Convergent neural correlates of prenatal exposure to air pollution and behavioral phenotypes of risk for internalizing and externalizing problems: Potential biological and cognitive pathways

Abstract

Humans are ubiquitously exposed to neurotoxicants in air pollution, causing increased risk for psychiatric outcomes. Effects of prenatal exposure to air pollution on early emerging behavioral phenotypes that increase risk of psychopathology remain understudied. We review animal models that represent analogues of human behavioral phenotypes that are risk markers for internalizing and externalizing problems (behavioral inhibition, behavioral exuberance, irritability), and identify commonalities among the neural mechanisms underlying these behavioral phenotypes and the neural targets of three types of air pollutants (polycyclic aromatic hydrocarbons, traffic-related air pollutants, fine particulate matter < 2.5 microns). We conclude that prenatal exposure to air pollutants increases risk for behavioral inhibition and irritability through distinct mechanisms, including altered dopaminergic signaling and hippocampal morphology, neuroinflammation, and decreased brain-derived neurotrophic factor expression. Future studies should investigate these effects in human longitudinal studies incorporating complex exposure measurement methods, neuroimaging, and behavioral characterization of temperament phenotypes and neurocognitive processing to facilitate efforts aimed at improving long-lasting developmental benefits for children, particularly those living in areas with high levels of exposure.

1. Introduction

Humans are ubiquitously exposed to air pollutants. Although there are many studies documenting deleterious physical health outcomes associated with exposure to air pollution (Cohen et al., 2017; Genc et al., 2012; Landrigan, 2017), a growing number of epidemiologic studies also link exposure to risk for a range of psychiatric and neurodevelopmental outcomes (Khan et al., 2019; Lu, 2020; Volk et al., 2021). Concerningly, exposure during pregnancy to even low levels of air pollution – levels that may have no adverse effects on an adult – may disrupt rapid and foundational fetal brain development, potentially leading to life-long functional impairments (Adams et al., 2000; Barker, 2004; Grandjean & Landrigan, 2014; Landrigan & Goldman, 2011; Lanphear, 2015; Rice & Barone, 2000). Prenatal exposure to air pollution has been linked with internalizing/anxiety symptoms (Loftus et al., 2020; Perera et al., 2011, 2012) and externalizing/ADHD symptoms (Alemany et al., 2018; Loftus et al., 2020; Min & Min, 2017; Newman et al., 2013; Perera et al., 2011, 2012; Siddique et al., 2011; Yorifuji et al., 2017), sometimes within the same participants (Loftus et al., 2020; Perera et al., 2011, 2012). Although most studies of prenatal exposures have focused on psychiatric outcomes in childhood or adolescence, a small and growing literature has linked prenatal exposure to neurotoxicants, including air pollution, with early behavioral phenotypes (Cowell et al., 2019; Gartstein & Skinner, 2018; Stroustrup et al., 2016), such as behavioral inhibition (BI), behavioral exuberance (BE), and irritability, that are known risk factors for later internalizing and externalizing psychopathology (Fox et al., 2021; Hirshfeld-Becker et al., 2002; Sjöwall et al., 2017). These behavioral phenotypes have distinct neural mechanisms (Kircanski et al. 2018) that may be vulnerable to prenatal exposure to air pollution. Although the effects of stress and genetics on these behavioral phenotypes (i.e., behavioral inhibition) and later psychiatric risk are well known (McGrath et al. 2012; Monk, Lugo-Candelas, and Trumpff 2019; Smoller 2016; Wiggins et al. 2014; Pagliaccio et al. 2018), the effects of the chemical environment on these early life markers remains understudied. Importantly, these prenatal exposures represent modifiable risk factors amenable to public health policy that can improve children’s mental health outcomes.

In this review, we explore the premise that prenatal exposure to air pollution may contribute to early emerging behavioral phenotypes that increase risk of internalizing and externalizing problems. We first briefly discuss three human behavioral phenotypes: behavioral inhibition, behavioral exuberance, and irritability and decompose, or back-translate, these phenotypes into constituent behavioral analogs measured in animal experimental paradigms. Table 1 presents examples of the applied methods, paradigms, and measured behaviors that correspond to each human behavioral phenotype. Given that behavioral inhibition and behavioral exuberance are often measured via the same paradigm, we discuss them jointly. Although there is a small literature examining the neural correlates of prenatal exposure to air pollution in humans (Guxens et al., 2018; Mortamais et al., 2019; Peterson et al., 2015), these studies are limited by their observational designs, whereas animal models allow for the experimental manipulation of air pollution types, levels, and timing. Thus, we next review findings from animal models of prenatal exposure to three commonly studied air pollutants (polycyclic aromatic hydrocarbons [PAH], traffic-related air pollutants [TRAP], fine particulate matter < 2.5 microns [PM2.5]) and identify commonalities among the neural targets of air pollution and neural mechanisms underlying the animal analogs of behavioral inhibition, behavioral exuberance, and irritability. Overall, this literature points to environmental contributions to behaviorally inhibited and irritable phenotypes that increase risk for later psychopathology, with less evidence of contributions to a behaviorally exuberant phenotype. We conclude with a discussion of directions for future research.

Table 1.

Examples of animal behaviors analogous to human behavioral phenotypes (BI/BE and irritability) for psychiatric risk

Human behavioral phenotype	Method/paradigm	Model/strain	Measured behavior in animal model	Example References
BI/BE	Novel social/nonsocial arena + juvenile reciprocal social interaction	Rats/Sprague-Dawley (outbred)	Longer latency to approach novelty, lower locomotion in both a novel social and nonsocial arena, more exploration of arena rather than following or sniffing unfamiliar rat (BI) Shorter latency to approach and higher locomotion in both a novel social and nonsocial arena, following, chasing, or sniffing unfamiliar rat (BE)	(Caruso et al., 2014; Cavigelli et al., 2007; Michael et al., 2020; Terranova et al., 1993)
	Novel social/nonsocial arena + open field, light/dark box, elevated plus maze, elevated zero maze	Mice/C57BL6 (inbred)	Longer latency to approach novelty, lower locomotion, less time spent in anxiogenic and interaction parts of the test apparatus (BI) Shorter latency to approach novelty, higher locomotion, more time spent in anxiogenic and interaction parts of the test apparatus (BE)	(Miao et al., 2019)
	Predator exposure paradigm	Rats/Sprague-Dawley (outbred)	Sum of freezing and hypervigilance: absence of locomotion (BI)	(Nanda et al., 2008; Qi et al., 2010)
	Non-spatial cue-based learned approach-avoidance paradigm	Rats/Long-Evans	High avoidance of conflict (BI) High approach of conflict (BE)	(Schumacher et al., 2018)
	Eyeblink conditioning	Rats/Wistar-Kyoto; Sprague Dawley	Wistar-Kyoto rats, an animal model of BI, show facilitated acquisition of the classically conditioned eyeblink response (CCER)	(Janke et al., 2015)
	Ultrasonic Vocalizations (USV)	Rats/Wistar	Increased isolation-induced ultrasonic vocalizations (BI) Decreased isolation-induced ultrasonic vocalizations (BE)	(Wöhr & Schwarting, 2008)
	Selective breeding (open field, light/dark box, elevated plus maze)	Rats/Bred “low responders” & bred “high responders”	Bred low responders (LR); decreased locomotion, increased latency to approach novelty, less time spent in anxiogenic parts of the test apparatus (BI) Bred high responders (HR); increased locomotion, decreased latency to approach novelty, more time spent in the anxiogenic sections (BE)	(Clinton et al., 2011, 2021; Flagel et al., 2010; McCoy et al., 2019; Piazza et al., 1991; Sanna et al., 2015, 2017; Simmons et al., 2012; Thiel et al., 1999)
Irritability	Frustrative non-reward (FNR); omission of expected reward	Fish/Atlanic Salmon Smolts (commercial strain, Aquagen AS)	Aggressive acts like charges, nips and chases elicited by delayed feeding	(Vindas et al., 2014)
	Frustrative non-reward (FNR); omission of expected reward	Mice/C57BL/6J	Increase in operant responding (lever presses/minute)	(Martín-García et al., 2015)
	Frustrative non-reward & threat/opponent confrontation	Rats/Lister, Long-Evans, Sprague-Dawley, Wister	Mice emit USVs in the same frequency in response to both frustrative non-reward and threat	(Knutson et al., 2002)

2. Behavioral inhibition and behavioral exuberance

BI and BE are conceptualized in terms of negative and positive reactivity to novel stimuli (for a full discussion see Perez-Edgar & Fox, 2018), which can be reliably measured even in infancy and predict later BI/BE traits (Dollar et al., 2017; Filippi, Sachs, et al., 2020; Fox et al., 2001; Kagan et al., 1984, 2007). BI is characterized by high negative affect and avoidance behaviors such as fearfulness or wariness when faced with novel people, objects, or challenging situations (Capitanio, 2018; Fox et al., 2005; Kagan et al., 1984). In contrast, BE is characterized by high positive affect and high approach behaviors, lack of fear, impulsivity, and high motor activity (Fox et al., 2001; Tarullo et al., 2011). In addition, BI has been characterized by increased monitoring of the environment and higher levels of reactive control, that is, the tendency to react to an event with corrective behavior (Buzzell et al., 2018). BI/BE positive/negative reactivity phenotypes exist on a continuum, such that children exhibiting neither high negative nor high positive affect are characterized as low reactive and considered normative along the BI-BE continuum (Fox et al., 2001; see Filippi, Sachs, et al., 2020 for a description of a non-inhibited sample). Prior literature documents sex differences in BI; however, the findings are mixed. One set of findings suggests that high negative reactivity in infancy predicted BI in toddlerhood and early childhood only for boys, but not for girls (Fox et al., 2015; Henderson et al., 2001). However, it was also reported that girls are more inhibited than boys at 14, 21, and 52 months (Kagan et al., 1994; Majdandzić & van den Boom, 2007). Boys have also been reported as being more exuberant than girls in early childhood (Gross & Stright, 2019; Majdandzić & van den Boom, 2007; Tarullo et al., 2011).

Longitudinal studies show that BI and BE are risk markers for future psychopathology. BI in childhood is associated with withdrawal and reticence in social environments, and represents one of the most robust early predictors of future anxiety disorders (Frenkel et al., 2015; Rubin et al., 2002) and specifically social anxiety disorder (Buzzell et al., 2017; Fox et al., 2021). BE is associated with adaptive outcomes, including higher levels of social competence and positive affect (Degnan et al., 2011; Dollar et al., 2017; Fox et al., 2001), but also impaired outcomes, including increased anger, frustration, impulsivity, and externalizing problems (Degnan et al., 2011; Morales et al., 2016), attention-deficit/hyperactivity disorder (ADHD; Bunford et al., 2015; Forslund et al., 2016; Salari et al., 2017), disruptive behavior disorder with or without comorbid mood disorder, and conduct disorder (Degnan et al., 2011; Egger & Angold, 2006; Forbes et al., 2017; Hirshfeld-Becker et al., 2002).

The neurobiology of BI/negative reactivity has been reviewed recently (Clauss, 2019; Clauss et al., 2015; Fox et al., 2021). A recent meta-analysis of task-related functional activation found that relative to non-inhibited individuals, those with BI show greater functional activation in the amygdala, basal ganglia, and prefrontal cortex during a range of tasks (Clauss et al., 2015). Structural imaging studies of the neural correlates of BI in children are lacking (Kircanski et al., 2018), but in one study, children with BI had reduced amygdala volume change from age 10 to 12 years relative to those with BE (Filippi, Sachs, et al., 2020). In addition to the amygdala and frontostriatal circuits, some models suggest that the hippocampus may be important in BI (Villard et al., 2021), but that it has been overlooked in research (Clauss et al., 2015). Recent work also points to the possible role of the bed nucleus of the stria terminalis in BI (Clauss, 2019). In addition, electrophysiology studies show that right resting frontal asymmetry appears to be a biomarker of BI (Fox et al. 2001; R. Liu, Calkins, and Bell 2020; Rutherford and Lindell 2011). Relatedly, BE infants were found to exhibit left frontal asymmetry that persisted at age 4 (Fox et al., 2001; Hane et al., 2008). Though research is limited, BE/positive reactivity has similar yet distinct neural correlates when compared to BI.

Animal analogs of BI and BE-related behaviors can be found in two types of studies: experimental designs that elicit inhibited or exuberant behavior or selective breeding of animals at either end of the BI-BE spectrum (see Table 1). To measure animal behavior, which we argue parallels BI and BE, behavioral paradigms typically expose rodents to novel stimuli (e.g., novel social/non-social arena and exposure to predator cues) or use an approach-avoidance paradigm. In these tasks, longer latency to approach, decreased locomotion, avoidance of a conflict arm (presenting simultaneously learned appetitive i.e., sucrose, and aversive i.e., foot shock, cues), increased hypervigilance, and freezing are all characteristic of BI phenotypes (Schumacher et al., 2018; Caruso et al., 2014; Cavigelli et al., 2007; Michael et al., 2020; Nanda et al., 2008; Qi et al., 2010). On the other hand, shorter latency to approach (including to conflict arm versus a neutral arm) and increased locomotion reflect BE-like behavior. In addition, in selective breeding experiments, “low-responder” mice show BI-like behaviors, such as decreased locomotion, increased latency to approach novelty, and less time spent and less entry frequency into anxiogenic regions (i.e., in the middle of an open field or in the open arms of an elevated plus maze). “High-responder” mice show BE-like behavior, manifesting as increased exploration and interaction with novelty, decreased latency to approach novelty, and more time spent in and times entered into anxiogenic areas (Clinton et al., 2011, 2021; Flagel et al., 2010; Simmons et al., 2012; McCoy et al., 2019; Thiel et al., 1999; Giorgi et al., 2019; Klein et al., 2014; Sabariego et al., 2019; Sanna et al., 2015, 2017).

3. Irritability

Irritability is defined as a low threshold for anger in response to frustration. In children, irritability can include a persistently angry/grumpy mood, termed tonic irritability, and/or temper outbursts (physical or verbal aggression), termed phasic irritability (Brotman et al., 2017). Normative in toddlerhood, irritability typically declines from childhood into adolescence (Copeland, Brotman, and Costello 2015; Wiggins et al. 2014; Pagliaccio et al. 2018). Persistent irritability over childhood or into adolescence, however, may be a marker of current or future psychopathology (Pagliaccio et al., 2018) and severe irritability has been conceptualized as Disruptive Mood Dysregulation Disorder in the Diagnostic and Statistical Manual 5^th edition (“Diagnostic and Statistical Manual of Mental Disorders, 5th Edition (DSM-5)” 2021). Findings suggest that children with irritability are at greater risk for future disorders, such as anxiety, depression, ADHD, conduct disorder, and oppositional defiant disorder (Dougherty et al., 2013, 2015; Galera et al., 2021; Vidal-Ribas et al., 2016). Previous research suggests mixed findings regarding sex differences in the prevalence of irritability. Although some researchers have reported higher levels of irritability in adolescent girls (Leibenluft et al., 2006), others failed to find sex differences in irritability (Copeland et al., 2015). The association between irritability and later psychopathology may also vary by sex. For example, irritability was more strongly related to externalizing than to internalizing problems in boys, whereas the association was equally likely in girls (Humphreys et al., 2019). The neurobiology and cognitive mechanisms underlying irritability have also been reviewed recently and include deficient reward learning, elevated sensitivity to receipt or omission of reward, and impaired threat processing and cognitive control (Brotman et al., 2017; Leibenluft, 2017).

In animal models, analogs of irritability have been examined as aberrant emotional and behavioral responding, such as increased motor activity and aggressive behaviors in response to frustrative non-reward (FNR) (Brotman et al., 2017). In FNR paradigms (Table 1) an animal receives repeated reinforcement of a target behavior, e.g., an expectation of a reward is established by associating the reward with a signal, placing it at the end of a runway or maze, or providing it after the push of a lever. This is followed by the omission of the expected reward, which then elicits aggression or increased motor activity and reflects a negative prediction error (de Almeida & Miczek, 2002; Vindas et al., 2014).

4. Measuring air pollutants

Monitoring or estimating human exposures to air pollutants, including PAH, TRAP, and PM, from indoor and outdoor sources can be conducted using personal air sampling, regional monitoring stations, or via computer modeling using combinations of station data, satellite images, and meteorological variables. Portable personal air monitors can measure real-time exposure with high spatial and temporal resolution. Air monitors rely on filters that accumulate particles over a set period of time or measure particles in space through light-based continuous real-time monitoring of particulate exposures. Critical to the utility of these devices is their ability to capture diurnal and seasonal alterations in pollutants as well as the ability to impute values using monitor networks. Satellite imaging-based models continue to prove valuable in air pollution epidemiology (for a contemporary review, see Holloway et al., 2021). Additionally, PAH, TRAP, and PM are broad categories of pollutants made of constituent parts. Over 100 PAHs are known (e.g., benzo[a]pyrene, chrysene, benzo[b]fluoranthene) as are various TRAP components (e.g., black carbon, nitrogen oxides, tire dust, etc.) and PM that includes any particles from a variety of natural and human origins within nanometer size categories.

Studies linking air pollutant exposures to adverse neuropsychiatric outcomes must account for toxicokinetic and toxicodynamic properties of the pollutants and individual differences in susceptibility based on polymorphisms in xenobiotic metabolism genes (e.g., GSTM1; (Yang et al. 2009) and the distinctions between toxicity of parent chemicals and toxic primary and secondary metabolites. For instance, some PAH congeners, such as benzo[a]pyrene, are more toxic as a reactive intermediate metabolite than the original PAH ((+)-Benzo[a]pyrene-7,8-dihydrodiol-9,10-epoxide; Miller & Ramos, 2001). However, distinctions in parent compound vs. metabolite toxicities likely differ when considering genotoxic carcinogenic effects relative to other toxicological influences that might disrupt neurodevelopment. Individual and combined toxicological features of air pollutants, including routes of exposure, metabolism, mechanisms of action, and systemic elimination, as well as exposure science factors, including sampling and analytical approaches, are critical in linking exposures to health outcomes in humans and animal models. The paradigm for exposure in animal models also impacts outcomes. Particulate exposures, including ultrafine particle chambers using particle condensers (e.g., the Harvard ultrafine concentrated ambient particle system; Gupta, Demokritou, & Koutrakis, 2004), and other forms of ambient exposure in rodent models (e.g., housing rats adjacent to air pollutants of interest such as automobile exhaust; Patten et al., 2020) facilitate major strides in resolving mechanisms of neuropsychiatric toxicity of air pollution. Technical reviews of experimental approaches in particulate and inhalation toxicity are provided elsewhere (Costa et al., 2014; Shang & Sun, 2018; Zavala et al., 2020).

5. Convergence of neural targets of air pollution and mechanisms of behavioral phenotypes for psychiatric risk

We highlight three potential areas of overlap between the neural consequences of prenatal air pollution exposure and the neural mechanisms of behavioral phenotypes: 1) changes in dopamine signaling in the prefrontal cortex, dorsal striatum, and ventral striatum (nucleus accumbens); 2) neuroinflammatory responses; and 3) reduced brain-derived neurotrophic factor expression in the hippocampus (Table 2). For each of these mechanisms, we present the exposure-related neural consequences and the overlapping neural mechanism of behavioral phenotypes. When present, we also highlight behavioral effects that are consistent with animal behaviors analogous to human behavioral phenotypes (BI, BE, irritability). When available, we include relevant findings from human neuroimaging studies of BI, BE, and irritability.

Table 2.

Overlap between neural correlates of prenatal exposure to air pollution and behavioral phenotypes

Publication	Model	Air pollutant	Behavioral phenotype (paradigm)	Target of air pollution exposure	Corresponding neural correlate of behavioral phenotype
Disturbances in the dopamine system
(Yokota et al., 2009)	Mice*	TRAP	BI (spontaneous motor activity) DE-exposed mice had decreased spontaneous locomotor activity.	Decreased dopamine turnover in dorsal striatum and nucleus accumbens	BI: rats have decreased dopamine activity in nucleus accumbens and dorsal striatum (Flagel et al., 2010; Sanna et al., 2015, 2017; Thiel et al., 1999)
(Suzuki et al., 2010)	Mice*	TRAP	BI (spontaneous locomotor activity) DE-exposed mice had decreased spontaneous locomotor activity.	Decreased dopamine turnover in dorsal striatum Increased dopamine, dopamine turnover and noradrenaline in PFC; decreased noradrenaline turnover in PFC	BI: rats have decreased dopamine activity in dorsal striatum (Flagel et al., 2010; Sanna et al., 2015, 2017; Thiel et al., 1999), and increased dopamine in PFC (Piazza et al., 1991)
(Gao et al., 2017)	Zebrafish	PAH	BI (novel tank test, T-maze test) Benzo-a-Pyrene-exposed adult fish had decreased locomotor activity, learning, and memory.	Decreased dopamine, DOPAC, and NE in adulthood, and induced loss of dopaminergic neurons that resulted in neurodegeneration. Increased DNA methylation level of DRD4-related sequence, causing significant decrease in DRD4-related sequence mRNA levels in the whole brain.	BI: rats have decreased dopamine activity in nucleus accumbens and dorsal striatum (Flagel et al., 2010; Sanna et al., 2015, 2017; Thiel et al., 1999)
(Yokota et al., 2016)	Mice*	TRAP	Irritability (resident-intruder test) DE-exposed mice showed great social isolation-induced territorial aggression behavior. No significant differences in locomotion.	Increased levels of dopamine and its metabolites in PFC and nucleus accumbens	Irritability: enriched dopamine-DARPP-32 signaling pathway in frontal cortex and ventral striatum genes (Martín-García et al., 2015)
(Cui et al., 2019)	Mice	PM2.5	BE (open field test) PM2.5-exposed mice had increased locomotor and exploratory behavior.	Upregulation of l-dopa (precursor of dopamine) and DRD4 expression levels in the whole brain. DRD4 antagonism ameliorated locomotor activities of mice to levels indistinguishable from controls.	BE: rats have increased dopaminergic activity in dorsal and ventral striatum, including the nucleus accumbens (Piazza et al., 1991; Sanna et al., 2015, 2017; Thiel et al., 1999)
Decrease in hippocampal BDNF expression
(Zhou et al., 2020)	Mice**	PM2.5	No relevant tests.	Reduced expression of BDNF in hippocampus	Irritability: fish have downregulated BDNF in telencephalon (Vindas et al., 2014) BI: rats have decreased hippocampal BDNF (Janke et al., 2015; Miao et al., 2019)
Hippocampal neuroinflammation and structure
(Nephew et al., 2020)	Rats*behavior testing	PM2.5	BI (elevated platform task, predation test, marble burying task) PM2.5-exposed rats had lower levels of whole cage social play and allogrooming as well as increased anxiety-like behaviors.	Decreased expression of proinflammatory interleukin-18 (IL-18) Ex vivo brain MRI analysis revealed disrupted neural integrity in the hippocampus of PM exposed juveniles	BI: individual differences are correlated with the activation of various stressresponsive brain regions, including CA3 region of the hippocampus (Qi et al., 2010).
(Tseng et al., 2019)	Rats	PM2.5	BI (MWM test, novel object and location recognition test. PM2.5-exposed rats had decreased recognition, decreased long-term, working, and spatial memory, and decreased exploratory behavior (though not significantly different).	Exposed rats had decreased dendritic branches and terminals in the CA1 and CA3 of the hippocampus. Increased expression of proinflammatory cytokines (TNF-α, monocyte chemoattractant protein-1 (MCP-1), interleukin 1 beta (IL-1β), interleukin-15 (IL-15), and vascular endothelial growth factor	BI: inactivation of ventral CA1 (Schumacher et al., 2018) Irritability: CA1 area of hippocampus activated (Panagiotaropoulos et al., 2009) BI: heightened proinflammatory responses (glucocorticoid and IL-6) (Michael et al., 2020) Irritability: increased proinflammatory cytokines (Takahashi et al., 2018)
(Woodward et al., 2018)	Mice*	TRAP	No effects on BI, BE or irritability (novel object in context recognition; elevated zero maze test) TRAP-exposed rats had decreased discrimination of novel object but no effect on object exploration. No differences in elevated zero maze test.	Increased levels of pro-inflammatory cytokines, and reduced levels of anti-inflammatory cytokines (IL-4, IL-10, IL-13) Microglia were activated in DG and CA1 subfields (35% more Iba1 [inflammatory marker]).	BI: heightened proinflammatory responses (glucocorticoid and IL-6) (Michael et al., 2020) Irritability: increased proinflammatory cytokines (Takahashi et al., 2018)) BI: inactivation of ventral CA1 (Schumacher et al., 2018)
(Berg et al., 2020; Patten et al., 2020)	Mice†,†	TRAP	Mixed BI/BE in females (pup ultrasonic vocalizations, reciprocal social interaction, open field, novel object recognition, fear conditioning) TRAP-exposed female pups had decreased USV calls, spent more time following/chasing in social interaction, and preferred novel objects (BE), but had decreased exploratory behavior (BI).	Increased microglial infiltration in CA1 (inflammation), reduced astrogliosis in DG, increased hippocampal anti-inflammatory IL-10 in females	BI: inactivation of ventral CA1 produces avoidance (Schumacher et al., 2018)
(Bolton et al., 2013)	Mice†	TRAP	BI (fear conditioning, elevated zero-maze) TRAP-exposed males had increased freezing and decreased exploratory behavior.	Increased proinflammatory interleukin (IL)-1β and decreased anti-inflammatory IL-10 in males	BI: heightened proinflammatory responses (glucocorticoid and IL-6) (Michael et al., 2020) Irritability: increased proinflammatory cytokines (Takahashi et al., 2018)

Note.

^*indicates only male offspring included in study

^**indicates only female offspring included in study

^†indicates sex-specific effect.

5.1. Dopaminergic signaling

In animal models, prenatal exposure to air pollution alters levels of dopamine and its metabolites across several key regions: the prefrontal cortex, the ventral striatum, including the nucleus accumbens, and the dorsal striatum (Table 2; Haber, 2014). These regions belong to parallel frontostriatal loops which are implicated in limbic (e.g., error monitoring and reward processing), associative (e.g., cognitive control), and sensorimotor (e.g., locomotion; Alexander et al., 1986; Botvinick et al., 2004; Shenhav et al., 2016; Tremblay et al., 2015) processes. In the striatum, specifically, there is a ventromedial to dorsolateral gradient, with the former striatal subregions subserving limbic functions, the latter subserving motor functions, and those in between subserving mostly associative functions (Tremblay et al., 2015).

The two extant whole brain analyses exploring the effects of prenatal air pollution on dopaminergic signaling in animal models have yielded equivocal findings. Prenatal exposure to PAH, specifically benzo-[a]-pyrene (BAP), reduced dopamine levels and induced dopaminergic neuron loss across the zebrafish brain (Gao et al., 2017). In contrast, prenatal PM2.5 exposure in mice upregulated a dopamine precursor, l-dopa, and DRD4 receptor expression across the whole brain (Cui et al., 2019). Further study is required to better understand different mechanisms of action associated with these different exposure types and how effects may be localized to specific dopaminergic pathways. Different air pollutants might not exert the same influences in select neurotransmitter systems or in shared brain regions.

Prenatal TRAP exposure more consistently affects dopamine signaling. Two early studies show that prenatal TRAP yields decreased dorsal striatal dopamine turnover (Suzuki et al., 2010; Yokota et al., 2009) and increased levels of dopamine in the prefrontal cortex (Suzuki et al., 2010). In a later study using environmentally relevant levels of exposure (roughly half that of Suzuki et al., 2010), exposed animals continued to show increased dopamine in the prefrontal cortex (Yokota et al., 2016). Notably, equivocal evidence links prenatal air pollution exposure with dopamine and its metabolites in the nucleus accumbens with one study reporting decreased (Yokota et al., 2009) and a second reporting increased levels (Yokota et al., 2016). In sum, prenatal TRAP exposure yielded decreased dorsal striatal dopamine turnover and increased prefrontal dopamine.

The effects of air pollution on dopamine signaling converge with the neural mechanisms of animal behaviors consistent with BI, suggesting that air pollution exposure may increase risk for this behavioral phenotype. For example, low-responder (behaviorally inhibited) rats demonstrate lower striatal dopamine responses to novelty (Thiel et al., 1999) and reduced expression of functionally active (high affinity) D2 receptors in the dorsal striatum (Flagel et al., 2010), which parallel the decreased dorsal striatal dopamine turnover evident in rats prenatally exposed to TRAP (Suzuki et al., 2010; Yokota et al., 2009). Behaviorally, exposure-related striatal or whole brain reductions in dopamine (or in its turnover) are possibly accompanied by decreased spontaneous locomotion (Gao et al., 2017; Suzuki et al., 2010; Yokota et al., 2009) (but see Cui et al., 2019; Yokota et al., 2016), a behavioral pattern characteristic of low-responder rat strains (Clinton et al., 2011, 2021; Flagel et al., 2010; Simmons et al., 2012). Additionally, increased monitoring associated with BI (Buzzell et al., 2018) may be paralleled by increased prefrontal (including the cingulate) dopamine associated with TRAP exposure (Suzuki et al., 2010; Yokota et al., 2009; 2016). We thus propose that prenatal air pollution exposure may increase risk for BI in infancy via disruptions across frontostriatal circuits.

Likewise prenatal air pollution exposure may also increase risk for irritable temperament. Negative prediction errors elicited by FNR are signaled by phasic reductions in striatal dopamine (Chang et al., 2016), and TRAP-induced dopamine signaling disruptions may alter prediction error coding in rats (Suzuki et al., 2010; Yokota et al., 2009). TRAP exposure also increases prefrontal dopamine (Suzuki et al., 2010; Yokota et al., 2016), which is implicated in prediction error processing under state uncertainty (Starkweather et al., 2018) as well as in cognitive control (Ott & Nieder, 2019). Similarly, TRAP-related increases in prefrontal dopamine have been shown to increase aggression (shorter attack latency) in the residential intruder paradigm (Yokota et al., 2016), a behavioral pattern consistent with FNR outcomes and irritability. Given TRAP-related dopaminergic effects, we propose that increased prenatal TRAP exposure may disrupt negative prediction error processing and impair cognitive control, thereby increasing risk for irritability.

5.2. Neuroinflammatory responses and hippocampal structure

Prenatal exposure to PM 2.5 and TRAP results in systemic neuroinflammatory responses. Prenatal PM 2.5 exposure increased serum levels of proinflammatory cytokines (TNF-α, monocyte chemoattractant protein-1 [MCP-1], interleukin 1 beta [IL-1β], IL-15, and vascular endothelial growth factor) (Tseng et al., 2019) but notably, in a separate study, it decreased proinflammatory IL-18 in plasma (Nephew et al., 2020). In addition to increased levels of pro-inflammatory cytokines caused by PM2.5 exposure, reduced serum levels of anti-inflammatory cytokines (IL-4, IL-10, IL-13) were observed in animals exposed to ecologically valid levels of prenatal TRAP, taken from roadways and tunnels (Woodward et al., 2018). Thus, further study is required to resolve some of the conflicting findings of PM2.5- and TRAP-related effects on pro-inflammatory and anti-inflammatory cytokine responses. However, studies generally suggest that exposure causes systemic increased proinflammatory and decreased anti-inflammatory cytokine expression.

One study has specifically examined neuroinflammatory responses across the whole brain and two have examined neuroinflammatory responses in hippocampal subfield regions. Whole brain analysis points to sex-specific effects of prenatal TRAP exposure on increased proinflammatory IL-1β and decreased anti-inflammatory IL-10 in male but not female fetal brains (Bolton et al., 2013). In contrast, female TRAP-exposed offspring showed increased hippocampal anti-inflammatory IL-10 (Patten et al., 2020). Prenatal TRAP also resulted in activated microglia as marked by increased immunofluorescence (Iba1) in the dentate gyrus (DG) polymorphic layer and in the neuropil layer of the CA1 subregion (Woodward et al., 2018) in both males and females; whereas, only female TRAP-exposed offspring showed increased microglial infiltration in the CA1 region, as well as reduced astrogliosis in the DG in a second study (Patten et al., 2020). Notably, prenatal PM 2.5 exposure also resulted in altered hippocampal morphology, including decreases in the number of dendritic branches and terminals in the CA1 and CA3 subregions (Tseng et al., 2019) and disrupted neural integrity in the hippocampus (Nephew et al., 2020). Additionally, TRAP exposure resulted in reduced neurogenesis in the DG (Woodward et al., 2018). Findings linking the neuroinflammatory processes subsequent to prenatal viral infection with altered morphology in the dentate gyrus (Li et al., 2014) suggest that the neuroinflammation secondary to prenatal air pollution exposure may cause TRAP-related morphologic changes in the hippocampus.

The broad neuroinflammatory effects of PM2.5 and TRAP converge with neural mechanisms that underlie behaviors in animal models consistent with BI and irritability. Compared to non-inhibited rats, behaviorally inhibited rats show heightened proinflammatory -- glucocorticoid and IL-6 -- responses to stress (Michael et al., 2020) similar to increases in proinflammatory cytokines caused by prenatal PM2.5 and TRAP exposure. Human studies also show associations between frontal EEG asymmetry, a marker of BI and avoidance (Rutherford & Lindell, 2011), and IL-6 (Shields & Moons, 2016). Increased proinflammatory cytokines are also associated with increased aggressive behavior in animal models (Takahashi et al., 2018) paralleling human irritability. Human findings also point to associations between TNF-related inflammatory cytokine gene expression and neural activity associated with FNR and aggressive behaviors in pediatric bipolar disorder (Barzman et al., 2014).

The effects of prenatal PM2.5 and TRAP exposure on hippocampal structure converge with neural mechanisms that underlie BI and irritability. In one rodent model, ventral CA1 and CA3 subregions had distinct roles in processing approach versus avoidance behavior to conflict (Schumacher et al., 2018). CA1 is implicated in avoidance behaviors and is affected by TRAP and PM2.5 exposure. TRAP- and PM2.5-exposed animals also showed behaviors consistent with BI: lower center time in an open field test (Berg et al., 2020), greater latency to climb down from an elevated platform, increased fecal boli and risk assessment behavior in a cricket predation task, and less whole-cage social play and allogrooming (Nephew et al., 2020). In contrast, prenatally exposed animals showed reduced neonatal pup ultrasonic vocalizations and altered juvenile reciprocal social interactions, and preferred a novel versus familiar object (Berg et al., 2020), which, although traditionally considered a marker of memory, may instead reflect an increased interest in novelty, which is more consistent with BE than BI. Moreover, TRAP exposure altered CA3 morphology, which has been shown to underlie approach behavior (Schumacher et al., 2018). Finally, adults with higher BI, as measured by sensitivity to punishment, show greater hippocampal volumes (Barrós-Loscertales et al., 2006; Cherbuin et al., 2008; Levita et al., 2014). With respect to irritable temperament, the FNR paradigm yielded a higher number of cFos immunopositive cells in the CA1 subregion (Panagiotaropoulos et al., 2009) that is targeted by TRAP and PM2.5. Given these findings, exposure-related effects on CA1 may increase risk for BI and irritable temperament in infancy and effects on CA3 may increase risk for BE.

5.3. Brain-derived neurotrophic factor expression in the hippocampus

Prenatal PM2.5 exposure caused reductions in brain-derived neurotrophic factor (BDNF) expression in the hippocampus (Zhou et al., 2020). Tasks measuring behavioral phenotypes were not investigated. Although there is only a single study documenting effects of prenatal exposure on hippocampal BDNF expression, a model of postnatal exposure showed a similar pattern of results such that PM2.5 exposure on postnatal day 3–15 resulted in decreased hippocampal BDNF expression (Liu et al., 2019). In humans, PAH-DNA adducts, a marker of exposure, and BDNF in umbilical cord blood were compared in two successive birth cohorts after a highly polluting, coal-fired power plant in Tongliang County, China ceased operation. Infants born after the closing of the coal burning factory had decreased PAH-DNA adducts and increased levels of mature BDNF (Tang et al., 2014).

The effects of air pollution on hippocampal BDNF also converge with the neural mechanisms of BI and irritability. For example, hippocampal BDNF is negatively correlated with avoidant behavior (Miao et al., 2019). Facilitated eye-blink, another marker of BI, is related to reductions in hippocampal BDNF (Janke et al., 2015), which also arises from prenatal PM2.5 exposure (Zhou et al., 2020). In addition, postnatal PM2.5 exposure caused reductions in hippocampal BDNF and accompanying anxiety-like symptoms as measured by lower number of entries on open arms in an elevated maze task (Liu et al., 2019). However, reductions in BDNF have been linked with increased spontaneous locomotion (Monteggia et al., 2004; Rios et al., 2001), a back-translated marker of BE, but not in studies of exposure to air pollutants. Despite some conflicting evidence in locomotion measures, given direct observations of BI-like behaviors in postnatally exposed animals, we propose that reductions in hippocampal BDNF may increase risk for BI. With respect to irritability, down-regulated BDNF expression in the telencephalon of Atlantic salmon has also been observed after omission of expected reward (i.e., FNR; Vindas et al., 2014), paralleling effects of exposure on reduced BDNF, although not localized to the hippocampus (fish do not have an anatomical hippocampus). PM2.5 exposure may alter BDNF expression through hypermethylation of the BDNF promoter IV (Zhou et al., 2020), but further study is required to understand the mechanisms underlying PM2.5 exposure, decreased BDNF expression, and behavior in FNR paradigms.

6. Discussion

We have reviewed animal models of prenatal exposure to three types of air pollutants and identified overlapping neural circuits, regions, and processes that are both targets of exposure and putative mechanisms underlying behavioral phenotypes that confer risk for later life psychopathology, specifically internalizing and externalizing problems. Additionally, we discuss animal models that represent analogues of human phenotypes which are predictive of later psychopathology. Although these pollutants have also been shown in animal models to affect other processes in the brain such as decreased cytochrome oxidase activity (Crépeaux et al. 2012), the behavioral phenotypic consequences of such exposures are yet to be explored. The convergence of evidence reviewed herein points to neural and cognitive pathways through which novel exposure-related behavioral phenotypes may emerge. Although much of the evidence converges, there are inconsistencies and many understudied areas.

Across different air pollutant exposure models, evidence points to an increased risk for BI and irritable phenotypes. Prenatal TRAP exposure alters levels of dopamine and its metabolites in regions critical for dopamine signaling. TRAP-related reductions in striatal dopamine turnover and increases in prefrontal dopamine (Suzuki et al., 2010; Yokota et al., 2009, 2016) may increase risk for BI and irritability as evidenced by reduced locomotion (Gao et al., 2017; Suzuki et al., 2010; Yokota et al., 2009) and increased aggression in animals (Yokota et al., 2016), respectively. Further, findings from human studies of prenatal exposure to PAH point to associations between exposure and reduction in caudate volumes (Mortamais et al. 2017), consistent with altered dopamine functioning reported in animal models. Prenatal TRAP and PM2.5 also increase risk for BI and irritability through neuroinflammatory effects across the brain (Nephew et al., 2020; Tseng et al., 2019; Woodward et al., 2018) and through effects in the hippocampus localized in the CA1 subregion (Patten et al., 2020; Woodward et al., 2018), which has been theorized to subserve avoidance behaviors and shows higher cFos reactivity after FNR (Panagiotaropoulos et al., 2009). Moreover, in vitro findings point to links between TRAP exposures, neuroinflammatory processes (phagocytic activation of microglial NADPH oxidase), and selective damage to dopamine cells, including cell death (Block et al., 2004). Finally, PM2.5 exposure increases risk for BI and irritability through reduced BDNF in the hippocampus, which has been previously linked with BI and anxiety (Janke et al., 2015; Miao et al., 2019) and with omission of reward in FNR (Vindas et al., 2014). Less convincing evidence points to potential associations between exposure to these three air pollutants and BE, suggesting that this phenotype may not be pollution-related and pollution-related risk for ADHD may not depend on early-life BE. Further, in some cases, exposures have been shown to confer an advantage such as when cigarette smoking which contains PAH is protective against Parkinson symptoms (Mappin-Kasirer et al. 2020). Thus, we propose that prenatal exposure to these air pollutants increases risk for BI and irritable phenotypes and suggest that future studies investigate these effects in longitudinal designs incorporating complex exposure measurement methods, neuroimaging, and behavioral characterization of temperament and neurocognitive processing from infancy through childhood.

6.1. Pollution-related BI versus irritable phenotypes

In human clinical settings, anxiety has been reported to co-occur with irritability (Cornacchio et al., 2016; Shimshoni et al., 2020), and in animal models FNR has been shown to elicit irritable behavior and anxiety (Cuenya et al., 2012; Taylor et al., 2019). However, in human infants, BI and irritable phenotypes are uncorrelated and have distinct neural underpinnings (Filippi, Subar, et al., 2020). Our review suggests that in animal models, prenatal exposure to air pollution increases risk for each phenotype via effects on their neural correlates. Future studies could examine why some infants manifest one versus the other phenotype and how this contributes to risk for later psychopathology. Yet unidentified factors, including genetic vulnerability to effects of air pollution, may moderate effects of exposure.

6.2. Proposed cognitive mechanisms underlying exposure-related BI and irritability

Cognitive pathways to BI and irritability arise via altered cognitive control, specifically error monitoring (Buzzell et al., 2018), and disrupted prediction error processing (Chang et al., 2016). Increased prefrontal (including in the cingulate) dopamine associated with TRAP exposure (Suzuki et al., 2010; Yokota et al., 2009) may cause increased environmental monitoring which is associated with BI (Buzzell et al., 2018). TRAP-induced decreases in striatal dopamine turnover (Suzuki et al., 2010; Yokota et al., 2009) may lead to abnormal negative prediction error processing associated with FNR (Chang et al., 2016). Human studies have shown that irritability may derive from impaired prediction error processing during reward learning (Brotman et al., 2017; Meyers et al., 2017). Future studies may be able to better understand effects of pollution on irritability using computational models, such as Q learning. For example, since prenatal TRAP exposure yielded decreased dorsal striatal dopamine turnover and increased prefrontal dopamine, we would expect to see decreased learning from positive prediction errors and increased learning from negative prediction errors.

The role of prenatal or concurrent exposure to air pollution on error monitoring and reinforcement learning has yet to be explored. Future studies can examine the impact of exposure both in animal models and in human cognitive and neuroimaging paradigms. In animal models, equivocal evidence links prenatal air pollution exposure to dopamine levels in the nucleus accumbens (Yokota et al., 2009, 2016) and whole brain dopamine (Cui et al., 2019; Gao et al., 2017). Human neuroimaging studies can now study effects of PAH exposure on negative prediction errors using model-based fMRI reinforcement-learning tasks (O’Doherty et al., 2007; Maia & Frank, 2011; Series, 2020).

6.3. Prenatal exposure to air pollution magnifies risk pathways in developmental models of anxiety

Air pollution exposure has been associated with increased risk for internalizing disorders and anxiety (Volk et al., 2021), as well as reductions in working memory (Rivas et al., 2019), attention (Chiu et al., 2016; Cowell et al., 2015), self-regulation (Margolis et al., 2016), and inhibitory control capacities (Margolis et al., 2021), which modulate the pathway from BI in young children to anxiety symptoms and disorders in older children (Fox et al., 2021; White et al., 2011). Importantly, these executive capacities are modulated by dopamine signaling (Ott & Nieder, 2019), which also is associated with anxiety symptoms, genetic risk, and treatment response, and as such has been hypothesized to play a critical role in the interaction between BI and inhibitory control on the manifestation of anxiety (Gunther & Pérez-Edgar, 2021). We propose that prenatal exposure to air pollution disrupts dopamine signaling thus contributing to BI and reductions in cognitive control which ultimately increases risk for anxiety. Human imaging studies of dopamine function in individuals with BI have been challenged by the lack of noninvasive methods for imaging dopamine, but studies of dopamine receptor gene expression and pharmacologic manipulation of dopamine levels suggest that dopamine plays a role in social anxiety symptoms (Cervenka et al., 2012; Hood et al., 2010; Schneier et al., 2000). Innovative noninvasive magnetic resonance imaging pulse sequences now allow for visualization of neuromelanin, a catabolite of dopamine, which binds to iron and is thus available for imaging in the substantia nigra and the ventral tegmental area (Sulzer et al., 2018), opening new avenues for the study of the dopaminergic effects of prenatal exposure to air pollution in humans.

6.4. Chemical and social stressors interact

Recent models have investigated the interacting effects of early life exposure to neurotoxicants including social stress and chemical exposures such as air pollution on neurodevelopment (Padula et al., 2020). For example, prenatal exposure to PAH moderated associations between early life stress and attention and thought problems (Pagliaccio et al., 2020). The mechanisms underlying such synergy remain understudied. We propose that effects of prenatal exposure to air pollution on hippocampal BDNF can increase vulnerability to stress, given that higher levels of hippocampal BDNF buffer the deleterious effects of stress (Taliaz et al., 2011). Thus, pollution-related reductions in hippocampal BDNF may provide one mechanism by which air pollution and stress have synergistic effects on behavioral and psychiatric outcomes. In addition, stress has significant effects on dopaminergic signaling, such that dopamine is released in the striatum in response to noxious stressors (Belujon and Grace 2015; Carvalheiro et al. 2021), which can further potentiate the deleterious effects of air pollution on brain function. Importantly, developmental models should consider mixtures of chemical and social exposures to identify modifiable risk factors.

6.5. Sex-specific effects of prenatal exposure to air pollution

Studies of the outcomes associated with prenatal exposures in animal models have primarily focused on male offspring. However, it is evident that there are sex-specific effects and sex differences in outcomes associated with prenatal exposure to air pollution. In females, prenatal TRAP-exposure resulted in increased microglial infiltration in the CA1, reduced astrogliosis in the DG, and increased hippocampal anti-inflammatory cytokines (Patten et al., 2020). In males, prenatal TRAP-exposure resulted in increased neurogenesis in the DG (Patten et al., 2020) as well as increased proinflammatory but decreased anti-inflammatory cytokines (Bolton et al., 2013). In both studies, prenatal TRAP-exposure resulted in sex-specific effects on behaviors. Exposed female mice exhibited BE-like behaviors including decreased USV calls, more time following an unfamiliar mouse, and preference for a novel object, but also exhibited BI-like behavior such as decreased exploratory behavior in an open field task (Berg et al., 2020). Affected male mice exhibited BI with increased freezing and decreased exploratory behavior compared to females (Bolton et al., 2013). High levels of IL-10 are protective against behavioral changes due to microglial-driven neuroinflammation (Schwarz et al., 2011), which may explain why the male mice with decreased IL-10 were more vulnerable to BI-like impairments compared to the females with normative IL-10 levels. In contrast, exposed females showed increased IL-10 (Patten et al., 2020) and increases in BE-like behavior (Berg et al., 2020), suggesting a behavioral continuum for effects of IL-10. Sex-specific effects of air pollution on neurodevelopment have also been exhibited in humans; associations between prenatal PM2.5 exposure and adverse memory performance have been reported in females, but associations between exposure and lower IQ and poor attention have been reported in males (Chiu et al., 2016). Additionally, some studies have identified sex-specific effects of prenatal air pollutant exposure on child executive functions (e.g., attention, memory) that contribute to self-regulatory capacity in boys (Chiu et al., 2013, 2016; Cowell et al., 2015; Rivas et al., 2019). These findings suggest that increased prenatal air pollution exposure may have sex-specific behavioral and cognitive effects and underscore the need for further study of these sex-specific effects in human and animal studies. When evaluating the effects of air pollution exposure on females, it is particularly critical to evaluate the contribution of the estrous cycle. Studies have demonstrated that high-estrogen phases are associated with less depression and anxiety-like behavior relative to low-estrogen phases (Jaric et al., 2019; Marcondes et al., 2001; Walf & Frye, 2006). Therefore, failure to observe sex-specific effects may reflect natural hormonal variation due to the estrous cycle that was not considered in the design or analysis.

6.6. Conflicting findings from animal models: Proposed systems to explore

Much remains unknown about the effects of prenatal exposure to air pollution. There is an urgent need for future studies to examine whole brain processes across mechanisms rather than testing only for hypothesized or likely areas. Whole brain analyses of exposure have been rare. Only one study has examined whole brain effects of exposure-related neuroinflammatory processes (Bolton et al., 2013). Further study is required to resolve conflicting findings of sex-specific effects of PM2.5 and TRAP on pro-inflammatory and anti-inflammatory cytokine responses. In some cases, extant studies were conducted in different species making comparisons challenging. For example, prenatal PAH exposure reduced dopamine levels and induced dopaminergic neuron loss across the zebrafish brain (Gao et al., 2017), but in a mouse model prenatal PM2.5 exposure upregulated a dopamine precursor, l-dopa, and DRD4 receptor expression across the brain (Cui et al., 2019). Similar challenges exist in the human neuroimaging literature. For example, much work has focused on the role of the amygdala in BI and anxiety with a general lack of exploration of the role of the hippocampus (Clauss et al., 2015), although recent work in non-human primates also points to hippocampal involvement in BI (Villard et al., 2021). A recent study in humans documents links between prenatal exposure to PAH and reductions in hippocampal volumes (Lubczyńska et al. 2021). Whole brain analyses require large samples to achieve adequate power, and prenatal exposure studies require prospective birth cohorts or retrospective biomarkers, such as can be accomplished with dentine (Petrick et al., 2020). Large sample sizes with decades of prospective data are challenging to find; however, some studies are acquiring appropriate samples such as in the Environmental influences on Children’s health Outcomes (ECHO) study or the Columbia Center for Children’s Environmental Health studies.

Studies that examine particular brain regions based on prior findings or outcomes of interest limit what we know. Alternate pathways from prenatal air pollution exposure to behavioral phenotypes for psychiatric risk through neuroinflammation likely exist; however, we have not discussed such findings because direct evidence is lacking. For example, links between air pollution and neuroinflammation (Bolton et al., 2013; Nephew et al., 2020; Patten et al., 2020; Tseng et al., 2019; Woodward et al., 2018), neuroinflammation and central amygdala morphology (Althammer et al., 2020), central amygdala and BI (Kalin et al., 2016; Kalin & Shelton, 1989; Rogers et al., 2013) have all been separately shown; however, experiments have not yet examined the effects of air pollution on central amygdala structure. Future studies must profile effects of exposure comprehensively, rather than focusing on specific endpoints of interest based on prior findings or specific laboratory expertise.

Conflicting findings using spontaneous locomotion as a measure of the effects of air pollution on BI and BE suggest that further study is required. For example, reduced BDNF has been associated with both increased (Monteggia et al., 2004; Rios et al., 2001) and decreased (Liu et al., 2019) locomotion. Similarly, prenatal TRAP exposure caused reduced striatal dopamine and reduced spontaneous locomotion in one study (Yokota et al., 2009) but not in a second study (Yokota et al., 2016). TRAP exposed animals also preferred a novel versus familiar object (Berg et al., 2020), possibly pointing to increased BE with TRAP exposure. Finally, TRAP exposure altered CA1 morphology (Tseng et al., 2019), which is thought to underlie avoid behavior, but it also affected CA3 morphology which is thought to underlie approach behavior (Schumacher et al., 2018). Notably, some rodent and human studies document animals and infants who show features of both BI and BE, depending on the context. For example, rodents display both BI and BE in social and non-social experimental contexts (Cavigelli et al., 2007; Michael et al., 2020). Similarly, a child with BI is more likely to feel distress and anxiety upon encountering a new social environment and simultaneously be driven to interact with others to lessen these feelings (Degnan & Fox, 2007; Tarullo et al., 2011). This motivation is associated with left frontal asymmetry which typically indicates approach tendencies and mitigates the relation between BI and social wariness (Fox et al., 2001; Harmon-Jones & Gable, 2018). Furthermore, shy (BI) individuals have enhanced reward sensitivity and respond faster to incentives than non-shy individuals, reflecting increased approach behaviors in particular contexts (Barker et al., 2019; Hardin et al., 2006).

We also encourage basic studies to look at specific behavioral tasks. For example, studies examining effects of exposure on reinforcement learning and specifically on negative prediction errors are needed to understand mechanisms through which exposure yields altered outcomes in the FNR paradigm. Tasks that are designed and refined to define the phenotype will enhance our understanding of the neural targets of air pollution. Finally, well designed animal models can identify underlying biological mechanisms through which prenatal exposure to air pollutants affect offspring brain development; recent work suggests that exposure may act on gut microbiota (Liu et al. 2020) or maternal immune activation (Volk et al. 2020) potentially via effects on microglia which have been implicated in psychiatric illness (Frick and Pittenger 2016; Tay et al. 2017; Rahimian et al. 2021). More work is required to understand how these processes affect humans and thus may be prevented.

6.7. Measuring Air Pollution

The exposures in the studies reviewed herein can be measured in different ways in human epidemiologic studies (see Volk et al., 2021 for an in-depth description). PAHs are typically measured through personal exposure, for example a backpack with a filter that captures chemicals for analysis. PM2.5 and TRAP are more frequently measured with geocoded data based on a participant’s address and satellite data regarding air quality, or distance from a roadway or traffic on one’s street of residence. Studies using geocoded rather than personal exposure measures may be more at risk of conflating more general neighborhood level effects with effects of neurotoxicant exposure. Future studies could aim to disentangle these issues by collecting personal measures of neighborhood level variables to use in conjunction with geocoded data.

7. Conclusion

Human studies show associations between prenatal exposure to air pollution and internalizing and externalizing behaviors as well as cognitive problems that act as effect modifiers of associations between BI/BE and internalizing and externalizing behaviors. Our review of animal models of both the neural targets of exposure and the neural mechanisms of these behavioral phenotypes suggests that these effects may arise from several distinct mechanisms. One may operate through hippocampal effects, specifically effects of prenatal air pollution exposure on hippocampal morphology, inflammation, and BDNF expression, suggesting that the role of the hippocampus in BI, irritability, and anxiety should be further investigated, as others have recently argued (Clauss et al., 2015; Villard et al., 2021). A second putative mechanism is through dopamine and prediction error processing which may increase the likelihood of BI and irritable phenotypes. Even small effect sizes at the individual level translate to costly consequences at the population level making the effects of prenatal exposure to air pollution a significant public health problem with respect to poor developmental outcomes. These findings taken together point to the need to study BI and irritable phenotypes associated with prenatal exposure to air pollution and to search for appropriate treatment targets that may have long-lasting developmental benefits for children, particularly those living in areas with high levels of exposure.

Figure 1. Convergence of evidence from animal models of prenatal exposure to air pollution and altered behavioral phenotypes with developmental models of human psychiatric risk.

The gradient bar illustrates how the studies reviewed herein cover animal models (red) and human studies (blue) and where these overlap conceptually (cross-species phenotypes; shown in purple). Light blue arrows from “exposure” to “neural correlates” and dark blue lines from “exposure” to analogous – back-translated – ”animal phenotypes” show potential mechanisms of exposure on human psychiatric risk. Solid versus dashed lines indicate more versus less persuasive evidence (Table 2). Observational studies linking human prenatal exposure to air pollution and alterations in brain structure are described in the text (see Discussion). Animal phenotypes akin to behavioral inhibition (BI) and behavioral exuberance (BE) were assessed in animal models of anxiety (elevated plus maze test, open field test, etc. see Table 1), whereas those akin to irritability were assessed using frustrative non-reward tasks (Table 1). Arrows from “exposure” to “neural correlates” to “animal phenotypes” and from “human phenotypes” to “human psychiatric risk” represent the hypothesized causal model positing that exposure increases psychiatric risk via effects on the brain and behavioral reactivity. ADHD = Attention Deficit hyperactivity Disorder; ODD=Oppositional Defiant Disorder.

Highlights.

• Air pollution alters dopamine, BDNF, and hippocampal structure in animal models.

• Neural mechanisms of risk markers of psychopathology are vulnerable to exposure.

• Exposure may increase risk for social anxiety and irritability via these mechanisms.