Neuroimaging markers of risk and pathways to resilience in autism spectrum disorders

Abstract

Autism spectrum disorder (ASD) is a complex, heterogeneous neurodevelopmental condition of largely unknown etiology. This heterogeneity of symptom presentation, combined with high rates of comorbidity with other developmental disorders and a lack of reliable biomarkers, makes diagnosing and evaluating life outcomes for individuals with ASD a challenge. We review the growing literature on neuroimaging-based biomarkers of risk for the development of autism and explore evidence for resilience in some autistic individuals. The current literature suggests that neuroimaging during early infancy, in combination with pre-birth and early genetic studies are promising tools for identifying biomarkers of risk, while studies of gene expression and DNA methylation have provided some key insights into mechanisms of resilience. With genetics and the environment contributing to both risk for the development of ASD and conditions for resilience, additional studies are needed to understand how risk and resilience interact mechanistically, whereby factors of risk may engender conditions for adaptation. Future studies should prioritize longitudinal designs in global cohorts, with the involvement of the autism community as partners in research to help identify domains of functioning that hold value and importance to the community.

1. Heterogeneity, risk and resilience in autism

Standardized criteria for the diagnosis of autism spectrum disorder (ASD) specify persistent deficits in social communication and interaction, and restricted, repetitive behaviors and interests (1,85). A diagnosis of autism is still determined based on clinician observation in combination with a detailed developmental history (2) rather than biomarker-based characterization (3). Contributors to heterogeneity in autism include diversity in phenotypic presentation, which overlaps with other conditions and the general population, and a broad spectrum of symptom severity leading to a range of prognoses (4–6).

Genetic and environmental studies have provided evidence that both genetic and nongenetic factors contribute to an increased susceptibility to developing autism (7–9). Nongenetic factors that increase the risk of autism include epigenetic and environmental factors (10), although the relative contribution of these factors is still not well understood (11,12). Furthermore, interactions between genetic and nongenetic factors may also contribute to autism risk in complex ways (13).

In this review, we will use the terms ‘autism’, ‘autistic’, ‘ASD’, and ‘persons on the autism spectrum’ interchangeably, out of respect for self-advocates (2,14,15) while maintaining consistency with existing terminology in the neuroimaging literature and the nomenclature in the DSM-V (1). Recently, many autism self-advocates have adopted the framework of neurodiversity, seeing autism as an example of diversity in a wide range of brains, none of which are considered ‘normal’ and all of which are different (14,16). The ‘disorder’ and ‘neurodiversity’ views both agree that autism is neurodevelopmental, in that it is characterized by variation in brain function and development (17). However, research has yet to uncover robust and clinically meaningful biomarkers of autism, including neuroimaging-based markers (18).

Resilience is a dynamic process of adaptability in the face of hardship or adversity, that enables adaptation to the context and a better-than-expected outcome. Resilient outcomes do not have to mean supernormal functioning, but rather that the individual is able to maintain or regain a good outcome despite significant adversity (17,19–21). This positive adaptation arises from a combination of biological and psychosocial factors that may be specific to the individual (such as IQ, personality traits), the community (family, school), or the culture (customs and expectations) and includes anything that can favorably impact a person’s adaptational systems. Factors promoting resilience may also be specific to a life-stage, or present across the lifespan whereby the restorative effect is sometimes seen years after the initial stressor (22,23).

The construct of resilience in psychology emerged out of the adaptive outcomes observed among persons with schizophrenia beginning in the 1970’s, and was then expanded to include various adversities, such as adverse childhood experiences, catastrophic life events and chronic illness (for a review, see (24)). Decades of research have identified a range of protective factors, also called ‘assets’ or ‘resources’ that contribute to resilience across different contexts of adversity, including: i) higher cognitive abilities or lack of comorbid intellectual disability; ii) higher socioeconomic status; iii) the presence of supportive adults, and iv) access to intensive, high quality interventions (21,22,24–26). There are also factors that impact a positive outcome more specifically in research on ASD, such as v) early diagnosis; vi) higher initial functioning prior to treatment, especially early language function; vii) being female; and viii) being an unaffected sibling (27–30).

Researchers have proposed various markers of resilience, including the achievement of developmental milestones, cultural competencies, or the absence of psychopathology (21). However, it is important to consider that resilience is dynamic, and positive adaptation in one domain does not necessarily translate to another domain, and depending on the measurement criteria used, resilient functioning may not be captured by researchers (24). This highlights the importance for researchers to work with the autism community to help define which domains of resilience are relevant to the community and would have a valuable impact on quality of life (20). In fact, the Interagency Autism Coordinating Committee (IACC) has called for a ‘paradigm shift in how we approach autism’ towards research that will have a more immediate and direct impact on the daily lives of autistic people and their families (31).

In this review, we highlight how neuroimaging studies of infants and children have provided potential biomarkers of risk for the development of autism and recent genetic and epigenetic studies on the oxytocin receptor gene (OXTR), as a possible biomarker of resilience. Considering the dynamic relationship between risk and protective factors, we discuss how multidisciplinary approaches can contribute to parsing heterogeneity in the autism phenotype and elucidating mechanisms of resilience. In order to improve diagnostic biomarkers for autism and trajectories of resilience throughout the lifecourse, we propose that longitudinal, dimensional research is needed, that considers both biological and psychosocial factors that may be common to mechanisms of both risk and resilience (Figure 1).

Figure 1. Risk factors for development of autism and factors affecting resilience across the lifespan.

In the prenatal period, genetic and epigenetic factors influence risk for development of ASD. Infant sibling studies examining the first years after birth have identified features including brain structure, connectivity and function that may serve as candidate biomarkers of risk for autism. The identification of early genetic and neural risk factors is critical for ensuring accurate early diagnosis, which translates to the development of therapy and intervention strategies targeting the early life period. As children age, family involvement and the school environment become critical to resilience during childhood and adolescence. Community support and access to clinical resources become increasingly important for optimal outcomes for autistic individuals during the transition to adulthood and in later life.

2. Developmental neuroimaging studies reveal neural markers of risk

In the current system of diagnosis based on developmental history and clinician observation, children may not be diagnosed with ASD until the age of 18 to 36 months (32), and may face sociodemographic, geographic, cultural and other barriers delaying intervention, despite evidence of early interventions leading to better outcomes (33,34). While overt behavioral signs of autism are rarely observable in the first year, risk of autism appears to affect a range of systems in the developing brain within this early period (13,35,36). Non-invasive neuroimaging techniques, including magnetic resonance imaging (MRI) and electroencephalography (EEG) are useful tools that may be used to identify early biomarkers for the development of ASD, predict symptom severity, and parse heterogeneity within the ASD phenotype to guide intervention and treatment decisions (10,37).

Table 1 lists studies published in the last decade that have used MRI or EEG to investigate infants at risk for autism (38–75). We draw on this literature to highlight key findings.

Table 1. Neuroimaging studies of infants at risk for autism.

Studies were identified through systematic searches in PubMed and Google Scholar with the key words ‘autism’ OR ‘ASD’ AND ‘infancy’ OR ‘infant’ AND ‘risk for ASD’ OR ‘sibling’ OR ‘risk’ AND ‘MRI’ OR ‘resting state’ OR ‘task’ OR ‘DTI’ OR ‘fMRI’ OR ‘connectivity’ OR ‘EEG’. As revealed by our literature review summarized here, no neuroimaging studies assessing risk or predicting outcomes for autism over the past decade have been conducted in countries other than the United States or the United Kingdom, and even in those studies, race or ethnicity of participants was not reported in most cases. Consequently, almost nothing is known about the influence of race and culture on the lifespan trajectories of autistic individuals (2,126–128).

Study	Number of subjects (HR, LR)	Age range (months)	Neuroimaging method	Biomarker feature	Outcomes predicted	Country of data collection/Race or ethnicity of subjects (N)	Longitudinal or cross-sectional
Blasi et al., 2015 (62)	33(15, 18)	4–7	Task-based fMRI	Brain responses and sensitivity to human auditory stimuli	Language impairments; social outcomes	UK/not reported	Cross-sectional
Bosl et al., 2011 (57)	79(46, 33)	6–24	EEG	mMSE computed from resting state EEG signals	Risk for ASD	US/not reported	Longitudinal* and Cross-sectional
Brito et al., 2019 (56)	129(not reported)	12 h-36	EEG	Neonatal EEG	Risk for atypical behaviors	US/not reported	Longitudinal**
Ciarrusta et al., 2019 (54)	36(18, 18)	1–52 d	fMRI	Synchronous neural activity in the right fusiform and left parietal cortex; anterior segment of the left insula and cingulate cortices	Social processing	UK/not reported	Cross-sectional
Ciarrusta et al., 2020 (101)	40(20, 20)	39–44 PMA w	fcMRI	Short- and long- range functional connectivity	ASD risk	UK/not reported	Cross-sectional
Elison etal., 2013 (44)	97(56, 41)	7–25	DTI	Visual orienting and white matter organization in fiber tracts of the corticospinal pathways and splenium and genu of the CC	LR infants	US/not reported	Longitudinal
Elsabbagh et al., 2011 (47)	16(16, 0)	9–40	EEG	ERP responses to static faces with either a direct or averted gaze	Social outcomes; stereotyped behaviors and restricted interests severity	UK/not reported	Longitudinal**
Elsabbagh et al., 2013 (63)	104(54, 50)	6–36	EEG	ERP responses to dynamic eye gaze shifts	ASD diagnosis	UK/not reported	Longitudinal**
Emerson et al., 2017 (91)	59(11, 48)a	6–24	fcMRI	Functional connectivity	ASD diagnosis; ASD symptom severity	US/not reported	Longitudinal**
Fingher et al., 2017 (42)	97(68, 29)	13–51	DTI	Diffusion measures in the temporal CC tract	ASD symptom severity	US/not reported	Longitudinal**
Garbard- Durnam et al., 2015 (54)	108 (57, 51)	6–18	EEG	Frontal EEG alpha asymmetry	ASD risk	US/not reported	Longitudinal*
Garbard- Durnam et al., 2019 (55)	171(102, 69)	3–36	EEG	EEG power dynamics	ASD diagnosis	US/not reported	Longitudinal*
Hazlett et al., 2011 (48)	Time point 1: 97(59b, 38); Time point 2: 59(38b, 21)	18–59	sMRI	Cerebral gray and white matter volumes and cortical thickness	ASD diagnosis	US/not reported	Longitudinal*
Hazlett et al., 2012 (64)	134(98, 36)	6	sMRI	Head circumference; brain volume	No differences in head circumferen ce or brain volume in 6 month old HR or LR infants	US/not reported	Cross-sectional
Hazlett et al., 2017 (65)	435(318, 117) [Neuroima ged: 148(106, 42)]	6–24	sMRI	Brain growth rate, total brain volume and surface area	ASD diagnosis; social outcomes	US/Non-white: [HR/LR: 53], White: [HR/LR: 376], Not reported: [HR/LR: 6]; [Neuroima ged: (Not reported)]	Longitudinal*
Jones et al., 2016 (41)	88(43, 45)	6–24	EEG	EEG responses to social stimuli	Social outcomes	US/not reported	Longitudinal**
Kolesnik et al., 2019 (40)	143(116, 27) [EEG analyzed: 94(80, 14)]	8–36	EEG	Cortical ‘hyperreactivity’ in response to a non- linguistic auditory oddball paradigm	ASD diagnosis; language outcomes; social outcomes	UK/not reported	Longitudinal**
Levin et al., 2017 (39)	48(29,19)	3–36	EEG	Frontal spectral power	Expressive Language outcomes	US/not reported	Longitudinal*
Lewis et al., 2014 (49)	136(113, 23)	24	DTI	Network efficiency	ASD Symptom severity	US/not reported	Cross-sectional
Lewis et al., 2017 (66)	260(not reported); [Longitudinally neuroimag ed: 116(81, 35)]	6–24	DTI	Network efficiency	ASD symptom severity	US/not reported	Longitudinal*
Liu et al., 2019 (67)	34(19, 15)	6 w-36	DTI	White matter pathways of the dorsal language network	Language impairments; ASD symptom severity	US/Non-white: [HR/LR: 24], White: [HR/LR: 10]	Longitudinal**
Lombardo et al., 2015 (68)	103(60b, 24, 19)c	12–48	Task-based fMRI	Superior temporal cortices	Language Impairments	US/not reported	Longitudinal**
McKinnon et al., 2019 (69)	12 mos: 118(87, 31); 24 mos: 87(67, 20)	12–24	fcMRI	Functional connectivity	Restricted and repetitive behaviors	US/not reported	Longitudinal*
Orekhova et al., 2014 (56)	54(28, 26)	14–36	EEG	Brain connectivity assessed using dbWPLI	ASD diagnosis; restricted and repetitive behaviors	UK/not reported	Longitudinal**
Pote et al., 2019 (70)	50(24, 26)	4–36	sMRI	Whole brain, Cerebellar and Subcortical volumes	Repetitive behaviors	UK/not reported	Longitudinal**
Righi et al., 2014 (57)	54(28, 26)	6–12	EEG	Functional connectivity in response to speech sounds	ASD diagnosis	US/not reported	Longitudinal*
Schumann et al., 2010 (58)	118(74b, 44) [Neuroima ged: 88(41b, 44)]	18–60	sMRI	Brain volume growth (gray matter and white matter)	ASD diagnosis	US/not reported	Longitudinal*
Seery et al., 2013 (50)	108(62, 46)	6–36		ERP responses to native and non-native speech	ASD diagnosis	US/not reported	Cross-sectional and longitudinal*
Seery et al., 2014 (52)	80(35, 45)	6–18	EEG	ERP responses to repeated speech	Risk for ASD; language outcomes	US/[% Nonwhite (HR: 2.9, LR: 13.6)]	Longitudinal**
Shen et al., 2013 (71)	64(41, 23)	6–36	sMRI	Extra-axial CSF volume	ASD diagnosis; ASD symptom severity	US/not reported	Longitudinal*
Shen et al., 2017 (72)	343(221, 122)	6–24	sMRI	Extra-axial CSF volume	ASD diagnosis; ASD symptom severity	US/not reported	Longitudinal*
Shen et al., 2018 (73)	236(132, 27, 77)d	24–42	sMRI	Extra-axial CSF volume	ASD diagnosis; sleep disturbances; non-verbal ability	US/not reported	Cross-sectional
Shephard et al., 2020 (59)	77(42, 35)	7 −7 y	EEG	ERP responses to faces and visual noise stimuli	HR ASD; Social- communicati on	UK/not reported	Longitudinal**
Swanson et al., 2017 (60)	525(382, 143) [Neuroimaged: 368(264, 104)]	6–24	sMRI	Subcortical volumes	Language impairments	US/White [82,80,83,8 1e]; African American: [1,5,2,5e]; one race: [14,10,10,1 1e]; not answered: [3,5,4,2e]Asian: [0,0,1,1e]; more than	Longitudinal**
Tierney et al., 2012 (38)	122(65, 57)	6–24	EEG	Frontal spectral power	ASD diagnosis	US/not reported	Longitudinal*
Wilkinson et al., 2019 (51)	101(58, 43)	24–36	EEG	Resting frontal gamma power	HR ASD; Language impairments	US/not reported	Longitudinal*
Wolff et al., 2012 (74)	92(28, 64)a	6–24	DTI	White matter tract connectivity	ASD diagnosis	US/not reported	Longitudinal*
Wolff et al., 2015 (53)	378(270, 108)	6–24	DTI	CC area and thickness	ASD diagnosis	US/not reported	Longitudinal*
Wolff et al., 2017 (75)	217(44, 173)a	6–24	DTI	Genu and cerebellar white matter connectivity	Repetitive behaviors; sensory response	US/not reported	Longitudinal*

^aHR+, HR−

^bP-ASD

^cASD, TD, LD/DD

^dNormal Risk, High Risk, TD

^eHR-ASD, HR-LD, HR-Neg, LR-Neg (%)

^*Included neuroimaging and behavior longitudinally

^**Included neuroimaging at a single time point, behavior longitudinally

ASD, Autism Spectrum Disorder; CC, corpus callosum; CSF, Cerebrospinal Fluid; d, days; dbWPLI, debiased weighted phase lag index; DD, developmentally delayed; DTI, diffusion tensor imaging; EEG, electroencephalogram; ERP, event-related potentials; fcMRI, functional connectivity magnetic resonance imaging; fMRI, functional magnetic resonance imaging; h, hours; HR, infants with an older sibling with ASD; HR+, infants with an older sibling with ASD and were later diagnosed with ASD; HR−, infants with an older sibling with ASD but were not later diagnosed with ASD; LD, language delay; mMSE, modified multiscale entropy; Normal Risk, came from a Simplex family in which they were the only child with ASD; p-ASD, provisionally diagnosed with ASD- at risk for ASD based on behavioral symptoms; PMA, post menstrual age; sMRI, structural magnetic resonance imaging; TD, typically developing; UK, United Kingdom; US, United States; w, weeks; y, years.

Brain Function

Neuroimaging studies of brain function have successfully differentiated high risk (HR) infants from those at low risk (LR). In the studies reviewed here, infants who have an older sibling diagnosed with autism are considered HR, and those who do not have a diagnosed older sibling are LR. As neuroimaging techniques are sensitive to motion artifacts, natural sleep paradigms permit researchers to conduct task-based fMRI studies in infants as young as 4–7 months and also to study more heterogeneous samples (62,76,77).

Deficits in communication and language are one of the hallmarks of ASD, thus, tasks designed to invoke language processing areas have been used as potential early markers for autism (68,76,78,79). In an fMRI study of brain responses to speech, autistic toddlers with subsequent poor language outcomes showed hypoactivation of the superior temporal cortex, an area known to be activated during language perception. Autistic toddlers who subsequently had good language outcomes showed similar brain responses to non-autistic comparison groups, robustly recruiting this brain area (68). While language trajectories in autism during the first years of life are highly unstable, identifying early language biomarkers represents a potentially meaningful way to parse phenotypic heterogeneity, and may improve early detection and intervention for those who are at risk of poor outcomes. Such findings also provide evidence of a possible neural correlate for resilience in language function. Interestingly, compensatory activity was found in left superior temporal cortex and other language areas in a study of language comprehension in 8 to 21 year olds diagnosed as high-functioning ASD in early childhood, who later achieved optimal outcomes (80).

In an event-related potential (ERP) study of the development of social brain networks in infants, neurotypical infants showed clear differentiation in brain responses to viewing faces with the eye gaze directed toward versus away from them. In contrast, infants who were later diagnosed with ASD showed little differentiation in their brain responses to these stimulus types (35,81). Furthermore, infants in the HR group who later exhibited typical behavioral outcomes showed a distinct difference in the processing of these stimuli relative to those who were later diagnosed with autism, as well as the control group. This ‘compensatory neural activity’ seen in the HR infants may be evidence of resilience, providing a neural correlate for the developmental shift into a ‘typical’ behavioral trajectory.

Further evidence of resilience was demonstrated in a group of children (49 – 77 months old) with ASD receiving 2 years of the Early Start Denver Model (ESDM), a comprehensive developmental behavioral intervention (82). Children who received ESDM showed significant improvements in IQ, language, adaptive behavior and autism symptoms compared with the children who received a community intervention (CI) (83). In a follow-up study, brain responses were compared to pictures of faces and toys – social and non-social stimulus categories - using EEG. Children in both intervention groups and neurotypical controls showed similar N170 responses, reflecting typical early perceptual processing of the stimuli. While the ESDM group showed similar brain activity to TD children reflecting deeper attentional and cognitive processes when viewing the faces, the CI group showed an opposite pattern of activity (82). Both studies provide evidence of resilience in the social domain, as shown by the positive behavioral outcomes following ESDM intervention, as well as the ‘normalization’ of patterns of brain activity to social stimuli. These results also support the need for early, intensive and high-quality interventions as a way to foster resilience.

Brain Structure

Brain development depends on the balance between cellular and synaptic growth and appropriately timed pruning of neurons and synapses. In at least some children with ASD, this balance appears to be disrupted (10). Structural MRI provides evidence for potential risk biomarkers for autism in the form of increased brain volume detected as early as 6 months (84–86). A longitudinal prospective study that used a deep-learning algorithm to examine hyperexpansion of the cortical surface area in children between 6 and 12 months of age was able to predict autism diagnosis at 24 months in HR children, with a positive predictive value of 81% and a sensitivity of 88% (65). One of the neuroendophenotypes identified as part of the Autism Phenome Project was boys with megalencephaly, a condition in which the brain is disproportionately large compared with body size (87). The estimated 15% of boys with autism with this phenotype were found to have more severe disabilities and had a poorer prognosis (88).

In addition to brain volume, measurements of extra-axial cerebrospinal fluid (CSF) have also been conducted in autism. Extra-axial CSF can be quantified by manual tracing of dura on successive coronal slices of the brain viewed with structural MRI that are summed to obtain intracranial volume. Semi-automated tissue segmentation is used to compute total cerebral volume, and the resulting space between the dura and cortical surface gives a measure of extra-axial CSF volume. From 6 months to 12 months, hyperexpansion of the cortical surface area and excessive CSF in the subarachnoid space (i.e., extra-axial CSF) precedes more global overgrowth and increased extra-axial fluid volume between 12 to 24 months (64,65,71,72). Recently, a large prospective study examined extra-axial fluid using MRI in subgroups of normal-risk (n=132) and HR (n=27) children with autism and neurotypical (n=77) controls, at age 3. Both autism groups had similar extra-axial fluid volumes, which were 15% greater than in neurotypical controls. Autism diagnosis was predicted at 83% (95% CI 76.2–88.3) by a machine learning algorithm (73). From the ages of 2 to 4 years, brain volumes in children with autism remain enlarged compared with those of their peers, particularly affecting the amygdala, frontal cortex, and temporal cortex, areas important for social cognition, emotion regulation, and language, respectively (89). Crucially, both early brain volume overgrowth and increased extra-axial fluid volume are associated with later autism symptoms - early brain volume overgrowth is associated with severity of social deficits and restricted and repetitive behaviors (65,70) and increased extra-axial fluid is predictive of communication deficits in ASD (71).

These results demonstrate the potential of using structural MRI, particularly to examine extra-axial fluid, as a risk marker in HR infants as young as 6 months, to and predict diagnosis and symptom severity for autism. Studies identifying early structural brain differences could also be used to investigate trajectories of resilience, if participants are followed longitudinally to assess symptom severity and adaptive outcomes through development and adulthood.

Brain Connectivity

Structural connectivity (SC) measured with diffusion tensor imaging (DTI), has also been used to identify early risk markers and correlates of symptom severity in ASD (75). Fractional anisotropy (FA) quantifies white matter tract microstructure, with an increase in FA indicating an increase in myelination (90). In HR infants as young as 6 weeks to 6 months, FA and lateralization of the superior longitudinal fasciculus (SLF) were found to be related to later language development and autism symptoms (67,74).

Network efficiency, or the capacity of the brain to exchange information locally and globally, has also been examined using SC metrics. In a longitudinal study of 116 infants scanned at 6 months and 12 months of age, network inefficiencies in regions involved in low-level sensory processing were seen as young as 6 months in HR infants who were later diagnosed with ASD. This metric also predicted symptom severity at 24 months (66). This work suggests that early network inefficiencies contribute to a developmental cascade affecting brain organization and behavior.

Recent work examining functional connectivity (FC), derived from resting-state fMRI data, has also demonstrated potential in identifying risk markers for HR infants (69,91). Machine learning applied to FC measures derived from fMRI of children at 6 months of age has been used to predict autism diagnosis in HR infants, and correctly classify infants who were not diagnosed (91).

Alongside whole-brain approaches examining long-distance FC, local functional alterations can be assessed using regional homogeneity (ReHo), which computes the coherence between a given voxel and those surrounding it. In neonates with a family history of ASD, significantly elevated ReHo has recently been observed (61). In a cross-sectional investigation of children and adolescents with ASD, higher local connectivity as measured with ReHo was associated with poorer social communication skills (92).

Taken together, developmental neuroimaging studies show promise in establishing biomarkers for early diagnosis, for parsing phenotypic heterogeneity, and for predicting behavioral outcomes. However, as there is substantial variability during the early developmental period we have focused on, a multi-disciplinary approach to studying risk and resilience in autism may help move us closer to more precise risk markers and more adaptive models of care. One such approach is to combine neuroimaging with genetics to provide corroborative data on carriers of risk alleles for a more unequivocal assessment of risk (2,13,93) and to identify mechanisms of resilience. Jack and colleagues (2020) investigated female-specific protective factors in ASD, and found that females with ASD showed a different pattern of brain activation than males with ASD during the viewing of biological motion point-light stimuli. This was also true of neurotypical females, who showed greater recruitment of salience and executive control networks compared with neurotypical males, a possible correlate of the protective effect of being female (29). In the genetic data, the authors found female-specific gene expression patterns (larger size of rare copy number variants, CNVs) in striatal cortex of females with ASD that may be related to the sex-differential processes of risk and resilience (37).

There are limited conclusions we can draw about mechanisms of resilience in ASD from existing neuroimaging studies, as studying the dynamics of resilience as an outcome requires a longitudinal approach. These studies still need to be done, incorporating promising multi-disciplinary approaches (37). Most importantly, evidence already exists that early, intense, supportive intervention can promote factors that increase resilience in the face of genetic or environmental risks (82,94).

3. Towards genetic and epigenetic markers of resilience

Resilience is not simply a reversal of pathological mechanisms, but involves specific psychosocial and biological processes that buffer a person against the impact of a stressor. Oxytocin (OXT) is a neurohormone that has frequently been implicated in affiliative and social behaviors (95). One of the ways OXT may affect social behavior is by attenuating the stress response through its actions on the neuroendocrine system (hypothalamic-pituitary-adrenal - HPA-axis) (96). OXT suppresses the HPA-axis, increasing interpersonal trust while reducing anxiety (97).

In recent years there has been enthusiasm about OXT as a potential pharmacological treatment for socio-emotional symptoms of autism (98,99). Functional neuroimaging studies of autistic children and adults have reported significant changes in activity after OXT administration in brain areas involved in social information processing and reward processing (100–106). These studies suggest that variation in the OXT system contributes to social phenotypes, regardless of diagnosis (107).

To identify subgroups of individuals who are most likely to benefit from pharmacological treatment with OXT, Voinsky and colleagues looked at variations in expression of OXTR and other genes in children with ASD, their neurotypical siblings, and neurotypical controls aged 3 to 16 years. Across all groups, higher OXTR expression levels correlated with i) better outcome as measured by the Vineland Adaptive Behavior Scale (VABS), ii) less severe social impairment as measured by the Social Responsiveness Scale (SRS) and iii) less severe behavior problems as measured by the Aberrant Behavior Checklist (ABC). Based on their findings, the authors propose that relevant gene expression levels should be evaluated as prognostic biomarkers by clinical trials or before administration of other tentative autism treatments (108).

As a result of the dynamic balance between risk and protective factors, resilience is a process that arises out of the interaction of genetic, epigenetic and environmental influences. The control of gene expression may provide a possible genetic mechanism of resilience. One key mechanism of epigenetic control of gene expression is DNA methylation. DNA methylation is the process by which methyl groups are added to cytosine–phosphate–guanine (CpG) sites on the regulatory or promoter regions of genes to silence transcription (109). This process is under genetic control, but subject to environmental influences (110,111).

Andari and colleagues investigated epigenetic modification of the OXTR gene in adults with ASD. They found that participants showing hypermethylation in the intron 1 area of OXTR showed fewer deficits in social interaction and communication as measured by the Autism Diagnosis Interview (ADI-R) and hypo-connectivity between superior temporal and posterior cingulate cortices, brain areas involved in theory of mind (112). Thus, OXTR hypermethylation may be a potential epigenetic biomarker for resilient social phenotypes in autism, and DNA methylation may be an epigenetic mechanism that contributes to this phenotype.

Several common alleles of the OXTR gene may increase risk for ASD (113,114), and a recent study has found that increased OXTR risk-allele dosage was associated with symptom severity in a group of 41 high-functioning youth with ASD, as well as with reduced connectivity of the nucleus accumbens (NAcc) to other components of the reward system. On the other hand, youth in the neurotypical control group showed increased NAcc connectivity with frontal brain regions involved in mentalizing, which was correlated to better scores on the SRS. Thus, additive genetic effects as measured by risk-allele dosage, may be an important factor in conferring risk but also in revealing resilient endophenotypes (93).

Accumulating evidence suggests that genetic and epigenetic factors have significant impact on risk during developmental processes pre- and postnatally (13). Early life stressors may alter the epigenome at various sites, but the specific alterations and psychiatric outcomes depend on multiple factors specific to the individual. DNA methylation contributes to the regulation of OXTR gene expression, which is associated with both biological and behavioral outcomes. This may be a direct effect, through OXT influencing stress regulation via the HPA-axis, or an indirect effect whereby risk and resilience have a common cause. In either case, risk-allele dosage and gene expression may serve as potential biomarkers for resilience.

4. Future Directions and Conclusions

Researchers have primarily worked towards identifying biomarkers that categorize autistic persons as a separate group from neurotypical individuals. However, the research domain criteria (RDoC), a novel framework for investigating mental disorders that integrates information at multiple levels, presents practicable alternatives (115). This is important for both the study of risk for autism, a heterogeneous disorder of largely unknown etiology, and for the study of resilience, a dynamic process of adaptation similarly influenced by genetic and environmental factors. RDoC highlights the importance of understanding mental health on a comparative scale that crosses boundaries set by the DSM categories (116).

While substantial progress has been made by infant studies which typically track children until diagnosis of ASD (117), longitudinal studies with at least three lifecourse time points are needed, that concurrently track changes in the clinical, neurocognitive, functional, and anatomical trajectories to ascertain the prognostic value of biomarkers (118). Furthermore, multi-modal biomarkers – such as those combining genetics and neuroimaging (93,119) likely have improved prognostic as well predictive value relative to markers based on one modality (120,121), and will help to develop more precise interventions and effective treatments targeted specifically to individuals who stand to benefit from them.

While key protective factors in autism have been defined (21,25–28), the long-term goal is for research to describe the mediating mechanisms for risks and protective factors (22,24). As resilience is a dynamic process, there is inter-individual variability in adaptation through time and by domain. To investigate mechanisms of this multi-faceted process, longitudinal studies are needed to establish causation from risk to resilience, as well as to capture individual differences throughout the lifecourse (122).

Conclusion

Current understanding of the etiology of autism is based on the interaction of multiple genes with each other and with environmental factors, leading to neurodevelopmental processes that result in the expression of autism. Similarly, the resilience of each individual in the face of adversity is the result of the interaction between vulnerability and protective factors. In this review, we have highlighted important early markers of risk, and explored potential mechanisms of resilience. Taken together, this work suggests that neuroimaging in early development can provide evidence of risk biomarkers, which can be refined through multi-modal approaches. Further, it may be possible to strengthen resilience through psychosocial and biological means at different timepoints during development to improve outcomes across the lifespan for autistic individuals.

The meaningful impact of research in autism depends on the integration of information from genetic, neural, and behavioral approaches over the lifecourse. Most importantly, in looking towards the future, views about autism are changing, driven by autistic self-advocates and their allies (123,124). Research has the potential to transform the lives of autistic people and their families, but only when it is relevant, valued and effectively implemented (125). The relevance and influence of autism research will depend on researchers getting involved with the community they are researching. The next chapter of autism research will be improved by multidisciplinary global collaboration between researchers and clinicians, the involvement of families, and the autism community itself.