Autism spectrum disorder (ASD) is a neurodevelopmental disorder that includes repetitive behaviors, impairment in reciprocal social interaction, difficulty communicating, and sensory sensitivities (American Psychiatric Association [APA], 2013). Environmental and genetic factors have been implicated in the etiology of ASD (Feinberg et al., 2015; Schmidt et al., 2011; Volk et al., 2014). Given the complex nature of ASD, gene-environment interaction research may further elucidate the etiology of ASD and point towards potential preventive opportunities. Few studies have used SNPs from a broad selection of targeted genes to investigate gene-by-environment contributions to autism risk.

The fetus, neonate and young child are more sensitive to exposures due to their small size, higher absorption rates, rapid growth, and development of cellular structures, but inferior ability to detoxify exogenous chemicals (Bondy and Campbell, 2005; Grandjean and Landrigan, 2006). Several reviews cite replicated findings that environmental factors are associated with ASD (de Cock et al., 2012; Fujiwara et al., 2016; Hertz-Picciotto et al., 2018; Kalkbrenner et al., 2014). In addition, parental occupational exposures have been found to be associated with ASD; in particular, parental occupational exposure to solvents (McCanlies et al., 2012; McCanlies et al., 2019). Solvents may be absorbed through the skin or lungs and are metabolized into toxic secondary substances including methyl-butyl ketone or n-hexane and are associated with abnormal white matter, smaller corpus callosum volume, and cerebellar atrophy (Hurley and Taber, 2015). Infants of mothers with solvent exposure show cognitive delays, attention deficit hyperactivity disorder, delayed speech, and motor functioning (Bemanalizadeh et al., 2022; Grandjean and Landrigan, 2006). Mothers occupationally exposed to solvents were 1.5 times more likely to have a child with ASD compared to a typically developing child further implicating solvents in the risk for ASD (McCanlies et al., 2019). Similarly, decades of genetic studies provide overwhelming evidence of linkage between ASD and multiple genes on virtually every chromosome (Butler et al., 2015; De Rubeis et al., 2014; Gaugler et al., 2014; Wong et al., 2014), which nevertheless, does not explain most cases of ASD.

As with most complex diseases, causal pathways likely involve interactions between inherited genetic variants and several environmental, chemical, and physical agents that influence immune, endocrine, and neuro-developmental processes (Dietert and Dietert, 2008; Doumouchtsis et al., 2009; Hertz-Picciotto et al., 2008; Pessah et al., 2008). Growing evidence also points to the increased risk for neurocognitive or behavioral impairments from epigenetic changes, which themselves are modulated by environmental factors (Cheroni et al., 2020; Mordaunt et al., 2020; Ramaswami et al., 2020). Moreover, the overlap in regulatory pathways disrupted by both gene mutations and environmental factors highlights convergence between genetic susceptibility and toxic substances (Cheroni et al., 2020; Mordaunt et al., 2020; Ramaswami et al., 2020). Whereas research on ASD, until the last decade had primarily focused on clinical aspects and genetics of autism, an emerging body of evidence is uncovering environmental or occupational exposures appearing either as risk or protective factors. Yet, little research has been conducted to evaluate gene-environment interactions (Gaugler et al., 2014; Kalkbrenner et al., 2014; Lyall et al., 2014; McCanlies et al., 2019). Studies that have been done have primarily focused on a single gene (Volk et al., 2014), genes involved in a single metabolic pathway(Schmidt et al., 2011), or genome-wide copy number variant burden (Kim et al., 2017). Other recent emerging work on such interactions in autism has focused on epigenetic markers at the interface of genes and the environment (Feinberg et al., 2015; Mordaunt et al., 2020; Ramaswami et al., 2020; Schmidt et al., 2011; Zhu et al., 2019). Given the relationship between parental occupational exposure and ASD, evaluating potential parental occupational exposure to solvents in conjunction with relevant SNPs may contribute to a better understanding of the etiology of ASD, and indicate promising molecular pathways and avenues for prevention. Thus, the current study investigates associations between ASD and gene-by-occupational solvent exposure interactions.

Descriptive statistics were used to characterize the study population. The outcome was ASD vs. TD. The solvent exposure, SNP, and SNP x solvent exposure served as the primary predictor variables. We present the analysis of mothers’ and fathers’ data combined. Potential confounders were selected from the literature with reference to the directed acyclic graph (DAG) (Weng et al., 2009). Based on the DAG, all statistical models were adjusted for parents’ age, maternal smoking, length of breastfeeding, mom’s birthplace, regional center, alcohol consumption, and total years of education.

Herein, we report the combinatorial influence of parental solvent exposure and SNP data on the risk of ASD. We identified statistically significant multiplicative and additive interactions between 31 genes and parental occupational exposure to solvents in their relationships to confirmed ASD diagnoses. To our knowledge, this is one of the first studies to evaluate gene x solvent interaction in the risk of ASD.

Results of additive interactions can indicate which exposures are associated with the highest risk of disease and therefore, which subgroup is the most appropriate to target for intervention (Lash et al., 2021). Although there were several sub-additive relationships indicating that some gene SNPs in the presence of solvents may be protective of ASD, this also suggests that the wildtype allele may confer higher risk than the minor allele, placing more individuals at risk of ASD given solvent exposure. While it is prudent to prevent parental occupational solvent exposure in all workers, results here indicate that some individuals may be more sensitive to the effects of solvent exposure than others. For these individuals, any solvent exposure may put them at the highest risk of ASD. Further research needs to be done to better understand the gene, solvent relationship, and how best to protect those at greatest risk.

In addition to public health effects, additive interactions may also correspond more closely to mechanistic interaction than statistical interaction (Lash et al., 2021). Our results suggest synergism in the sufficient cause framework, indicating that the risk of ASD is higher in individuals with both the genes observed here and solvent exposure compared to those who have one or none of the risk factors. Super additive interactions indicated that the risk of ASD is even higher in the presence of the PON1, RORA, and TGFβ2 gene SNPs and solvent exposure. It’s important to note that all the gene SNPs that showed additive interactions, in addition to HTR1F, PSMB9, and TAP1*PSMB9 also showed positive multiplicative interactions, which can also suggest underlying biological mechanisms or sufficient cause interaction (Lash et al., 2021).

A few of the genes we identified have previously been shown to be associated with ASD (e.g. CNTNAP2, RELN, RORA) (Bai et al., 2020; Carter and Blizard, 2016; National Center for Biotechnology Information [NCBI], 2017; Shehabeldin et al., 2018; Stamou et al., 2013), but many have not. However, based on their known functional roles, they are plausible candidates in the etiology of ASD (Supplementary Table 1), being involved in neuronal migration or development (HT, CNTNAP2, ST7*WNT2) (Gilbert and Man, 2017; Muller et al., 2016; Stamou et al., 2013; Stephan, 2008; Watts, 2008), oxidative stress (GLRX 3, SOD2,) (Bowers et al., 2011; Giulivi et al., 2010; Stamova et al., 2013), detoxification (ALDH, NAT1 and NAT2, PON1) (Sabbioni et al., 2006; Vasiliou and Nebert, 2005), or an immune response and inflammation (IL, TGFβ2, HLA) (Ashwood et al., 2011a; Ferrante et al., 2003; Krakowiak et al., 2017; Torres et al., 2002; Warren et al., 1996). Furthermore, the functional role of these genes suggests how they may be interacting with solvents in the risk of ASD, which we will further discuss below.

We identified joint interactions with solvents and several serotonin genes (HTR1A, HTR1B, HTR1F, HTR2A, HTR4, HTR7), RELN, CNTNAP2, and ST7*WNT. These genes are associated with neurodevelopment or embryonic development. Serotonin specifically is associated with neuronal differentiation, neurogenesis, synaptogenesis and controls the activity of GABAergic interneurons, which have also been found to be affected in autistic children (Watts, 2008; Zafeiriou et al., 2009). Our results are also consistent with research showing an association between serotonin receptor genes and ASD (Butler et al., 2015; Muhle et al., 2004; Whitaker-Azmitia, 2001). While there is little research specifically evaluating solvents and serotonin genes, alcohol consumption can cause cell apoptosis of neurons particularly serotoninergic neurons and xylene exposure has an inhibitory effect on GABA, a product of serotonin (Boschen and Klintsova, 2017; Niaz et al., 2015; Pruett et al., 2013), suggesting that solvents may interact either with serotonin genes directly or their products such as GABA, interfering with neural development. Further research could elucidate the molecular basis for a solvent-HTR interaction mechanism in ASD (Chen et al., 2004; Lin et al., 2009; Niaz et al., 2015).

The joint interaction between RELN rs56041591 and parental occupational exposure to solvents was also associated with ASD. RELN encodes for an extracellular matrix protein that controls cell-cell interaction critical for cell positioning and neuronal migration during brain development and plays an important role in synaptic connectivity and plasticity (Shehabeldin et al., 2018) (Gilbert and Man, 2017). Solvent exposure is associated with several neurological effects and changes including interfering with the glial guidance processes which inhibit neuritic outgrowth (Bondy and Campbell, 2005; Hurley and Taber, 2015). Thus, the interaction between RELN and parental occupational exposure to solvents may reflect converging or intersecting pathways that interfere with critical aspects of brain development (Gilbert and Man, 2017).

Similarly, CNTNAP2 is a synaptic protein and a member of the neurexin family that mediates cell-to-cell communication and may be involved in axon differentiation and neuronal migration, while ST7*WNT2 is in the same region as RELN on chromosome 7 (National Center for Biotechnology Information [NCBI], 2019; Rodenas-Cuadrado et al., 2014). It is expressed during development in several tissues including the nervous system (Katoh, 2002; National Center for Biotechnology Information [NCBI], 2019). Solvents may interact directly with CNTNAP2 or ST7*WNT2 SNPs, or their protein product interfering with cell-to-cell communication, neural connectivity, or migration, increasing the risk of ASD.

In the presence of solvents, RORA rs67288758 was protective of ASD. In contrast, we saw significant odds of ASD in the presence of parental solvent exposure and both RORA rs75941956 and EGF rs11569014, as well as additive and multiplicative interactions between solvents, RORA, SOD2, and GLRX3 SNPs. These results are consistent with the observation that solvent exposure may result in oxidative stress and the formation of reactive oxygen species (ROS) or reactive nitrogen species (RONS) (Khan and Wang, 2018; Moro et al., 2012). Solvent exposure triggers an inflammatory response and can cause neuronal apoptosis (Fisseler-Eckhoff et al., 2011; Pruett et al., 2013). RORA, on the other hand, protects neurons from inflammation and oxidative stress (Hu, 2012). Our results suggest that in the presence of solvents, RORA rs67288758 may be able to protect neurons from oxidative stress while the rs75941956 SNP can’t (Hu, 2012). It’s unclear how solvents may interact with RORA SNPs in the risk of ASD, perhaps it directly interacts with the RORA SNP triggering an inflammatory response causing neuronal apoptosis, or inflammation that the SNP is unable to mitigate, or solvents may interfere directly with RORAs ability to protect neurons from inflammation and oxidative stress (Fisseler-Eckhoff et al., 2011; Pruett et al., 2013).

Like RORA, EGF, SOD2 and GLRX3 are associated with buffering oxidative stress(Esparham et al., 2015; Maher, 2006; Stamova et al., 2013). EGF is involved in redox regulation and signaling and promotes cell differentiation and proliferation in neural progenitor cells and has been shown to be associated with ASD (Behring et al., 2020; Galvez-Contreras et al., 2017) (National Institutes of Health [NIH], July 16, 2019). Similarly, SOD2 and GLRX3 have been shown to offset or reduce oxidative stress (Stamova et al., 2013) (Bowers et al., 2011; Maher, 2006). GLRX3 is thought to be important in maintaining nerve cell function, which may also partially explain its association with ASD (Bowers et al., 2011). Solvent exposure may interfere with either the genes or the gene products, reducing their ability to buffer oxidative stress, increasing the risk of ASD.

PON1 has a multitude of functions including altering the expression of numerous genes associated with oxidative stress, but also plays a role in detoxification, specifically, it detoxifies organophosphate pesticides (OP) (Carter and Blizard, 2016; Mackness and Mackness, 2015). PON1 variants interact with OPs in the risk of autism (D’Amelio et al., 2005). Whether PON1 SNPs interact with solvents like OPs, or results in oxidative stress increasing the risk of ASD is unclear, further research is necessary to clarify this relationship and how solvent exposure may be interacting with PON1 SNPs to increase the risk of ASD.

ALDH5A1, NAT1, NAT2 are involved in detoxification and drug metabolism (National Center for Biotechnology Information [NCBI], 2017; National Center for Biotechnology Information [NCBI], 2019; Vasiliou and Nebert, 2005). Variations in ALDH5A1 are associated with developmental delays and other neurological complications (National Center for Biotechnology Information [NCBI], 2019; Vasiliou and Nebert, 2005). Our results may indicate that ALDH5A1 variants are involved in the metabolism of solvents and poor metabolism may be associated with ASD (National Center for Biotechnology Information [NCBI], 2019). NAT encodes for enzymes that help metabolize xenobiotics and drugs (National Center for Biotechnology Information [NCBI], 2017). NAT2 fast acetylation was associated with neuropsychological impairment in solvent exposed dock workers (Dick et al., 2002). Our results indicate that NAT1 and NAT2 may be involved in the biotransformation of solvents influencing the risk of ASD.

Several studies suggest that neuroinflammation may be involved in pathogenesis of ASD (Ashwood et al., 2008; Ashwood et al., 2011a; Ashwood et al., 2011b; Ashwood et al., 2011c; Kelder et al., 1998; Krakowiak et al., 2017; Matta et al., 2019; Pardo et al., 2005). In the presence of solvents, several inflammatory gene SNPS in the IL, TGFβ2, HLA class I and class I MHC genes, including SNPs in GABBR1, PSMB9, TAP1*PSMB9, and TAP2 SNPs were associated with ASD. Inflammatory cytokines are expressed in the developing brain, affecting the function and development of neuronal and glial cells, and a large literature implicates maternal immune activation in ASD (Zawadzka et al., 2021). Similarly, TGFβ2 is important in embryonic development and regulates the immune system (National Institutes of Health [NIH], July 16, 2019). The joint interaction between IL or TGFβ2 gene SNPs and solvents may trigger an immune response, interfere with the glial guidance process in infants, interfere with the genes resulting in inflammation, or cause cell apoptosis, increasing the risk of ASD (Barragan-Martinez et al., 2012; Bondy and Campbell, 2005; Hurley and Taber, 2015; Pruett et al., 2013).

Lastly, class I HLA proteins are important in synaptic plasticity and neuronal connections (Boulanger and Shatz, 2004). Independent of the inflammatory response, HLA-class II is expressed in human neurons and microglia and may be important in embryonic neural development (Vagaska et al., 2016). Immune challenges may change levels of MHC-I proteins in the brain, indicating an important link between immune activation and brain wiring. Solvents may affect immune responses or increase auto-immune tendencies(Barragan-Martinez et al., 2012; Gerhardsson et al., 2021; Khan and Wang, 2019), suggesting ASD risk could be influenced by HLA genes interacting with solvents in an immunologic cascade affecting brain development, wiring, or neuronal cell death or in the case of antagonistic relationships, be protective against damage (Barker et al., 2001; Kahn et al., 1964).

Solvent exposure is associated with several neurological effects and changes including oxidative stress, inflammation, and cell apoptosis of neurons (Hurley and Taber, 2015; Pruett et al., 2013). For example, in infants, solvent exposure interferes with the glial guidance process which inhibits neuritic outgrowth (Bondy and Campbell, 2005). It has an inhibitory effect on GABA and has been found to bind directly to the GABA_A receptor (Boschen and Klintsova, 2017; Niaz et al., 2015). The mechanisms linking solvent exposure to ASD suggests that the interaction between solvents and the genes identified in this study may trigger inflammation, oxidative stress, or possibly interfere with neuronal development. However, further functional research needs to be conducted to confirm these findings and to help elucidate the causal pathway between gene, solvent exposure, and ASD.

This study has both strengths and limitations. Strengths include: use of gold standard diagnostic instruments for confirmation of case status and research reliable psychometricians, resulting in accurate, consistent developmental classification(Hertz-Picciotto et al., 2006). We employed an efficient strategy to enhance power of gene-environment analyses by selection of candidate gene SNPs based on established or likely role in the etiology of ASD. The analytic methods were designed to reduce confounding through screening and control of many unknown and suspected risk factors for ASD. Additionally, population-based recruitment of participants reduced selection bias, enhancing the representativeness of the target population and thus, increasing the generalizability of the results. Despite the sample size being larger than most gene-environment interaction studies in ASD, it may still have been too small to see potential SNP or solvent associations with ASD, and even more so to identify interactions between occupational solvent exposures and some candidate SNPs in relation to ASD risk. However, after correcting for multiple testing, several interaction p-values remained significant. An additional limitation of the study may be that the selection of genes discovered originates from primarily European ancestry populations, while our cohort has a substantial proportion of individuals of other ancestries. However, it is also true that most of the studies on autism (and most other disorders, as recognized by the NIH) have been in European-derived populations (National Institutes of Health [NIH], 2019). Therefore, any candidate gene analysis based on the literature would face the same limitations. There have been several instances of cross-population associations in autism and other disorders, so we feel that our choice of candidate genes is reasonable (Keys et al., 2020). Additional analyses that consider genes found in other populations are warranted, though further genetic research in non-European populations would need to be conducted. The proportion of the CHARGE cohort with non-European ancestry, including non-white race and Hispanic ethnicity is about 45%.

Obtaining accurate exposure data can be challenging. Here we used IH-assessment based on parent reported job title, tasks, and responsibilities; a methodology that is less affected by recall bias than asking parents to report their specific workplace exposures (Teschke et al., 2002). Factors that may affect the accuracy of estimating exposure include, the industrial hygienists’ familiarity with specific jobs, variability in solvent exposure within each job, the use of personal protective equipment, and in some instances, access to accurate job information. Nonetheless, while IH generated exposure assessment is less sensitive, the specificity is generally more stable, resulting in less misclassification bias and attenuation of the odds ratios (Benke et al., 2001b). Misclassification bias can be further reduced if information such as responsibilities, task, and duties is also available as it was in this study (Benke et al., 2001a; Teschke et al., 2002). However, father’s job histories completed by the mother may be less accurate than those completed by the father, which could have led to misclassification of exposure and decreased precision in ORs. Lastly, although we did not have three or more IHs to assess occupational exposure, use of two IHs (as we did) generally improves reliability and validity over a single IH (Fritschi et al., 2003; Siemiatycki et al., 1997).