The interaction between polygenic risk and environmental influences: a direct test of the 3P model of insomnia in adolescents

Abstract

Background:

Stress is a universal phenomenon and one of the most common precipitants of insomnia. However, not everyone develops insomnia after experiencing a stressful life-event. This study aims to test aspects of Spielman’s ‘3P model of insomnia’ (during adolescence) by exploring the extent to which: 1) insomnia symptoms are predicted by polygenic scores; 2) life-events predict insomnia symptoms; 3) the interaction between polygenic scores and life-events contribute to the prediction of insomnia symptoms; 4) gene-environment interaction effects remain after controlling for sex.

Methods:

The sample comprised 4629 twins aged 16 from the Twin Early Development Study who reported on their insomnia symptoms and life-events. Polygenic scores (PGS) for insomnia were calculated. In order to test the main hypothesis of this study (a significant interaction between PGS and negative-life events), we fitted a series of mixed effect regressions.

Results:

The best fit was provided by the model including sex, PGS for insomnia, negative life-events and their interactions (AIC=26158.7). Our results show that the association between insomnia symptoms and negative life-events is stronger for those with a higher genetic risk for insomnia.

Conclusions:

This work sheds light on the complex relationship between genetic and environmental factors implicated for insomnia. This study has tested for the first time the interaction between genetic predisposition (PGS) for insomnia and environmental stressors (negative life-events) in adolescents. This work represents a direct test of components of Spielman’s 3P model for insomnia which is supported by our results.

Introduction

Insomnia is the most prevalent sleep disorder and is among the most common complaints in clinical practice (Morin et al., 2015). Insomnia is characterized by both nocturnal and diurnal symptoms. These symptoms include difficulties initiating or maintaining sleep and impairments of daytime functioning (Morin et al., 2015, Ohayon, 2002). The prevalence of insomnia varies widely depending on the criterion used to define insomnia. For example, one review found that it ranges from 6% (when diagnosed according to the DSM-IV) to around one third of the population (when considering those with at least one symptom of insomnia) (Ohayon, 2002). Insomnia is related to psychiatric disorders such as depression, anxiety, attention deficit hyperactivity disorder (ADHD), schizophrenia and bipolar disorder among others (Alvaro et al., 2013, Gregory et al., 2017, Lewis et al., 2019, Robertson et al., 2019, Madrid-Valero et al., 2020). Insomnia and poor sleep quality are also included as diagnostic criteria for some of these disorders (American Psychiatric Association, 2013). Despite the importance of insomnia at all stages of the lifespan, insomnia during adolescence is under-recognised, under-diagnosed and under-treated even though it is a highly prevalent condition (de Zambotti et al., 2018).

In order to reduce the impact of insomnia, it is of interest to understand the etiology of this disorder. An understanding of its etiology will enable us to identify who may be at risk of insomnia, and who may benefit from preventative interventions. Twin studies have shown that insomnia is substantially influenced by genetic factors, accounting for between 39-40% of the variability in general population samples (Madrid-Valero et al., 2021b, Barclay et al., 2021). This has also been found in Genome-Wide Association Studies (GWAS), although the SNP-based heritability estimate is lower than twin heritability (Watanabe et al., 2022).

A moderate heritability for insomnia means that there is also substantial environmental influence. Investigation of environmental factors, in combination with knowledge of underlying genetic factors, will increase our predictive ability to identify individuals at risk for insomnia. There are multiple risk factors for insomnia such as increased age, female sex and genetic factors (Morin et al., 2015, Madrid-Valero et al., 2017). However, adult literature also shows numerous social determinants of insomnia-related phenotypes. For example, sleep disturbances are more common in individuals who are unemployed, on low income, or with low educational attainment (Ford and Kamerow, 1989, Ohayon, 2002, Grandner et al., 2010). Of note, the most frequently reported precipitating factors for insomnia are stressful life events related to health, family relationships, work or school, and which are perceived as negative events (Bastien et al., 2004, Ellis et al., 2021). The literature regarding how genes and environments interact is scarce. In twin modelling, it has been shown that genetic factors that may influence sleep quality may be shared with those that influence exposure to stressful life events, reflecting gene-environment correlation (Barclay et al., 2011).

The possibility of gene-environment interaction (GxE) aligns with Spielman’s ‘3P model of insomnia’ (Spielman et al., 1987). The model suggests that insomnia manifests when a predisposing factor (e.g. a genetic factor) is triggered by a precipitating factor (such as an environmental stressor). A recent study in support of this idea has demonstrated that those with high sleep reactivity (a tendency toward cognitive or physiological hyperarousal at night or in response to stress) have a stronger relationship between weekly stressful life events and insomnia (Walker et al., 2022). This reinforces the idea that a predisposing factor could exacerbate the effects of a precipitating factor.

Stress is a universal phenomenon and one of the most common precipitants of insomnia (Morin, 2022). However, not everyone develops insomnia following a stressful life-event (Morin, 2022). Therefore, making use of PGS, this study aims to test the diathesis-stress model, where stressful life events may activate a vulnerability, transforming the potential for insomnia into actual symptoms. This study explores the extent to which: 1) insomnia symptoms are predicted by the PGS derived from a reference GWAS; 2) life-events predict insomnia symptoms; 3) the interaction between PGS and life-events contribute to the prediction of insomnia symptoms; 4) GxE effects remain after controlling for sex since there are large differences between men and women when it comes to the prevalence of insomnia (Zhang and Wing, 2006).

Method

Participants

The sample comprised 10,346 twins from the Twins Early Development Study (TEDS). TEDS is a community-based, longitudinal study of twins born in England and Wales between 1994 and 1996 (Haworth et al., 2013, Rimfeld et al., 2019). This sample is reasonably representative of the general population (Kovas et al., 2007). Zygosity was assessed using either DNA testing or a questionnaire which has shown an accuracy of over 95% (Price et al., 2000). Ethical approval for TEDS was provided by the King’s College London ethics committee (ref: PNM/09/10-104). Written informed consent was obtained prior to each wave of data collection from parents and from twins themselves from age 16 onwards.

This work focuses on data collected at age 16. In this wave 19,874 families were contacted and invited to participate and 47.3% of these families provided data. Differences between participating and non-participating families have been reported elsewhere (Zavos et al., 2016). Analyses here focused on a study within this wave of data collection called LEAP-1 which includes participants who provided data about insomnia [N=7,908], life-events [N=10,136], and had PGS for insomnia [N=10,346]. Of the LEAP-1 participants, 4634 participants had data for all three variables. The sample comprised both MZ (25%) and DZ (75%) twins and was 54% female. Five participants were excluded due to medical reasons which affected their participation in the study.

Measures

Insomnia was assessed using the Insomnia Severity Index questionnaire (ISI,(Bastien et al., 2001)). This questionnaire is composed of seven items enquiring about symptoms of insomnia over the previous month. Scores vary from 0 to 28 where a higher score represents greater insomnia symptoms (Bastien et al., 2001). Psychometric properties of this questionnaire have shown to be adequate (Morin et al., 2011). Cronbach’s alpha in this sample was .89.

Negative life-events were assessed using a reduced version of the Coddington Life Event Scale (CLES) (Coddington, 1972). This scale is composed of the 20 items most relevant to adolescents. This version has been previously used (Wootton et al., 2017, Shakoor et al., 2016). Participants self-reported whether they had experienced any of the events in the past six months. Additionally, if the event had occurred, participants also informed whether they found the experience: “Very unpleasant”, “Moderately unpleasant”, “Neither unpleasant nor pleasant”, “Moderately pleasant” or “Very pleasant”. In this study we focused on negative life events (those rated as “very unpleasant” or moderately unpleasant”). This variable was calculated by summing ‘Moderately unpleasant’ (1) and ‘Very unpleasant’ (2) responses. The responses ‘No’ and ‘Neither unpleasant nor pleasant’ were coded as 0. All questions are available elsewhere (Supplementary Table 4; (Peel et al., 2023)).

Polygenic Scores (PGS)

PGS for individuals are the sum of the number of trait-associated allelic genetic variants weighted by their association effect size based on summary statistics. In this study we used summary statistics from the GWA study by Jansen et al., (2019) which used a sample of more than 1.3 million participants (cases and controls for insomnia, see Jansen et al., (2019) supplementary Table 1 and 2 for a full description). The SNP-based heritability was estimated at 7%—which is the upper limit of the predictive value of this PGS for insomnia using this GWAS. PGS were calculated using LDpred which re-weights the variants effect sizes using a prior on their effect size (based on the heritability and assumed fraction of causal markers that influence the trait), adjusting for the linkage disequilibrium in the sample (Vilhjálmsson et al., 2015). Details about genotyping can be found elsewhere (https://www.teds.ac.uk/datadictionary/studies/dna.htm). The TEDS genotypic sample was composed of data from both the Affymetrix and OEE platforms, which were combined, and quality control procedures were applied (these methods are described in detail in S1 Methods, Supplementary Methods, Selzam et al.,(2018). The initial sample comprised 11869 samples. From this initial sample, 1523 samples were removed due to possible non-European ancestry, heterozygosity anomalies, genotype call rate <0.98. PGS were statistically adjusted for the first ten principal components, chip and plate using the regression method, and were z-standardized (mean=0, SD=1) to avoid potential effects due to population stratification and genotyping. Three thresholds for PGS were available (p=0.01, p=0.3 and p=1), which represent an assumption on the fraction of genetic markers that are causally influencing the trait. For example, p=1 represents the assumption that all genetic variants contribute to trait development. Analyses were performed using the p=1 threshold since it was the threshold with the highest predictive value (p=0.001 and r2=0.002) for insomnia symptoms. The p=0.01 threshold was not significantly associated with symptoms of insomnia.

Statistical analyses

This study was preregistered on the Centre for Open Science Website (https://osf.io/qsd2a). In order to test the main hypothesis of this study we fitted a series of mixed effect regressions to control for the relatedness of the sample (i.e. including family as a random effect). The analyses were performed using the lme4 package in R (Bates, 2015). First, we fitted the full model (life-events, insomnia PGS and the interaction terms) including sex as a covariate. The interactions between the moderator and predictors (i.e. insomnia PGS x sex and life-events x sex) were also included as is standard practice in gene-by-environment interaction (GxE) studies (Keller, 2014). Then, reduced nested models were compared in order to check if one or more parameters could be dropped without a decrease in model fitting. The fit of the models was compared using the Akaike Information Criterion (AIC). We also reported the p-value for the comparison among models. However, the use of p-values using mixed models is discouraged (and intentionally not reported by lme4) and the method to obtain these values has certain limitations (Luke, 2017). As recommended, we are not focusing exclusively on p-values when using mixed regression models (Luke, 2017). Negative life events and symptoms of insomnia showed positive skewness, within an acceptable range (1.71 and 1.39 for symptoms of insomnia and negative life-events respectively). All continuous variables were mean-centred in the regression models in order to avoid possible multicollinearity.

Results

Descriptive statistics are displayed in Table 1. Females reported higher levels of insomnia symptoms than males (=4.25 vs =3.28; p<.05). Regarding negative life events males and females reported a similar score — albeit significant differences were found due to the large sample size (=0.89 vs =0.81, p = 0.009), similar values were also found for PGS for insomnia (=0.01 vs =0, p = 0.769). A substantial proportion of the sample (N=761; 16.4%) reported a score of more than 8 points which has been previously suggested as a cut-off to detect individuals at risk of insomnia (Morin et al., 2011). Regarding negative life-events, scores ranged from 0 to 5 and 21.8% of the sample (N=1011) reported a score ≥2.

Table 1:

Descriptive statistics

	Male	Female	Total
Total, N (%)	2125 (46)	2504 (54)	4629 (100)
Age (SD)	16.15 (0.68)	16.14 (0.68)	16.15 (0.68)
Mean ISI (SD)	3.28 (3.9)	4.25(4.5)	3.81 (4.2)
Mean negative life-events (SD)	0.81 (1.0)	0.89 (1.1)	0.85 (1.1)
Mean PGS insomnia (SD)	0 (1.0)	0.01 (1.0)	0 (1.0)

ISI: Insomnia Severity Index; PGS: Polygenic scores

Note: higher scores represent more symptoms of insomnia, negative life-events or a higher genetic predisposition.

The full model and reduced models are presented in Table 2. Overall, the best fit (as indicated by the lowest AIC) was provided by the full model (AIC=26158.7) (Table 2). This model shows that an increase in negative life-events is associated with an increase in insomnia symptoms and this increase is more pronounced for those with higher genetic vulnerability for insomnia. For example, we found differences in ISI scores (depending on the genetic vulnerability for insomnia [i.e. PGS for insomnia]) of around 0.5 points for participants reporting 2 negative life-events and differences of 1 ISI point or higher for those reporting 4 or more negative life-events. All models are compared against the full model (model I) except model III which is compared to model II to provide information about the interaction term without covariates. For the sake of clarity, comparisons of reduced models are here presented first. First of all, reduced models removing PGS of insomnia (Model V; AIC= 26161.8; p=0.028) and negative life-events (Model VI; AIC= 26383.0; p=<0.001) showed a lower fit, and fitted significantly worse, as compared to the full model. Secondly, the interaction between PGS for insomnia and negative life-events was a significant predictor of symptoms of insomnia. Models removing the interaction but retaining sex (Model IV; AIC=26160.0; p=0.071, compared to model I) and excluding sex (Model III; AIC=26226.3; P=0.039, compared to model II) showed a poorer fit. Note that the p-value for the comparison between Model I and Model IV is slightly above .05 but the AIC value is lower for model I (AIC=26158.7) and hence this is focused on as our model of best fit. Furthermore, the model including the interaction term (Model I) explained 0.10% more of the variance as compared to the model without the interaction term (Model IV). Sex was a significant predictor of insomnia symptoms since dropping this variable resulted in a significant worse fit (AIC=26224.1; p<0.001) (Table 2). We performed a sensitivity analysis using the p=0.3 threshold and the same pattern of results was found as compared to the p=1 threshold. Figure 1 clearly shows the effect of the interaction between PGS for insomnia and negative life-events. Results are displayed in this figure for high genetic predisposition (+1SD in PGS for insomnia), medium genetic predisposition (mean value) and low genetic predisposition (−1SD in PGS for insomnia). In this figure differences of around one point in insomnia symptoms (ISI) are found, depending on the genetic predisposition group for those with high levels of negative life-events. There are no differences between ISI scores based on genetic predisposition groups for those reporting low levels of negative life-events.

Table 2:

Models predicting insomnia symptoms

Model (model for comparison)		Estimate	SE	df	t-value	Standardized estimate	AIC	R2	Model comparison (p- value)
I							26158.7	0.064
	Negative life-events	1.025	0.076	4612	13.477	0.255
	PGS insomnia	0.085	0.082	4573	1.040	0.020
	Negative life-events* PGS insomnia	0.104	0.058	4621	1.806	0.026
	Sex	−0.926	0.120	4589	−7.715	−0.109
	Sex* Negative life-events	−0.397	0.115	4514	−3.454	−0.064
	Sex* PGS insomnia	0.119	0.120	4556	0.992	0.019
II (I)							26224.1	0.050	<0.001
	Negative life-events	0.873	0.058	4413	14.96	0.217
	PGS insomnia	0.144	0.062	4197	2.343	0.034
	Negative life-events* PGS insomnia	0.119	0.057	4617	2.056	0.030
III (II)							26226.3	0.049	0.039
	Negative life-events	0.878	0.058	4420	15.066	0.218
	PGS insomnia	0.143	0.062	4200	2.313	0.034
IV (I)							26160.0	0.063	0.071
	Negative life-events	1.034	0.076	4612	13.621	0.257
	PGS insomnia	0.087	0.082	4574	1.066	0.021
	Sex	−0.093	0.120	4589	−7.746	−0.110
	Sex* Negative life-events	−0.406	0.115	4516	−3.540	−0.066
	Sex* PGS insomnia	0.111	0.120	4553	0.923	0.017
V (I)							26161.8	0.063	0.028
	Negative life-events	1.038	0.076	4613	13.684	0.258
	Sex	−0.931	0.120	4588	−7.754	−0.110
	Negative life-events*Sex	−0.407	0.115	4513	−3.545	−0.066
VI (I)							26383.0	0.016	<0.001
	PGS insomnia	0.145	0.084	4585	1.731	0.034
	Sex	−1.006	0.123	4577	−8.187	−0.118
	PGS insomnia *Sex	0.064	0.123	4536	0.519	0.010

Note: higher scores represent more symptoms of insomnia, negative life-events or a higher genetic predisposition. Continuous variables were mean-centered.

AIC: Akaike information criterion; ISI: Insomnia Severity Index; PGS: Polygenic scores; SE: standard error.

Model in bold represents the model with the lowest AIC value and which is selected as our model of best fit.

R² expressed as a proportion.

Figure 1: Regression model for the interaction between standardized polygenic scores for insomnia and life-events.

Three groups for polygenic scores are presented: 1) high genetic predisposition (+1SD); 2) medium genetic predisposition (mean value) and; 3) low genetic predisposition (−1SD).

Discussion

This study has tested for the first time the interaction between genetic predisposition (PGS) and environmental stressors (negative life-events) for insomnia symptoms in adolescents. This work represents a direct test of key components of the 3P model of insomnia (Spielman et al., 1987). Results from this study support this model. Here we find that negative life-events may have a bigger impact on insomnia symptoms in those with a higher as compared to lower genetic predisposition for insomnia. Conversely, the intensity of the precipitating factor (environmental stress) may need to be higher for those with a lower genetic risk for insomnia (predisposing/protecting factor) in order to trigger the disorder.

Life-events and PGS for insomnia are both significant predictors of insomnia symptoms. Our results confirm the robust association between negative life-events and insomnia symptoms, those who report stressful life-events report higher scores for insomnia symptoms (Ellis et al., 2021, Short et al., 2022). Furthermore, our study has also revealed that PGS calculated in adults (Jansen et al., 2019) is a significant predictor of insomnia in adolescence — although, the proportion of the variance explained by PGS of insomnia is still small. PGS has its predictive limit value in heritability (Choi et al., 2020). In a previous study, using a twin design, we confirmed that insomnia is substantially influenced by genetic factors in this sample (41%) (Madrid-Valero et al., 2020). However, there is still a gap between estimates from twin studies and SNP-based heritability which has been estimated at 7-8% (Watanabe et al., 2022, Jansen et al., 2019). Sex was also a significant predictor of insomnia symptoms, but similar mean PGS values were found for men and women. This chimes well with the previous literature: women usually report higher levels of insomnia (Zhang and Wing, 2006) but the magnitude of genetic and environmental influences on insomnia symptoms seems to be similar for both sexes (Madrid-Valero et al., 2021b). This has also been confirmed at a molecular level (Jansen et al., 2019).

This work also sheds light on the complex relationship between genetic and environmental factors implicated in insomnia during a critical period of the lifespan. As stated by Morin in a recent editorial, it is of interest to elucidate the mechanisms underlying the vulnerability to insomnia (Morin, 2022). This study advances our knowledge about the etiology of insomnia by showing that genetic vulnerability acts as a threshold (with ample inter-individual variation) that could be exceeded in the event of environmental stressors such as negative life-events. Genetic predisposition for insomnia could be related to a wide variety of biological and psychological factors such as hyperarousal (Riemann et al., 2010, Schneider et al., 2019a), neurotransmitters implicated in sleep-wake regulation (Morin et al., 2015), and a tendency towards worry or excessively ruminative (Clancy et al., 2020) and also dysfunctional beliefs about sleep (Schneider et al., 2019b). Furthermore, our results also converge with the study by Walker et al., (2022) which demonstrated a stronger relationship between greater stressful life-events and greater insomnia symptoms for those with high ‘sleep reactivity’ (predisposition to experience insomnia following acute stressors). Interestingly, PGS for insomnia symptoms became non-significant once we added negative life-events and the interaction between negative life-events and PGS for insomnia. This reinforces the idea that genetic influences cannot be understood without taking environmental influences into account — genetic influences need the environment to be expressed (at least in the context of insomnia). Insomnia is a complex disorder that is influenced by multiple factors, both genetic and environmental, and there is a complex inter-relationship among them.

Results from this study have implications for both clinical and research practice. First, we have shown that PGS scores could be a useful tool to identify people at risk of suffering from insomnia during adolescence. These findings could eventually pave the way to more accurate and tailored prevention programs. Furthermore, we have also shown that this genetic predisposition interacts with environmental stressors (i.e. negative life-events) so in the future, it might be fruitful for preventive programs to focus on participants with a high genetic predisposition and especially following negative life-events. A high heritability is often misunderstood as genetic determinism. As we previously demonstrated these conceptions could potentially influence treatment outcomes since patients might interpret this information as suggesting that certain treatments are not valuable (Madrid-Valero et al., 2021a). In fact, efforts should be aimed at supporting participants at high risk of experiencing insomnia. The role of protective factors such as coping skills, social support and resilience should be further investigated and tested (especially) among people at high risk. Finally, sleep reactivity can also be reduced by using cognitive behavioural therapies (Chan et al., 2021).

Strengths and limitations

This study has several advantages such us the use of a large sample and a powerful design. Furthermore, this study represents a direct test of a theory-driven research question. However, it also has some limitations and conclusions from this work should be considered in light of them. First, we focused on symptoms of insomnia rather than the diagnosis. However, this approach allows us to generalize our results to the general population and reflects that notion that symptoms of insomnia are on a continuum. Second, our research focus was on adolescents, yet no previous PRS of insomnia has focused on this developmental stage, meaning that we had to focus on a PRS derived from an adult sample. Nonetheless, in the context of this research we demonstrated that an adult sample derived PRS can be applied to an adolescent population too. Whilst focusing on PRS it is also noteworthy that while the PRS for insomnia captures the main genetic effects on insomnia it does not necessarily imply vulnerability to stress. There could be meaningful differences between mechanisms underlying insomnia and stress vulnerability. Further research is needed in order to identify the specific pathways of underlying these complex relationships (e.g. hypothalamic-pituitary-adrenal axis functioning). Third, the proportion of variance explained by PGS is still limited due to methodological and sample size issues. Fourth, it was not possible to perform a sensitivity analysis using summary statistics from the most recent GWAS of insomnia (Watanabe et al., 2022) as these data were not available at the time of analyses. Of note, we acknowledge that this more recent GWAS found a very similar SNP-based heritability as compared to the GWAS of focus in these analyses (Jansen et al., 2019). Finally, statistical power is reduced when it comes to interaction effects. We used a large sample and our statistical power to detect small effect sizes (0.1%) was around 80% (calculated using the R package “InteractionPowerR”). Nevertheless, future studies should attempt to replicate our results using even larger sample sizes, in different populations and using the most up-to-date GWAS information.

Conclusions

This study investigated for the first time the interaction between polygenic risk and environmental influences on insomnia. Our results show that there is a significant interaction between both variables which supports the 3P model of insomnia. This is clear evidence of the diathesis-stress model for insomnia. In other words, the amount of “stress” needed to trigger insomnia is different based on the genetic vulnerability. Notwithstanding the above caveats, further research is needed to extend these results to other samples (e.g. adult samples) and use different conceptualisations of environmental risk for insomnia.

Key points.

• Insomnia is the most common sleep disorder and is linked to almost every psychiatric disorder.

• Not everyone develops insomnia after experiencing a stressful life-event.

• Negative life-events have a greater impact on insomnia symptoms in those with a higher as compared to lower genetic risk for insomnia.

• The intensity of the precipitating factor (environmental stress) may need to be greater in those with a lower genetic risk (predisposing/protecting factor) as compared to a higher genetic risk in order to trigger insomnia.

Data availability:

under restrictions. JJMV had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis