
Figure 1 provides an overview of our conceptual model, illustrating that we first assess the predictive power of socio-demographic factors, but also genetic factors related to reproductive behavior (timing, number of children) and biological reproductive traits (e.g., menarche, sperm defects) on childlessness. We also study the interplay between these socio-demographic and genetic factors and finally, assess GxG and GxE correlations. Although we are unable to distinguish between voluntary (childfree) and involuntary childlessness, genetic factors related to biological reproductive traits may serve as a proxy for involuntary childlessness. Due to biological differences between men and women, we likewise examine sex differences (Verweij et al., 2017), and explore differences by ethnic groups.
5.2.1 Socio-demographic factors
The central socio-demographic factors, which we also include in our model, are education, occupation, religion, marriage and birth year (Balbo et al., 2013). A core predictor of childlessness is higher education, particularly for women (Tropf & Mandemakers, 2017), as well as working full-time and working in a high status or particular occupation (Balbo et al., 2013). Core reasons for women’s postponement, which sometimes leads to childlessness, are related to role incompatibility, work-life reconciliation and ability to find a partner by a particular age (Balbo et al., 2013). The picture is different for men, as highly educated and full-time employed men are more attractive marriage partners and less often remain childless (Balbo et al., 2013; Keizer et al., 2008). Particularly in the US and during the period studied, religion is an important predictor for normative differences in reproductive behavior, with Catholic and Protestant women on average having more children (Frejka & Westoff, 2008). Finding a partner and age at marriage are also key factors (Keizer et al., 2008), particularly in the period that we examined in the US, where most childbearing happened within marriage (Ventura & Bachrach, 2000).
5.2.2 Polygenic risk scores
In this study we adopt a novel approach to include both biological reproductive scores (e.g., sperm count, endometriosis) along with behavioral genetic scores (i.e., timing and number of children). These cover diverse genetic facets of reproduction, with the assumption that biological reproductive scores are potentially stronger predictors of involuntary childlessness due to infertility, and reproductive behavioral genetic scores are more linked to reproductive choice and voluntary childlessness.
Biological reproductive trait scores. These PGSs are proximal phenotypes, or in other
words, we anticipate the biological pathway between the genetic markers and outcomes to be more traceable and directly related to infertility. For women, ovulatory, cervical, fallopian tube and uterine problems are most likely to cause infertility (Blundell, 2007). We aim to capture these genetic risks by including PGSs for polycystic ovary syndrome (PCOS) (Hayes et al., 2015), (which mainly cause ovulatory problems), endometriosis (Painter et al., 2011) (which influences the ovaries and fallopian tubes), age at menarche (Day et al., 2017), and
age at menopause (Day et al., 2015) (which determine women’s reproductive life span). For men, sperm defects are the most likely cause for infertility, therefore we include PGSs for azoospermia (Aston & Carrell, 2009) and testicular dysgenesis syndrome (TDS) (Dalgaard et al., 2012).
Endometriosis affects 10% of premenopausal women and is characterized by endometrial-like tissue outside of the uterus causing pelvic pain and subfertility (Nnoaham et al., 2011). The risk of infertility is about 20 times greater for women with endometriosis (Strathy et al., 1982) and 25-50% of women with infertility have endometriosis, with 30-50% of women with endometriosis experiencing infertility. It is a complex trait, caused by genetic and socio-environmental factors and the interaction between the two (Rahmioglu, Montgomery, & Zondervan, 2015). Twin studies show a heritable component of endometriosis of around 52% (Treloar et al., 1999).
Polycystic ovary syndrome (PCOS) is characterized by chronic oligo- or anovulation, hyperandrogenism (biochemical and/or clinical evidence of male hormone excess) and polycystic ovarian morphology (Ehrmann, 2005). It accounts for 80% of the anovulatory infertility, with 5-15% of women in reproductive ages having PCOS, depending on the diagnostic criteria applied (Balen et al., 2016). Many studies have found familial aggregation, estimating heritability of around 72% (Jones & Goodarzi, 2016).
Age at menarche (i.e., first menstruation) is an indicator of female pubertal development, with heritability of around 50% (Snieder et al., 1998). On the one hand, an early age at
Figure 1 | Conceptual model on the pathways from the three sets of factors leading to (both
volun-tary and involunvolun-tary) childlessness. Abbreviations: PCOS, polycystic ovary syndrome; TDS, testicular
5.2 CONCEPTUAL MODEL AND EXPECTATIONS
Figure 1 provides an overview of our conceptual model, illustrating that we first assess the predictive power of socio-demographic factors, but also genetic factors related to reproductive behavior (timing, number of children) and biological reproductive traits (e.g., menarche, sperm defects) on childlessness. We also study the interplay between these socio-demographic and genetic factors and finally, assess GxG and GxE correlations. Although we are unable to distinguish between voluntary (childfree) and involuntary childlessness, genetic factors related to biological reproductive traits may serve as a proxy for involuntary childlessness. Due to biological differences between men and women, we likewise examine sex differences (Verweij et al., 2017), and explore differences by ethnic groups.
5.2.1 Socio-demographic factors
The central socio-demographic factors, which we also include in our model, are education, occupation, religion, marriage and birth year (Balbo et al., 2013). A core predictor of childlessness is higher education, particularly for women (Tropf & Mandemakers, 2017), as well as working full-time and working in a high status or particular occupation (Balbo et al., 2013). Core reasons for women’s postponement, which sometimes leads to childlessness, are related to role incompatibility, work-life reconciliation and ability to find a partner by a particular age (Balbo et al., 2013). The picture is different for men, as highly educated and full-time employed men are more attractive marriage partners and less often remain childless (Balbo et al., 2013; Keizer et al., 2008). Particularly in the US and during the period studied, religion is an important predictor for normative differences in reproductive behavior, with Catholic and Protestant women on average having more children (Frejka & Westoff, 2008). Finding a partner and age at marriage are also key factors (Keizer et al., 2008), particularly in the period that we examined in the US, where most childbearing happened within marriage (Ventura & Bachrach, 2000).
5.2.2 Polygenic risk scores
In this study we adopt a novel approach to include both biological reproductive scores (e.g., sperm count, endometriosis) along with behavioral genetic scores (i.e., timing and number of children). These cover diverse genetic facets of reproduction, with the assumption that biological reproductive scores are potentially stronger predictors of involuntary childlessness due to infertility, and reproductive behavioral genetic scores are more linked to reproductive choice and voluntary childlessness.
Biological reproductive trait scores. These PGSs are proximal phenotypes, or in other
words, we anticipate the biological pathway between the genetic markers and outcomes to be more traceable and directly related to infertility. For women, ovulatory, cervical, fallopian tube and uterine problems are most likely to cause infertility (Blundell, 2007). We aim to capture these genetic risks by including PGSs for polycystic ovary syndrome (PCOS) (Hayes et al., 2015), (which mainly cause ovulatory problems), endometriosis (Painter et al., 2011) (which influences the ovaries and fallopian tubes), age at menarche (Day et al., 2017), and
age at menopause (Day et al., 2015) (which determine women’s reproductive life span). For men, sperm defects are the most likely cause for infertility, therefore we include PGSs for azoospermia (Aston & Carrell, 2009) and testicular dysgenesis syndrome (TDS) (Dalgaard et al., 2012).
Endometriosis affects 10% of premenopausal women and is characterized by endometrial-like tissue outside of the uterus causing pelvic pain and subfertility (Nnoaham et al., 2011). The risk of infertility is about 20 times greater for women with endometriosis (Strathy et al., 1982) and 25-50% of women with infertility have endometriosis, with 30-50% of women with endometriosis experiencing infertility. It is a complex trait, caused by genetic and socio-environmental factors and the interaction between the two (Rahmioglu, Montgomery, & Zondervan, 2015). Twin studies show a heritable component of endometriosis of around 52% (Treloar et al., 1999).
Polycystic ovary syndrome (PCOS) is characterized by chronic oligo- or anovulation, hyperandrogenism (biochemical and/or clinical evidence of male hormone excess) and polycystic ovarian morphology (Ehrmann, 2005). It accounts for 80% of the anovulatory infertility, with 5-15% of women in reproductive ages having PCOS, depending on the diagnostic criteria applied (Balen et al., 2016). Many studies have found familial aggregation, estimating heritability of around 72% (Jones & Goodarzi, 2016).
Age at menarche (i.e., first menstruation) is an indicator of female pubertal development, with heritability of around 50% (Snieder et al., 1998). On the one hand, an early age at
Figure 1 | Conceptual model on the pathways from the three sets of factors leading to (both
volun-tary and involunvolun-tary) childlessness. Abbreviations: PCOS, polycystic ovary syndrome; TDS, testicular
menarche might be beneficial for early fertility (Guldbrandsen et al., 2014), but on the other hand it is found to have a negative effect on fertility (Weghofer, Kim, Barad, & Gleicher, 2013). Age at menopause (i.e., last menstruation) as well as very early menopause (before the age of 40) is often clustered in families, and heritability studies estimate a genetic component of 30 to 85% (Snieder et al., 1998).
For men, sperm defects most often cause infertility (Blundell, 2007), which we hope to capture with the PGSs for TDS and oligozoospermia or azoospermia. Some male infertility factors are associated with chromosomal abnormalities. However, familial aggregation of male infertility, in the absence of specific genetic aberrations, hint towards multifactorial inheritance. There are comparatively fewer male infertility-related GWA studies and discoveries, many studies characterized by very small samples (Dalgaard et al., 2012). Azoospermia, which is a zero sperm count, effects around 1% of the male population, and is the reason for infertility in 10-15% of men (Willott, 1982). Oligozoospermia is a sperm count of less than 15 or sometimes 20 million per milliliter in the semen (Cooper et al., 2010). TDS is a group of diseases related to male reproductive health, namely infertility due to poor sperm quality, testicular cancer, undescended testes (cryptorchidism) and hypospadias (abnormality where the urinary opening is not at the head of the penis) (Skakkebæk, 2003). Familial clustering of TDS has been shown (Joffe, 2007).
Reproductive behavioral scores. Medical research has often only examined
infertility-related outcomes on smaller sex-specific medical samples. A recent GWAS of reproductive behavior of both sexes isolated genetic markers related to the age at first birth (AFB) and number of children ever born (NEB) (Barban et al., 2016). These PGSs are viewed as behavioral since they likely measure not only biological and genetic components, but also psychological and choice aspects (e.g., personality, intentions) and socio-environmental contexts (e.g., childcare, work-family reconciliation) (Tropf et al., 2017). These GWASs, however, measure very concrete reproductive outcomes, and the identified genetic loci are related to markers found for endometriosis, PCOS, sperm defects, menarche, menopause and level of education.
5.2.3 Interplay between socio-demographic factors and genetic risk scores
We adopt two lines of reasoning to understand the interplay between socio-demographic and genetic factors. First, we anticipate a birth cohort effect based on studies that demonstrate changes in heritability, or in other words, that there are differences in the relationship between genes and reproductive outcomes over time (Tropf et al., 2017). In the US (Briley et al., 2015), the UK (Tropf, Barban, et al., 2015) and Denmark (Kohler, Rodgers, et al., 2002) findings indicate that heritability of reproductive-related outcomes was higher around the second demographic transition, for individuals born between 1935 and 1950, than in periods before and after. The suggested mechanism is that in times where individual freedom is more pervasive, genetic propensities and thus heritability should be higher due to the rise in realization of individual genetic drivers (Kohler, Billari, et al., 2002).
The second line of reasoning is the decreasing biological ability to conceive with age, especially for women after the age of 30 (Menken et al., 1986) and to a lesser extent men
(Eisenberg & Meldrum, 2017). For women, decreasing oocyte quality and uterine aging are key indicators, with both early age at menarche and early age at menopause related to diminished ovarian functioning with age, accelerated ovarian aging for women with early menarche (Kok et al., 2003) and diminished ovarian reserve for women with early menarche (Weghofer et al., 2013). Although early age at menarche might be beneficial for early fertility (Guldbrandsen et al., 2014), it may also have negative effects (Weghofer et al., 2013). Earlier menopause is related to higher levels of childlessness, having only one child, a large interval between the first and second child and miscarriages (Kok et al., 2003). We therefore expect that both early menarche and early menopause will have a stronger influence for women who postpone childbearing attempts. Since we do not have a direct measure of attempts to conceive, we use age at marriage as a proxy for age at first attempt to have a child, since the two are closely related in the US in the period of our study (Rindfuss, John, & John, 1983). For PCOS and endometriosis PGSs, the expectations are less clear. Several studies suggest that the natural decrease in the number of oocytes with age is compensated in women with PCOS due to their higher number of oocytes (Mellembakken et al., 2011) and higher levels of serum anti-Müllerian hormone levels up to higher ages (Piltonen et al., 2005). This results in higher ovarian volume and more regular menstrual cycles among older women with PCOS, and less frequently to early ovarian aging. We would thus expect that for older women, PCOS has a weaker effect on childlessness. For endometriosis, the impact on infertility has been found to be stronger for women under the age of 35 (Prescott et al., 2016), thereby suggesting an interaction effect with age.
For men, core factors are increasing sperm defects with age, including decreasing semen volume, total sperm count, sperm motility and morphologically normal sperm (Eisenberg & Meldrum, 2017). The relationship between age and sperm concentration is less clear (Eisenberg & Meldrum, 2017) thus leading us to hypothesize that there is possibly no relationship between azoospermia and oligozoospermia later in life. Low sperm count, however, can be compensated by higher motility and normal morphology (Wainer et al., 2004), which leads us to expect that for younger men, oligozoospermia can be compensated by sperm motility and morphology. Due to naturally decreasing motility and morphology this compensating mechanism will no longer be at work, resulting in the expectation that genes for oligozoospermia will have a stronger effect on childlessness as men get older.
5.2.4 Association between socio-demographic factors and genetic risk scores
In addition to examining the interaction between genes and environment, we investigate correlations between the PGSs and socio-environmental factors. A recent study examined gene environment correlations in a broad range of PGSs related to child behavior and environmental parental characteristics and found widespread covariation between the two (Krapohl et al., 2017). This correlation indicates that the genetic risk scores may partly overlap with environmental factors, which should be taken into account when interpreting genetic risk scores. There is reason to expect gene-environment correlations for our PGSs. If factors that influence reproduction are heritable (e.g., education, age at marriage), GWASs for AFB and NEB will also pick up these factors. Research from the UK and the US finds that about
menarche might be beneficial for early fertility (Guldbrandsen et al., 2014), but on the other hand it is found to have a negative effect on fertility (Weghofer, Kim, Barad, & Gleicher, 2013). Age at menopause (i.e., last menstruation) as well as very early menopause (before the age of 40) is often clustered in families, and heritability studies estimate a genetic component of 30 to 85% (Snieder et al., 1998).
For men, sperm defects most often cause infertility (Blundell, 2007), which we hope to capture with the PGSs for TDS and oligozoospermia or azoospermia. Some male infertility factors are associated with chromosomal abnormalities. However, familial aggregation of male infertility, in the absence of specific genetic aberrations, hint towards multifactorial inheritance. There are comparatively fewer male infertility-related GWA studies and discoveries, many studies characterized by very small samples (Dalgaard et al., 2012). Azoospermia, which is a zero sperm count, effects around 1% of the male population, and is the reason for infertility in 10-15% of men (Willott, 1982). Oligozoospermia is a sperm count of less than 15 or sometimes 20 million per milliliter in the semen (Cooper et al., 2010). TDS is a group of diseases related to male reproductive health, namely infertility due to poor sperm quality, testicular cancer, undescended testes (cryptorchidism) and hypospadias (abnormality where the urinary opening is not at the head of the penis) (Skakkebæk, 2003). Familial clustering of TDS has been shown (Joffe, 2007).
Reproductive behavioral scores. Medical research has often only examined
infertility-related outcomes on smaller sex-specific medical samples. A recent GWAS of reproductive behavior of both sexes isolated genetic markers related to the age at first birth (AFB) and number of children ever born (NEB) (Barban et al., 2016). These PGSs are viewed as behavioral since they likely measure not only biological and genetic components, but also psychological and choice aspects (e.g., personality, intentions) and socio-environmental contexts (e.g., childcare, work-family reconciliation) (Tropf et al., 2017). These GWASs, however, measure very concrete reproductive outcomes, and the identified genetic loci are related to markers found for endometriosis, PCOS, sperm defects, menarche, menopause and level of education.
5.2.3 Interplay between socio-demographic factors and genetic risk scores
We adopt two lines of reasoning to understand the interplay between socio-demographic and genetic factors. First, we anticipate a birth cohort effect based on studies that demonstrate changes in heritability, or in other words, that there are differences in the relationship between genes and reproductive outcomes over time (Tropf et al., 2017). In the US (Briley et al., 2015), the UK (Tropf, Barban, et al., 2015) and Denmark (Kohler, Rodgers, et al., 2002) findings indicate that heritability of reproductive-related outcomes was higher around the second demographic transition, for individuals born between 1935 and 1950, than in periods before and after. The suggested mechanism is that in times where individual freedom is more pervasive, genetic propensities and thus heritability should be higher due to the rise in realization of individual genetic drivers (Kohler, Billari, et al., 2002).
The second line of reasoning is the decreasing biological ability to conceive with age, especially for women after the age of 30 (Menken et al., 1986) and to a lesser extent men
(Eisenberg & Meldrum, 2017). For women, decreasing oocyte quality and uterine aging are key indicators, with both early age at menarche and early age at menopause related to diminished ovarian functioning with age, accelerated ovarian aging for women with early menarche (Kok et al., 2003) and diminished ovarian reserve for women with early menarche (Weghofer et al., 2013). Although early age at menarche might be beneficial for early fertility (Guldbrandsen et al., 2014), it may also have negative effects (Weghofer et al., 2013). Earlier menopause is related to higher levels of childlessness, having only one child, a large interval between the first and second child and miscarriages (Kok et al., 2003). We therefore expect that both early menarche and early menopause will have a stronger influence for women who postpone childbearing attempts. Since we do not have a direct measure of attempts to conceive, we use age at marriage as a proxy for age at first attempt to have a child, since the two are closely related in the US in the period of our study (Rindfuss, John, & John, 1983). For PCOS and endometriosis PGSs, the expectations are less clear. Several studies suggest that the natural decrease in the number of oocytes with age is compensated in women with PCOS due to their higher number of oocytes (Mellembakken et al., 2011) and higher levels of serum anti-Müllerian hormone levels up to higher ages (Piltonen et al., 2005). This results in higher ovarian volume and more regular menstrual cycles among older women with PCOS, and less frequently to early ovarian aging. We would thus expect that for older women, PCOS has a weaker effect on childlessness. For endometriosis, the impact on infertility has been found to be stronger for women under the age of 35 (Prescott et al., 2016), thereby suggesting an interaction effect with age.
For men, core factors are increasing sperm defects with age, including decreasing semen volume, total sperm count, sperm motility and morphologically normal sperm (Eisenberg & Meldrum, 2017). The relationship between age and sperm concentration is less clear (Eisenberg & Meldrum, 2017) thus leading us to hypothesize that there is possibly no relationship between azoospermia and oligozoospermia later in life. Low sperm count, however, can be compensated by higher motility and normal morphology (Wainer et al., 2004), which leads us to expect that for younger men, oligozoospermia can be compensated by sperm motility and morphology. Due to naturally decreasing motility and morphology this compensating mechanism will no longer be at work, resulting in the expectation that genes for oligozoospermia will have a stronger effect on childlessness as men get older.
5.2.4 Association between socio-demographic factors and genetic risk scores
In addition to examining the interaction between genes and environment, we investigate correlations between the PGSs and socio-environmental factors. A recent study examined gene environment correlations in a broad range of PGSs related to child behavior and environmental parental characteristics and found widespread covariation between the two (Krapohl et al., 2017). This correlation indicates that the genetic risk scores may partly overlap with environmental factors, which should be taken into account when interpreting genetic risk scores. There is reason to expect gene-environment correlations for our PGSs. If factors that influence reproduction are heritable (e.g., education, age at marriage), GWASs for AFB and NEB will also pick up these factors. Research from the UK and the US finds that about
40% of the genetic influence on NEB is shared with genetic influences on age at marriage and AFB (Briley et al., 2017). There is also a shared genetic basis between education and fertility (Barban et al., 2016), suggesting that the causal effect of education on fertility is mainly due to shared genetic and family influences (Nisén et al., 2013; Tropf & Mandemakers, 2017). We therefore examine the extent to which our PGSs for AFB and NEB correlate with education and marriage and if their effect on childlessness is mediated by education and marriage. Finally, we examine the association between the two types of behavioral (AFB, NEB) and biological fecundity-related PGSs to examine the extent to which the more distant pathways can be explained by the more proximal pathways.
5.2.5 Sex differences
As discussed previously, both socio-demographic and genetic factors have been found to have differential impacts on males and females due to both socio-environmental and biological differences. For this reason, we estimate sex-specific models. Higher education and occupational status increase childlessness in women but decrease childlessness probabilities in men, and being married seems to be more important for men (Keizer et al., 2008). Previous medical GWASs of our traits of interest have often only been carried out on sex-specific samples (e.g., endometriosis, PCOS only reported by women; sperm defects only for men). These sex-specific GWAS results are possibly capturing factors relevant to examine across the sexes. A recent study found that different genes influence childlessness in men and women (Verweij et al., 2017). For instance, lower testosterone levels in men are related to a lower sperm count and infertility while hyperandrogenism in women (increased levels of testosterone) results in adverse fertility outcomes (Barbieri, 2000). Levels of these sex hormones are heritable (Sluyter et al., 2000), and are possibly captured in our PGSs. Early female pubertal development generally has a positive effect on fertility (Guldbrandsen et al., 2014), while for men both late and early onset of puberty results in decreased sperm concentration and total sperm count (T. K. Jensen et al., 2016). There is a high genetic correlation between age at menarche in women and age at voice breaking in men (Day et al., 2017), but the dissimilar effect of pubertal aging on fertility could also hint towards sexual dimorphism. On the contrary, Anti-Müllerian hormone is important for both reproductive development in men and follicular growth in women (Piltonen et al., 2005).
5.2.6 Differences by ethnicity
Most GWA studies have been conducted in European-ancestry (i.e., White) populations due to the legacy of logistical and historical factors, but also to reduce spurious associations caused by population stratification (i.e., certain population subgroups have a history of interbreeding and differ in genetic ancestry which varies by allele frequency, patterns of linkage disequilibrium, and the prevalence of a trait). Here, we use the HRS to compare results from the White and Black samples. A previous study found that PGSs for educational attainment, depression and height stemming from European-ancestry GWASs have a considerably lower explanatory power in the Black HRS sample (Ware et al., 2017), indicating that different genetic factors might play a role in back individuals.
5.3 MATERIALS AND METHOD
5.3.1 Data, samples, and genotypingWe use two broadly comparable datasets of the Health and Retirement Survey and the Wisconsin Longitudinal Study in the US, with more detailed information and comparisons of the data described in the Online Appendix section 1.
Health and Retirement Survey (HRS). The HRS is a nationally representative sample of men
and women born between ~1920 and 1960 living in the US. This survey started in 1992 with a sample of men and women aged 51 to 61 and their partners, interviewed every two years. Extra cohorts have been added to create a representative sample of Americans over 50 years of age (Sonnega et al., 2014), consisting of over 27,000 respondents in 2010 (Health and Retirement Study, 2017). Between 2006 and 2012, the HRS genotyped 20,000 respondents with data from 15,445 currently available. We selected only the White Non-Hispanic sample (N=10,686) for reasons described previously with separate analyses conducted on the African American Non-Hispanic sample (N=2,433) (we removed the Hispanic sample and people with other ethnicities).
Wisconsin Longitudinal Study (WLS). The WLS is a random sample of one third of all men
and women who graduated from Wisconsin high schools in 1957 (N=10,317 graduates), and one of their siblings (N=8,734 siblings). It is broadly representative of White, non-Hispanic Americans who at least finished high school (Herd, Carr, & Roan, 2014). Respondents filled in questionnaires across six waves (1957, 1964, 1975, 1993, 2004, 2011). Between 2007 and 2011, 9,012 of the WLS respondents were genotyped. We select only those individuals who provided information about their number of children after they finished their reproductive period (age 45 for women or 50 for men), resulting in 8,284 individuals.
5.3.2 Measurements
Childlessness is measured from a direct question regarding the number of biological children
after reaching the end of their reproductive period (see Online Appendix 2 for details on all variables included).
Birth year of respondent is the birthdate reported in the first non-missing wave and is
standardized ((value-mean)/SD) for ease of comparison.
Years of education is the number of years of education and is calculated based of the
highest degree (asked at least once after the age of 30) and is also standardized.
Occupational field is measured in the HRS by job previous to their current occupation
distinguishing between ‘professionals’, ‘managers’, ‘clerks’, ‘sales’, ‘mechanics/production’, ‘services’, ‘operators’, ‘farming’, and ‘army’. In the WLS, it is measured by the first job they had after completing the highest level of schooling, distinguishing between ‘professional/ technical’, ‘administrators/managers’, ‘sales’, ‘clerks’, ‘manufacturing/construction’, ‘service’, ‘farming’., and ‘no first job’. In both datasets clerks were used as reference groups.
Age at first marriage is measured in both datasets using information from the total number
of marriages at each wave, using the answer at the last interview. It is dichotomized into never and ever married and for those who had been ever married, the age of their first
40% of the genetic influence on NEB is shared with genetic influences on age at marriage and AFB (Briley et al., 2017). There is also a shared genetic basis between education and fertility (Barban et al., 2016), suggesting that the causal effect of education on fertility is mainly due to shared genetic and family influences (Nisén et al., 2013; Tropf & Mandemakers, 2017). We therefore examine the extent to which our PGSs for AFB and NEB correlate with education and marriage and if their effect on childlessness is mediated by education and marriage. Finally, we examine the association between the two types of behavioral (AFB, NEB) and biological fecundity-related PGSs to examine the extent to which the more distant pathways can be explained by the more proximal pathways.
5.2.5 Sex differences
As discussed previously, both socio-demographic and genetic factors have been found to have differential impacts on males and females due to both socio-environmental and biological differences. For this reason, we estimate sex-specific models. Higher education and occupational status increase childlessness in women but decrease childlessness probabilities in men, and being married seems to be more important for men (Keizer et al., 2008). Previous medical GWASs of our traits of interest have often only been carried out on sex-specific samples (e.g., endometriosis, PCOS only reported by women; sperm defects only for men). These sex-specific GWAS results are possibly capturing factors relevant to examine across the sexes. A recent study found that different genes influence childlessness in men and women (Verweij et al., 2017). For instance, lower testosterone levels in men are related to a lower sperm count and infertility while hyperandrogenism in women (increased levels of testosterone) results in adverse fertility outcomes (Barbieri, 2000). Levels of these sex hormones are heritable (Sluyter et al., 2000), and are possibly captured in our PGSs. Early female pubertal development generally has a positive effect on fertility (Guldbrandsen et al., 2014), while for men both late and early onset of puberty results in decreased sperm concentration and total sperm count (T. K. Jensen et al., 2016). There is a high genetic correlation between age at menarche in women and age at voice breaking in men (Day et al., 2017), but the dissimilar effect of pubertal aging on fertility could also hint towards sexual dimorphism. On the contrary, Anti-Müllerian hormone is important for both reproductive development in men and follicular growth in women (Piltonen et al., 2005).
5.2.6 Differences by ethnicity
Most GWA studies have been conducted in European-ancestry (i.e., White) populations due to the legacy of logistical and historical factors, but also to reduce spurious associations caused by population stratification (i.e., certain population subgroups have a history of interbreeding and differ in genetic ancestry which varies by allele frequency, patterns of linkage disequilibrium, and the prevalence of a trait). Here, we use the HRS to compare results from the White and Black samples. A previous study found that PGSs for educational attainment, depression and height stemming from European-ancestry GWASs have a considerably lower explanatory power in the Black HRS sample (Ware et al., 2017), indicating that different genetic factors might play a role in back individuals.
5.3 MATERIALS AND METHOD
5.3.1 Data, samples, and genotypingWe use two broadly comparable datasets of the Health and Retirement Survey and the Wisconsin Longitudinal Study in the US, with more detailed information and comparisons of the data described in the Online Appendix section 1.
Health and Retirement Survey (HRS). The HRS is a nationally representative sample of men
and women born between ~1920 and 1960 living in the US. This survey started in 1992 with a sample of men and women aged 51 to 61 and their partners, interviewed every two years. Extra cohorts have been added to create a representative sample of Americans over 50 years of age (Sonnega et al., 2014), consisting of over 27,000 respondents in 2010 (Health and Retirement Study, 2017). Between 2006 and 2012, the HRS genotyped 20,000 respondents with data from 15,445 currently available. We selected only the White Non-Hispanic sample (N=10,686) for reasons described previously with separate analyses conducted on the African American Non-Hispanic sample (N=2,433) (we removed the Hispanic sample and people with other ethnicities).
Wisconsin Longitudinal Study (WLS). The WLS is a random sample of one third of all men
and women who graduated from Wisconsin high schools in 1957 (N=10,317 graduates), and one of their siblings (N=8,734 siblings). It is broadly representative of White, non-Hispanic Americans who at least finished high school (Herd, Carr, & Roan, 2014). Respondents filled in questionnaires across six waves (1957, 1964, 1975, 1993, 2004, 2011). Between 2007 and 2011, 9,012 of the WLS respondents were genotyped. We select only those individuals who provided information about their number of children after they finished their reproductive period (age 45 for women or 50 for men), resulting in 8,284 individuals.
5.3.2 Measurements
Childlessness is measured from a direct question regarding the number of biological children
after reaching the end of their reproductive period (see Online Appendix 2 for details on all variables included).
Birth year of respondent is the birthdate reported in the first non-missing wave and is
standardized ((value-mean)/SD) for ease of comparison.
Years of education is the number of years of education and is calculated based of the
highest degree (asked at least once after the age of 30) and is also standardized.
Occupational field is measured in the HRS by job previous to their current occupation
distinguishing between ‘professionals’, ‘managers’, ‘clerks’, ‘sales’, ‘mechanics/production’, ‘services’, ‘operators’, ‘farming’, and ‘army’. In the WLS, it is measured by the first job they had after completing the highest level of schooling, distinguishing between ‘professional/ technical’, ‘administrators/managers’, ‘sales’, ‘clerks’, ‘manufacturing/construction’, ‘service’, ‘farming’., and ‘no first job’. In both datasets clerks were used as reference groups.
Age at first marriage is measured in both datasets using information from the total number
of marriages at each wave, using the answer at the last interview. It is dichotomized into never and ever married and for those who had been ever married, the age of their first
marriage, categorized into ‘before 21’, ‘21-25’, ‘26-30’, ‘31-35’, ‘36-40’, and ‘older than 41’ years. In this time period most childbearing occurred within marriage, from 98% (1940 to 1960), to 94% in 1970, 90% in 1980 and 80% in 1990 (Ventura & Bachrach, 2000).
Religion. In the HRS respondents were asked their religious preference at each wave:
‘Protestant’, ‘Roman Catholic’, ‘Jewish’, ‘something else’, or ‘non-religious’. The answer from the first wave with non-missing information was used. In several waves of the WLS, respondents were asked about their current religious preference and could choose between 76 religions, which we collapsed into Roman Catholic, Protestant, other, and not religious. The answer from the first wave with non-missing information was used.
Ethnicity. In the WLS we used self-reported ethnicity, removing non-White respondents.
In the HRS, respondents were asked: “Do you consider yourself primarily: ‘White or Caucasian’, ‘Black or African American’, ‘American Indian’, or ‘Asian’?”. Respondents ware also asked if they identified as Hispanic, and the Hispanic respondents were removed from the sample. Since the HRS oversampled African Americans we were able to create a White and Black sample.
5.3.3 GWASs used to create PGSs
Single nucleotide polymorphisms (SNPs) and their summary statistics for NEB and AFB, which were obtained from a recent GWAS that used 251,151 European ancestry individuals for AFB and 343,072 individuals for NEB (Barban et al., 2016). For endometriosis a GWAS of 3,194 surgically confirmed endometriosis cases (of which 1,364 moderate-severe) and 7,060 controls from Australia and the United Kingdom was used (Painter et al., 2011). The PCOS GWAS consisted of 984 PCOS cases and 2,946 controls, all of European ancestry (Hayes et al., 2015). For age at menarche, defined by age at first menstrual period, we used the GWAS of 329,345 women from European ancestry (Day et al., 2017). For age at menopause, defined as the age at which a woman had her last menstrual period, data from the GWAS that included 69,360 women of European ancestry were used (Day et al., 2015). Azoospermia and
oligozoospremia data stem from a GWAS of 80 controls, 52 oligozoospermia cases and 40
azoospermia cases, including white individuals primarily of northern European descent (Aston & Carrell, 2009). For TDS results were used of a GWAS, that included 488 cases and 439 controls from Denmark (Dalgaard et al., 2012). Of these cases 107 were infertile with sperm count below 15 million per milliliter (ml) in the semen and testis volume below 15 ml, 212 with testicular germ cell tumors (TGCC), 138 with cryptorchidism and 31 with hypospadias.
5.3.4 Statistical analyses
To examine the impact of the genetic factors on childlessness we created separate PGSs, using GWAS summary statistics by calculating the sum of all risk alleles, weighted by their reported effect sizes. A PGS thus can be seen as the summary measure of the genetic risk for a trait (Wray, Goddard, & Visscher, 2007). For more detail on all methods, see the Online Appendix section 3.
We apply logistic regression models, adding variables over four steps: (1) PGSs for behavioral genetic reproductive outcomes (AFB, NEB) with the first 20 genetic principal
components (PCs), (2) PGSs for biological fecundity-related genetic outcomes (including PCs); (3) socio-demographic factors; and, (4) all variables. To compare the explanatory power of the genetic and socio-environmental factors, we compare odds ratios (with both PGS and continuous variables standardized) and use McFadden’s pseudo R2.
Since siblings are included in the WLS, we run multilevel models on respondents nested within households to adjust for non-independence (Snijders & Bosker, 2012). To estimate sex differences we used the HRS and WLS samples on both men and women, for which we apply multilevel models (siblings in the WLS and partners in the HRS sample) including interactions with sex. The interaction between the genetic predispositions and postponement of childbearing were examined by including all PGSs by age at first marriage interactions. To test whether genetic influences on fertility became stronger in more recent birth cohorts, we included a PGS (AFB and NEB) by birth year interaction. To properly control for confounding in gene*environment interaction models, Keller (2014) argues that interactions between confounders and genes as well as confounders and environment should be included. For that reason, we include interactions with the first five PCs and with education, birth year and religion.
To assess whether the effect of AFB and NEB PGSs is mediated/confounded by education, marriage or reproduction-related biological traits on the effect of on childlessness, simply comparing coefficients across models with and without confounding factors is not feasible, because unobserved heterogeneity differs between logistic regression models (Mood, 2010). We therefore apply the Karlson-Holm-Breen (KHB) method to equalize the scale of the log-odds across models (Karlson, Holm, & Breen, 2012). With these models we can assess the percentage of confounding due to the PGSs and socio-environmental factors, after we control for the first 20 genetic PCs. We furthermore examine correlations between the AFB and NEB PGSs and education and marriage using Pearson correlation coefficients, as well as the association between AFB and NEB PGSs with the biological traits PGSs (while controlling for the first 20 principal components). In addition we assess the LD-score genetic correlations between the reproductive behavior and the biological traits PGSs (see Online Appendix).
5.3.5 Descriptives
Descriptive statistics of the samples can be found in Supplementary Material Table 1. In the HRS 10.8% of women and 12.5% of men in the HRS remained childless, whereas for the WLS this was 6.6% and 6.3% (these estimates are in line with the levels of childlessness in the USA in these periods, see Supplementary Material Figure 1). In the HRS, around half of the respondents completed high school or less and around 20% of women finished college or above compared to 29% of men. In the WLS, around 58% of women finished high school only compared to 49% of men. Finishing college or more was 25% and 35% for women and men respectively. Only a very small percentage in both samples remained never married (between 3% and 4%). WLS respondents on average had younger ages at marriage than the HRS respondents.
marriage, categorized into ‘before 21’, ‘21-25’, ‘26-30’, ‘31-35’, ‘36-40’, and ‘older than 41’ years. In this time period most childbearing occurred within marriage, from 98% (1940 to 1960), to 94% in 1970, 90% in 1980 and 80% in 1990 (Ventura & Bachrach, 2000).
Religion. In the HRS respondents were asked their religious preference at each wave:
‘Protestant’, ‘Roman Catholic’, ‘Jewish’, ‘something else’, or ‘non-religious’. The answer from the first wave with non-missing information was used. In several waves of the WLS, respondents were asked about their current religious preference and could choose between 76 religions, which we collapsed into Roman Catholic, Protestant, other, and not religious. The answer from the first wave with non-missing information was used.
Ethnicity. In the WLS we used self-reported ethnicity, removing non-White respondents.
In the HRS, respondents were asked: “Do you consider yourself primarily: ‘White or Caucasian’, ‘Black or African American’, ‘American Indian’, or ‘Asian’?”. Respondents ware also asked if they identified as Hispanic, and the Hispanic respondents were removed from the sample. Since the HRS oversampled African Americans we were able to create a White and Black sample.
5.3.3 GWASs used to create PGSs
Single nucleotide polymorphisms (SNPs) and their summary statistics for NEB and AFB, which were obtained from a recent GWAS that used 251,151 European ancestry individuals for AFB and 343,072 individuals for NEB (Barban et al., 2016). For endometriosis a GWAS of 3,194 surgically confirmed endometriosis cases (of which 1,364 moderate-severe) and 7,060 controls from Australia and the United Kingdom was used (Painter et al., 2011). The PCOS GWAS consisted of 984 PCOS cases and 2,946 controls, all of European ancestry (Hayes et al., 2015). For age at menarche, defined by age at first menstrual period, we used the GWAS of 329,345 women from European ancestry (Day et al., 2017). For age at menopause, defined as the age at which a woman had her last menstrual period, data from the GWAS that included 69,360 women of European ancestry were used (Day et al., 2015). Azoospermia and
oligozoospremia data stem from a GWAS of 80 controls, 52 oligozoospermia cases and 40
azoospermia cases, including white individuals primarily of northern European descent (Aston & Carrell, 2009). For TDS results were used of a GWAS, that included 488 cases and 439 controls from Denmark (Dalgaard et al., 2012). Of these cases 107 were infertile with sperm count below 15 million per milliliter (ml) in the semen and testis volume below 15 ml, 212 with testicular germ cell tumors (TGCC), 138 with cryptorchidism and 31 with hypospadias.
5.3.4 Statistical analyses
To examine the impact of the genetic factors on childlessness we created separate PGSs, using GWAS summary statistics by calculating the sum of all risk alleles, weighted by their reported effect sizes. A PGS thus can be seen as the summary measure of the genetic risk for a trait (Wray, Goddard, & Visscher, 2007). For more detail on all methods, see the Online Appendix section 3.
We apply logistic regression models, adding variables over four steps: (1) PGSs for behavioral genetic reproductive outcomes (AFB, NEB) with the first 20 genetic principal
components (PCs), (2) PGSs for biological fecundity-related genetic outcomes (including PCs); (3) socio-demographic factors; and, (4) all variables. To compare the explanatory power of the genetic and socio-environmental factors, we compare odds ratios (with both PGS and continuous variables standardized) and use McFadden’s pseudo R2.
Since siblings are included in the WLS, we run multilevel models on respondents nested within households to adjust for non-independence (Snijders & Bosker, 2012). To estimate sex differences we used the HRS and WLS samples on both men and women, for which we apply multilevel models (siblings in the WLS and partners in the HRS sample) including interactions with sex. The interaction between the genetic predispositions and postponement of childbearing were examined by including all PGSs by age at first marriage interactions. To test whether genetic influences on fertility became stronger in more recent birth cohorts, we included a PGS (AFB and NEB) by birth year interaction. To properly control for confounding in gene*environment interaction models, Keller (2014) argues that interactions between confounders and genes as well as confounders and environment should be included. For that reason, we include interactions with the first five PCs and with education, birth year and religion.
To assess whether the effect of AFB and NEB PGSs is mediated/confounded by education, marriage or reproduction-related biological traits on the effect of on childlessness, simply comparing coefficients across models with and without confounding factors is not feasible, because unobserved heterogeneity differs between logistic regression models (Mood, 2010). We therefore apply the Karlson-Holm-Breen (KHB) method to equalize the scale of the log-odds across models (Karlson, Holm, & Breen, 2012). With these models we can assess the percentage of confounding due to the PGSs and socio-environmental factors, after we control for the first 20 genetic PCs. We furthermore examine correlations between the AFB and NEB PGSs and education and marriage using Pearson correlation coefficients, as well as the association between AFB and NEB PGSs with the biological traits PGSs (while controlling for the first 20 principal components). In addition we assess the LD-score genetic correlations between the reproductive behavior and the biological traits PGSs (see Online Appendix).
5.3.5 Descriptives
Descriptive statistics of the samples can be found in Supplementary Material Table 1. In the HRS 10.8% of women and 12.5% of men in the HRS remained childless, whereas for the WLS this was 6.6% and 6.3% (these estimates are in line with the levels of childlessness in the USA in these periods, see Supplementary Material Figure 1). In the HRS, around half of the respondents completed high school or less and around 20% of women finished college or above compared to 29% of men. In the WLS, around 58% of women finished high school only compared to 49% of men. Finishing college or more was 25% and 35% for women and men respectively. Only a very small percentage in both samples remained never married (between 3% and 4%). WLS respondents on average had younger ages at marriage than the HRS respondents.
5.4 RESULTS
5.4.1 PGSs for reproductive behavior and not biological traits explain childlessness
A main finding is that the PGSs for reproductive behavior (AFB and NEB) are to a small extent related to childlessness, while those related to biological traits do not. PGSs favoring higher NEB decreased the chance of not having children among both sexes, but only in the HRS (model 1 of Tables 1-4). PGSs for later AFB increase childlessness, especially among women (model 1 of Tables 1-4). The correlation between the AFB and NEB PGSs is relatively high, -0.36, -0.39, -0.30 and -0.28 in the female HRS, male HRS, female WLS and male WLS samples respectively (all significant at the 0.05 level). In Supplementary Material Figure 2 to 5 it is shown that when included in the model separately AFB is significant in all 4 samples and NEB in both female samples. However, if we include all socio-demographic variables in our models, the effect sizes of the genetic risk scores decrease or become insignificant (see model 4 in Tables 1-4, we elaborate on this further in the section on gene-environment correlations).
For the PGSs related to biological fecundity we find only small and mixed findings (model 2 of Tables 1-4). For men (in WLS), PGSs related to infertility due to low sperm count increased male childlessness, while the PGS for Azoospermia has an unexpected negative effect on childlessness. This smaller effect of biological fecundity PGSs is likely attributed to the lower-powered GWASs they are based on (see Supplementary Material Table 2). The relationship between the PGSs and childlessness using different p-value cutoffs are graphically displayed in Supplementary Material Figure 2 to 5, showing that in most cases the p-value cutoff of 1 resulted in the highest odds ratios and smallest confidence intervals.
5.4.2 Effect of socio-demographic factors as expected
Individuals from more recent birth cohorts that also postponed or did not marry and were not religious (in the WLS) were more likely to remain childless (see model 3 in Tables 1-4). Among women, those higher educated remained childless more often in contrast to lower childlessness for those who never worked (i.e., no reported first occupation) or were employed in the service sector. Education and occupation did not influence childlessness in men.
5.4.3 Predictive power of PGSs small compared to some socio-demographic factors
The effect sizes for the PGSs were modest: an increase of 1 SD in the AFB PGS increased the odds of remaining childless with 1.27, 1.23, 1.08 and 1.27 in the female HRS, female WLS, male HRS and male WLS respectively. However, in the models in which the socio-environmental factors were included these effects reduced to 1.026, 1.147, 1.105 and 0.977. This is relatively small compared to some socio-environmental factors, such as education, where a 1 SD increase in years of education resulted in an increase in the odds of remaining childless of 1.24, 1.53 in the female HRS and WLS samples respectively. For those who married after age 36, the odds of remaining childless are 4.6, 10.8, 8.8 and 44.8 times
higher than those who wed before the age 21, in the four samples respectively. Examining the McFadden R2, the goodness of fit in the models with only genetic factors is markedly
lower (between 0.01 and 0.02) than models with socio-demographic factors (0.19 and 0.46).
5.4.4 PGS for AFB especially relevant among women who married at higher ages
We find suggestive evidence that PGSs for AFB are particularly influential among women who married at higher ages (Figure 2, Supplementary Material Table 6 and Supplementary Material Table 7). Genes related to AFB do not seem to effect childlessness among women who married before 30, but have a positive effect for those who married after 30. We are only able to detect these effects in the HRS data, where more respondents marry at later ages. The interaction between AFB and age at marriage has a similar direction in the WLS sample (Supplementary Material Figure 6). Although we expected that the PGSs for AFB and NEB would have a stronger influence during the second demographic transition, we did not find these effects.
5.4.5 Gene-environment correlations in the expected direction and mixed results for the genetic correlations
The PGSs related to reproductive behavior (higher AFB, lower NEB) are related to higher education and a higher age at first marriage or never marrying (Table V), which is in line with our expectations. In females, we also see a positive correlation between the AFB PGS with PCOS but an unexpected negative correlation between the AFB PGS with endometriosis (Table VI). PGSs for higher age at menarche and higher age of menopause are related to higher AFB PGS (Table VI). For men, the results are almost all insignificant, it only seems to
Figure 2 | Age at first birth PGSs especially relevant among later married women in the HRS sample
(results from the model in Supplementary Material Table 7)
5.4 RESULTS
5.4.1 PGSs for reproductive behavior and not biological traits explain childlessness
A main finding is that the PGSs for reproductive behavior (AFB and NEB) are to a small extent related to childlessness, while those related to biological traits do not. PGSs favoring higher NEB decreased the chance of not having children among both sexes, but only in the HRS (model 1 of Tables 1-4). PGSs for later AFB increase childlessness, especially among women (model 1 of Tables 1-4). The correlation between the AFB and NEB PGSs is relatively high, -0.36, -0.39, -0.30 and -0.28 in the female HRS, male HRS, female WLS and male WLS samples respectively (all significant at the 0.05 level). In Supplementary Material Figure 2 to 5 it is shown that when included in the model separately AFB is significant in all 4 samples and NEB in both female samples. However, if we include all socio-demographic variables in our models, the effect sizes of the genetic risk scores decrease or become insignificant (see model 4 in Tables 1-4, we elaborate on this further in the section on gene-environment correlations).
For the PGSs related to biological fecundity we find only small and mixed findings (model 2 of Tables 1-4). For men (in WLS), PGSs related to infertility due to low sperm count increased male childlessness, while the PGS for Azoospermia has an unexpected negative effect on childlessness. This smaller effect of biological fecundity PGSs is likely attributed to the lower-powered GWASs they are based on (see Supplementary Material Table 2). The relationship between the PGSs and childlessness using different p-value cutoffs are graphically displayed in Supplementary Material Figure 2 to 5, showing that in most cases the p-value cutoff of 1 resulted in the highest odds ratios and smallest confidence intervals.
5.4.2 Effect of socio-demographic factors as expected
Individuals from more recent birth cohorts that also postponed or did not marry and were not religious (in the WLS) were more likely to remain childless (see model 3 in Tables 1-4). Among women, those higher educated remained childless more often in contrast to lower childlessness for those who never worked (i.e., no reported first occupation) or were employed in the service sector. Education and occupation did not influence childlessness in men.
5.4.3 Predictive power of PGSs small compared to some socio-demographic factors
The effect sizes for the PGSs were modest: an increase of 1 SD in the AFB PGS increased the odds of remaining childless with 1.27, 1.23, 1.08 and 1.27 in the female HRS, female WLS, male HRS and male WLS respectively. However, in the models in which the socio-environmental factors were included these effects reduced to 1.026, 1.147, 1.105 and 0.977. This is relatively small compared to some socio-environmental factors, such as education, where a 1 SD increase in years of education resulted in an increase in the odds of remaining childless of 1.24, 1.53 in the female HRS and WLS samples respectively. For those who married after age 36, the odds of remaining childless are 4.6, 10.8, 8.8 and 44.8 times
higher than those who wed before the age 21, in the four samples respectively. Examining the McFadden R2, the goodness of fit in the models with only genetic factors is markedly
lower (between 0.01 and 0.02) than models with socio-demographic factors (0.19 and 0.46).
5.4.4 PGS for AFB especially relevant among women who married at higher ages
We find suggestive evidence that PGSs for AFB are particularly influential among women who married at higher ages (Figure 2, Supplementary Material Table 6 and Supplementary Material Table 7). Genes related to AFB do not seem to effect childlessness among women who married before 30, but have a positive effect for those who married after 30. We are only able to detect these effects in the HRS data, where more respondents marry at later ages. The interaction between AFB and age at marriage has a similar direction in the WLS sample (Supplementary Material Figure 6). Although we expected that the PGSs for AFB and NEB would have a stronger influence during the second demographic transition, we did not find these effects.
5.4.5 Gene-environment correlations in the expected direction and mixed results for the genetic correlations
The PGSs related to reproductive behavior (higher AFB, lower NEB) are related to higher education and a higher age at first marriage or never marrying (Table V), which is in line with our expectations. In females, we also see a positive correlation between the AFB PGS with PCOS but an unexpected negative correlation between the AFB PGS with endometriosis (Table VI). PGSs for higher age at menarche and higher age of menopause are related to higher AFB PGS (Table VI). For men, the results are almost all insignificant, it only seems to
Figure 2 | Age at first birth PGSs especially relevant among later married women in the HRS sample
(results from the model in Supplementary Material Table 7)
