Association Between Childhood Green Space, Genetic Liability, and the Incidence of Schizophrenia

Abstract

Childhood exposure to green space has previously been associated with lower risk of developing schizophrenia later in life. It is unclear whether this association is mediated by genetic liability or whether the 2 risk factors work additively. Here, we investigate possible gene–environment associations with the hazard ratio (HR) of schizophrenia by combining (1) an estimate of childhood exposure to residential-level green space based on the normalized difference vegetation index (NDVI) from Landsat satellite images, with (2) genetic liability estimates based on polygenic risk scores for 19 746 genotyped individuals from the Danish iPSYCH sample. We used information from the Danish registers of health, residential address, and socioeconomic status to adjust HR estimates for established confounders, ie, parents’ socioeconomic status, and family history of mental illness. The adjusted HRs show that growing up surrounded by the highest compared to the lowest decile of NDVI was associated with a 0.52-fold (95% confidence interval [CI]: 0.40 to 0.66) lower schizophrenia risk, and children with the highest polygenic risk score had a 1.24-fold (95% CI: 1.18 to 1.30) higher schizophrenia risk. We found that NDVI explained 1.45% (95% CI: 1.07 to 1.90) of the variance on the liability scale, while polygenic risk score for schizophrenia explained 1.01% (95% CI: 0.77 to 1.46). Together they explained 2.40% (95% CI: 1.99 to 3.07) with no indication of a gene–environment interaction (P = .29). Our results suggest that risk of schizophrenia is associated additively with green space exposure and genetic liability, and provide no support for an environment-gene interaction between NDVI and schizophrenia.

Introduction

Schizophrenia is a severe mental illness with an estimated number of 20.9 million cases in 2016 globally. Even though schizophrenia is a low-prevalence disorder, associated societal and financial costs are high relative to other chronic conditions because of the severity of the disorder.^1–3 Despite decades of extensive scientific research, the underlying causes are still poorly understood. The causes are likely to be multifactorial and may include social conditions, genetic risk factors, and environmental risk factors.^4–8

Green space (parks, grasslands, forests, etc.) have recently been presented as a novel environmental risk factor for schizophrenia, with high values being associated with lower risk.⁹ Green space could potentially influence schizophrenia risk by supporting recovery of depleted psychological, physiological, and social resources through psychological restoration, decreasing noise and air pollution affecting cognition and brain development, and improving immune functioning.^10–13 However, it remains unclear whether green space and genetic liability are both independently and additively associated with schizophrenia or whether choosing to live surrounded by more or less green space could be genetically determined.

High heritability estimates for schizophrenia indicate that genetic liability is an important risk factor for the development of the disease.^6,14,15 A previous study showed that genetic liability was unable to explain the association between urbanization and schizophrenia in the Danish population.¹⁶ Still, genetic liability may confound the association between green space and schizophrenia, eg, if parents with higher genetic liability tend to live in less green areas or alternatively if parents that are genetically attracted to greener living areas also have lower schizophrenia liability. Furthermore, a gene–environment interaction may increase risk in areas with less green, as the “stress-vulnerability” model of etiological influence would predict increased expression of schizophrenia in at-risk persons following exposure to environmental risk factors.⁸

In this study, the aim is to investigate the association between schizophrenia risk, green space, and genetic liability for schizophrenia. Specifically, we investigate the following 4 questions: (1) is there an individual association between schizophrenia risk and the risk factors low green space exposure and genetic liability for schizophrenia; (2) what is the strength and direction of the associations between green space exposure and schizophrenia risk and between genetic liability and schizophrenia risk; (3) is the association between childhood green space exposure and rates of schizophrenia mediated by genetic liability for schizophrenia or is the association additive; and (4) is there a gene–environment interaction between green space and genetic liability and the association with schizophrenia risk?

Methods

Data Sources

The Danish Civil Registration System¹⁷ was established in 1968, and includes all individuals living in Denmark. Each individual is assigned a unique civil registration number at birth or immigration date, and this number is used in all Danish registration systems enabling accurate linkage within and between them. The Danish Civil Registration System also contains information on gender, date and place of birth, date of immigration, change of address, date of emigration, vital status, and links to family members. The Danish Psychiatric Central Research Register¹⁸ contains information on every psychiatric admission from 1969 onwards. In 1995, data on psychiatric outpatient treatment and emergency room contacts were included. Dates of admission and discharge, type of referral, and mode of admission are stored together with all discharge diagnoses assigned by the treating physicians. Diagnoses are coded according to the International Classification of Diseases (8th revision) (ICD-8)¹⁹ until the end of 1993, with the 10th revision (ICD-10)²⁰ applied thereafter. As a measure of green space, we used the normalized difference vegetation index (NDVI) values obtained from 30 m × 30 m resolution remote sensing images from the Landsat archive covering Denmark since 1985. Based on this data, we calculated the mean NDVI within a 210 m × 210 m square regions centered at each individual’s residence from birth until the 10th birthday. The genetic liability estimates based on polygenic risk scores were acquired from data from the Danish Newborn Screening Biobank.²¹ This agency maintains samples of residual dried blood spot samples after routine screening in the first days after birth for nearly all infants born in Denmark after 1981. The socioeconomic data were acquired from the Integrated Database for Longitudinal Labor Market Research, and the Danish Education Registers cover the entire population and contain yearly information from 1980 including income, education, and employment status.^22,23

Study Population

We used data from the iPSYCH case–cohort sample to identify all schizophrenia cases.²⁴ A random sample of 30 000 individuals were selected from a study population consisting of all singleton births between May 1, 1981 and December 31, 2005, who were alive and residing in Denmark at their first birthday and had a known mother (N = 1 472 762). All cases were diagnosed through the Danish Psychiatric Central Research Register. Diagnoses were made between 1994 and 2012 for individuals 10 years and above if they had been admitted to a psychiatric facility, received outpatient care, or visited a psychiatric emergency care unit with a diagnosis of schizophrenia (ICD-10: F20). We started follow-up at age 10, thus restricting the dataset to individuals born on or before December 31, 2002, and only included individuals living in Denmark from birth to age 10. Since information on green space is available from 1985, we included only individuals born after January 1, 1985. To ensure complete information on family history of mental disorder, only those individuals with a known mother and father both born in Denmark were retained. The final study population consisted of all genotyped individuals meeting the above criteria.

Childhood Green Space

Using information from the Civil Registration System, we created a complete history of current and former residential addresses including dates of changes in address for each individual within the first 10 years of life. Exact geographical coordinates were obtained by linking these data with information from the Danish address register. The coordinates were used to calculate for each individual the mean yearly NDVI within square-shaped zones of 210 m × 210 for each year between birth and the 10th birthday as a measure of green space exposure.²⁵ NDVI values range between −1 and 1, with low values indicating sparse vegetation and high values indicating dense vegetation. NDVI is a reliable measure of green space,²⁶ but smaller green spaces may be missed as NDVI was obtained from 30 m × 30 m resolution images. Further details on the calculation of NDVI are published previously.⁹

Polygenic Risk Score

DNA extracted from dried blood spot samples from the Danish Newborn Screening Biobank was genotyped using 3 different genotyping chips, Illumina Human 610-Quad BeadChip array, Illumina HumanCoreExome beadchip for the precursor studies²⁷ and Illumina Infinium PsychArray-24-v.1.1 BeadChip²⁸ for the iPSYCH-sample (further details are published previously^24,29). After quality control, the 3 samples comprised 81 561 individuals. For each data set, only the genotyped variants (minor allele frequency above 1%) were used for constructing polygenic risk scores as measures of genetic liability for schizophrenia. These scores capture each individual’s load of common genetic variants associated with schizophrenia. The polygenic risk scores were calculated separately accounting for linkage disequilibrium (LD) using LDpred for each of the datasets.³⁰ Results from the latest Psychiatric Genomics Consortium (PGC) schizophrenia sample excluding the Danish data were included in a meta-analysis yielding a discovery sample of 34 600 schizophrenia cases and 45 986 controls. The polygenic risk scores for schizophrenia were calculated using the summary statistics (effect allele, size, and direction) derived from the discovery sample. For the LD reference, we first randomly sampled 5000 individuals, and then excluded individuals more than 1 SD away from the average 1000 genomes CEU (northern and western European ancestry) genotype when projected onto a space spanned by the first 2 principal components in the 1000 genomes (phase 3), and cryptic-related individuals $(\widehat{\pi }>0.05)$. This subsample of 4456 unrelated individuals of European ancestry broadly reflects the LD pattern of the individuals on which the summary statistics are based. As parameters for LDpred, the LD radius was set to 100 single-nucleotide polymorphisms (SNPs) for the 2 precursor studies (3570 individuals), and 150 SNPs for the iPSYCH sample (78 003 individuals). The polygenic risk score corresponding to LDpred p-parameter (expected fraction of causal variants) of 0.3 was used as the prediction for the precursor studies, and LDpred-inf was used for the iPSYCH sample.

The calculated polygenic risk scores were standardized, initially separately for the 3 samples using arithmetic mean and SD among controls/subcohort to account for differences in genotyping procedures, eg, using different genotyping chips, and then for the entire sample based on mean and SD in the subcohort.

Covariates

Information on age, gender, and parents was obtained from the Civil Registration System. Parental history of mental disorder at the individual’s 10th birthday was extracted from the Danish Psychiatric Central Research Register and classified hierarchically as schizophrenia spectrum disorder (ICD-8: 295.x9, 296.89, 297.x9, 298.29–298.99, 299.04, 299.05, 299.09, 301.83; ICD-10: F20-29), affective disorder (ICD-8: 296.x9 [excluding 296.89], 298.09, 298.19, 300.49, 301.19; ICD-10: F30-39), or any other mental disorder (ICD-8: 290–315; ICD-10: F00-F99). Paternal and maternal gross income (in quintiles), their highest educational attainment level (primary school, high school/vocational training, higher education), and employment status (employed, unemployed, outside workforce for other reasons) were defined at age 10 for each individual. To prevent confounding from population stratification, ie, allele frequency differences due to systematic ancestry differences, we conducted a principal component analysis (PCA) using smartPCA of the EIGENSOFT 6.1.4 package to generate genomic principal components based on a relatedness-pruned set of individuals $(\widehat{\pi }>0.09)$ and a subset of SNPs (minor allele frequency > 0.001 and pruned for LD [r² < .05]). All individuals were subsequently projected back onto the principal component space based on genotypes and SNP loadings.

Statistical Analysis

Individuals were followed from their 10th birthday until admission for schizophrenia, emigration, death, or end of follow-up (December 31, 2012), whichever came first. Since cases were oversampled in the case-cohort design, resulting in 70 cases being selected into the sub-cohort, each individual’s risk time was weighted according to the case–cohort sampling probability using inverse probability weighting (Borgan’s estimator II³¹). The weight for cases was set to 1 since all cases were sampled, and the weight for noncases was calculated as the reciprocal of the sampling fraction.

We estimated the least square means of NDVI and polygenic risk score for schizophrenia cases and noncases, adjusted for age, gender, birth year, parental history of mental disorder, and parental socioeconomic status.³² Mean polygenic risk score was additionally adjusted for the first 10 genomic principal components to control for population stratification. The relationship between childhood NDVI and polygenic risk score for schizophrenia was estimated using Pearson’s correlation and linear regression analysis.

Crude and adjusted hazard ratios (HRs) for schizophrenia with 95% confidence intervals (CIs) were estimated by fitting weighted Cox regression models with robust standard errors. The basic model was adjusted for age, gender, and birth year (and first 10 genomic principal components). We then further adjusted for parental history of mental disorder and parental socioeconomic status. To evaluate the joint effect of childhood exposure to NDVI and polygenic risk score, we added a product term to assess interaction in the regression model.

We calculated Nagelkerke’s R² values and transformed them to the liability scale³³ with population prevalence of 1% and the proportion of cases in our sample, to quantify the variance explained by childhood NDVI, polygenic risk score for schizophrenia, and their interaction. The liability scale has proven better for combining estimates into a comprehensive risk model, particularly for joint models with environmental and genetic risk factors.³⁴ 95% CIs were obtained by bootstrapping (n = 10 000).

Previous research has identified higher risk of schizophrenia in urban compared to rural areas,^5,35 and we finally performed a sensitivity analysis including population density at age 10 as a more refined measure of urbanization. Population density was calculated as quintiles of the number of people residing within 500 m of each individual in the study population at their 10th birthday.

To improve accuracy of the polygenic risk score for schizophrenia derived in samples with European ancestry, we performed a sensitivity analysis excluding cryptic-related individuals $(\widehat{\pi }>0.02)$ and ancestral outliers. We selected European ancestry based on an ellipsoid in the space of the first 3 principal components centered and scaled using the mean and 6 SD of a subsample whose parents and grandparents were all known to have been born in Denmark. Genomic principal components were then recalculated using PLINK version 1.9.

Epidemiological analyses were performed using Stata (StataCorp LLC) version 15.1 and SAS statistical software (SAS Institute Inc.) version 9.4.

Results

A total of 19 746 individuals were included in the case–cohort study: 2636 cases with a schizophrenia diagnosis and 17 180 subcohort members. As subcohort members were selected by chance, 70 cases were also selected into the subcohort and subsequent treated as cases. Table 1 shows description of childhood NDVI and polygenic risk score for schizophrenia. Individuals with schizophrenia grew up surrounded by lower NDVI and had a higher polygenic risk score for schizophrenia (P < .0001) (table 1, figure 1). Further characteristics of the study population are provided in supplementary table S1.

Table 1.

Characteristics of the 19 746 Individuals in the Case–Cohort Study

Baseline Characteristics	Schizophrenia (n = 2636)	Subcohort (n = 17 180)
Childhood NDVI
Crude mean (SD)a	0.21 (0.19)	0.33 (0.18)
Decilesb
1	604 (22.91)	1718 (10.00)
2	519 (19.69)	1718 (10.00)
3	357 (13.54)	1718 (10.00)
4	287 (10.89)	1718 (10.00)
5	239 (9.07)	1718 (10.00)
6	178 (6.75)	1718 (10.00)
7	150 (5.69)	1718 (10.00)
8	114 (4.32)	1718 (10.00)
9	104 (3.95)	1718 (10.00)
10	84 (3.19)	1718 (10.00)
Polygenic risk score for schizophrenia
Crude mean (SD)	0.35 (1.02)	0.00 (1.00)
Deciles
1	144 (5.46)	1718 (10.00)
2	197 (7.47)	1718 (10.00)
3	190 (7.21)	1718 (10.00)
4	205 (7.78)	1718 (10.00)
5	210 (7.97)	1718 (10.00)
6	259 (9.83)	1718 (10.00)
7	293 (11.12)	1718 (10.00)
8	303 (11.49)	1718 (10.00)
9	345 (13.09)	1718 (10.00)
10	490 (18.59)	1718 (10.00)

Note: NDVI, normalized difference vegetation index.

^aThe crude mean shows the unadjusted hazard ratios for schizophrenia according to deciles of mean childhood NDVI and polygenic risk score for schizophrenia.

^bIntegers reports the number of cases/subcohort members, and numbers in parentheses shows the percentage of cases/subcohort members.

Fig. 1.

Mean childhood normalized difference vegetation index (NDVI) and mean polygenic risk score for schizophrenia according to schizophrenia. Means (95% confidence intervals) were calculated as least square means and adjusted for age, gender, birth year, parental history of mental disorder, and parental socioeconomic status. Mean polygenic risk score for schizophrenia was additionally adjusted for first 10 genomic principal components.

We found a weak inverse correlation between childhood NDVI and polygenic risk score for schizophrenia (ρ = −0.0779, P < .0001) for the subcohort members, meaning that the higher the NDVI, the lower the polygenic risk score for schizophrenia (β = −0.014, 95% CI: −0.017 to −0.011). However, this association was strongly attenuated after adjusting for age, gender, first 10 genomic principal components, birth year, parental history of mental disorder, and parental socioeconomic status (β = −0.002, 95% CI: −0.004 to 0.000). Estimates for all covariates are shown in supplementary table S2. In the Cox regression models, adjusted for age, gender, birth year, parental history of mental disorder, parental socioeconomic status (and first 10 genomic principal components), and both polygenic risk score for schizophrenia and NDVI were associated with schizophrenia (table 2). Increasing the NDVI was associated with decreased risk of schizophrenia (adjusted HR = 0.52, 95% CI: 0.40 to 0.66), and higher polygenic risk score for schizophrenia was associated with elevated risk of schizophrenia (adjusted HR = 1.24, 95% CI: 1.18 to 1.30). Additional adjustment for the other risk factors had virtually no impact on the estimated effects sizes (table 2). When including population density in the models, the association between childhood NDVI and schizophrenia was attenuated but remained statistically significant except for the mutually adjusted estimate (supplementary table S3). Finally, our results remained robust after excluding cryptic-related individuals and ancestral outliers (supplementary table S4) as expected since the study population was restricted to individuals born in Denmark by Danish-born parents.

Table 2.

Adjusted Hazard Ratios (95% Confidence Intervals) for Schizophrenia According to Childhood NDVI and Polygenic Risk Score for Schizophrenia (n = 19 746)

Adjusted for	Childhood NDVIa	Polygenic Risk Scoreb
Age, gender, calendar year	0.40 (0.32–0.50)	1.26 (1.21–1.32)
Age, gender, calendar year, parental history of mental disorder, and parental socioeconomic status	0.52 (0.40–0.66)	1.24 (1.18–1.30)
Age, gender, calendar year, parental history of mental disorder, parental socioeconomic status, and mutually adjusted	0.54 (0.42–0.70)	1.24 (1.18–1.30)

Note: NDVI, normalized difference vegetation index.

^aThe estimate for childhood NDVI measure the decreased risk of schizophrenia associated with 1-unit increase in mean NDVI during the first 10 years of life.

^bThe estimate for polygenic risk score measure the increased risk of schizophrenia associated with 1 SD increase in polygenic risk score. The estimate was additionally adjusted for first 10 genomic principal components.

The risk of developing schizophrenia was related to both polygenic risk score for schizophrenia and NDVI in a dose–response relationship (figure 2). Compared with individuals in the lowest decile, individuals in the highest NDVI deciles had lower risk of developing schizophrenia (HR = 0.62, 95% CI: 0.48 to 0.80), and individuals with polygenic risk score in the highest decile had higher risk (HR = 2.02, 95% CI: 1.62 to 2.52).

Fig. 2.

Hazard ratios (95% confidence intervals) for schizophrenia according to deciles of childhood normalized difference vegetation index (NDVI) and polygenic risk score for schizophrenia. Hazard ratios are mutually adjusted, and further adjusted for age, gender, birth year, first 10 genomic principal components, parental history of mental disorder, and parental socioeconomic status.

We found that childhood NDVI explained 1.45% (95% CI: 1.07 to 1.90) of the variance on the liability scale, while polygenic risk score for schizophrenia explained 1.01% (95% CI: 0.77 to 1.46) (table 3). Together they explained 2.40% (95% CI: 1.99 to 3.07), suggesting additivity. We found no evidence of an interaction between childhood NDVI and polygenic risk score for schizophrenia (P = .29). Including the interaction term in the model did not increase explanatory power.

Table 3.

Hazard Ratios (95% Confidence Intervals) and R² (95% Confidence Intervals) Calculated for Logistic Regression Models Estimating Schizophrenia Risk Associated With NDVI During Childhood and Polygenic Risk Score Singly or Jointly Along With Age, Gender, Birth Year, Parental History of Mental Disorder, and Parental Socioeconomic Status (and First 10 Genomic Principal Components) (n = 19 746)

	OR (95% CI)	Nagelkerke’s R2 (%)	R  2 on Liability Scale (%)
Model 1		2.21 (1.63–2.90)	1.45 (1.07–1.90)
Childhood NDVI	0.13 (0.10–0.18)
Model 2		1.54 (1.17–2.24)	1.01 (0.77–1.46)
Polygenic risk score for schizophrenia	1.32 (1.25–1.38)
Model 3		3.65 (3.02–4.63)	2.40 (1.99–3.07)
Childhood NDVI	0.14 (0.10–0.19)
Polygenic risk score for schizophrenia	1.32 (1.25–1.38)
Model 4		3.65 (3.03–4.65)	2.41 (2.00–3.08)
Childhood NDVI	0.13 (0.10–0.18)
Polygenic risk score for schizophrenia	1.29 (1.19–1.39)
Interaction term	1.10 (0.84–1.43)

Note: CI, confidence interval; OR, odds ratio; NDVI, normalized difference vegetation index.

Discussion

We found that childhood green space and genetic liability for schizophrenia appeared to be uncorrelated. Furthermore, they were independently associated with risk of schizophrenia in dose–response relationships: growing up surrounded by more green space is associated with lower risk, whereas having high genetic liability is associated with higher risk. We found no evidence of an interaction between green space and genetic liability for schizophrenia, indicating that a gene–environment interaction does not increase risk beyond each risk factor separately. The findings are consistent with additive effects of genetic liability and green space exposure on schizophrenia risk.

Growing up surrounded by more green space is associated with lower risk of developing schizophrenia later in life consistent with results from previous studies.^9,36 The association was not greatly attenuated by adjusting for genetic liability. However, as the polygenic risk score only captures a fraction of the expected genetic liability, the possibility of genetics contributing to the association between green space and schizophrenia cannot fully be rejected. However, this does strengthen the epidemiological evidence in support of green space being independently associated with schizophrenia risk and not just a correlate from individuals’ choice of living being genetically determined. This is further supported by our finding of a dose–response relationship between green space and schizophrenia risk, and the fact that the associations remained after adjusting for various confounders. Not unexpectedly, adjusting for family history of mental illness and parent’s socioeconomic status, slightly attenuated the association between green space and schizophrenia risk, indicating that part of the association is explained by family-related factors—eg, a family’s financial situation may affect where the family chooses to live. Despite adjusting for several potential confounders, our results may be influenced by unmeasured socioeconomic and environmental factors such as neighborhood deprivation and air pollution. Future studies should focus on investigating potential pathways through which green space may influence schizophrenia risk.

Genes may influence schizophrenia risk by altering sensitivity to environmental exposures⁸ including green space. Despite finding that people with schizophrenia grew up surrounded by less green space and had higher genetic liability, we found no evidence of a gene–environment interaction. Green space and genetic liability were weakly correlated, indicating a possible, but slight, gene–environment correlation. However, comparing models with the 2 factors alone to a model with both risk factors suggest almost perfect additivity. Interestingly, green space exposure explained more variation in schizophrenia risk than genetic liability. This could be because the current polygenic risk score only captures a small proportion of the variance attributed to genetic risk. Still, we speculate, that if green space influence schizophrenia risk, modifying exposure could be promising for the prevention of schizophrenia⁷ in addition to other health benefits.^13,37,38 Effects of green space exposure is consequently a scientific question that deserves further attention—especially in high-risk urban environments, where green space access is threatened globally by urban densification and development.³⁹

To our knowledge, this is the first study to jointly consider the role of green space exposure in childhood and genetic liability for schizophrenia risk. The major strengths of this register-based longitudinal study are the longitudinal design, the high resolution and precise residential-level green space exposure, and the large cohort of genotyped individuals. Having access to the extensive Danish register data allowed us to control for confounding by parental socioeconomic status and history of mental disorder, and population density.

Nevertheless, this study also has several limitations. First, the diagnosis of schizophrenia was based on registered hospital contacts, and consequently individuals with schizophrenia only treated in the primary healthcare are missing. However, as schizophrenia is a serious psychiatric disorder, typically requiring secondary care treatment, this misclassification error will be relatively small and likely introduces almost zero bias in our findings. Second, the study used residential addresses to define green space exposure, ignoring actual use of green space and exposure at day care, school, or during commuting time.⁴⁰ Future studies should aim at obtaining measures of individual’s use of their surrounding green space to better understand the association between green space and schizophrenia risk. Also, using higher resolution satellite images and a smaller exposure zone could provide more detailed information about green space exposure. Third, the polygenic risk score only captures a proportion of the variance attributed to underlying genetic risk. The polygenic risk score aggregates many common risk alleles, but does not include rare variants and copy number variants, which may be important for the development of schizophrenia.^4,41,42 Additional variants associated with schizophrenia are likely to be identified in future genome-wide associations studies covering larger population samples, and these may prove useful in the future for refining estimates of genetic risk for schizophrenia. Also, part of the association with green space may be explained by genetic nurturing. This hypothesis proposes that parents’ behaviors are determined by both transmitted and nontransmitted alleles, and thus an environmental effect (ie, green space) on children may be partly genetic if parents behavior are determined by a gene–environment interaction with nontransmitted alleles.⁴³ Finally, although our analyses were adjusted for a range of known confounding factors, the possibility of unknown or unmeasured confounding remains.

Detecting gene–environment interactions requires accurate exposure assessment, identification of relevant genetic measures, and more analytical power than 1-way associations.⁴⁴ To identify a gene–environment interaction, the effect has to be very large to be distinguishable from the additive effect of the 2 risk factors. Obtaining more complete measures of green space exposure and genetic liability could provide evidence of a gene–environment interaction in the future. An improved Danish polygenic risk score (PRS) for schizophrenia is coming, and at this point, we cannot conclusively dismiss that a gene–environment interaction exists between green space and genetic liability for schizophrenia.

In conclusion, growing up surrounded by more green space was associated with lower schizophrenia risk with no evidence of mediation by genetic liability or confounding by socioeconomic status, family history of mental illness, and population density. This result provides further support for the hypothesis that increasing green space exposure in childhood could be a protective environmental factor for schizophrenia. Green space and genetic liability were independently and consistently associated with schizophrenia risk, with slightly higher association for green space, indicating that these risk factors operate in an additive fashion. We found no evidence of an environment–gene interaction suggesting that any possible interaction between the 2 risk factors is small. Meanwhile, these results may help researchers across disciplines to better understanding the risk factors of schizophrenia and other psychiatric disorders.

Funding

This study was supported by the Lundbeck Foundation, Denmark (grant numbers R102-A9118 and R155-2014-1724); the Stanley Medical Research Institute; the Danish Strategic Research Council; the Novo Nordisk Foundation (Big Data Centre for Environment and Health, grant number NNF17OC0027864); and VILLUM FONDEN (VILLUM Investigator, grant number 16549 to J.-C. S.).