Neuropsychiatric “comorbidity” as causal influence in autism

Abstract

Behavioral comorbidity is the rule rather than the exception in autism spectrum disorder (ASD), and the co-occurrence of autistic traits with subclinical manifestations of other psychiatric syndromes (e.g. anxiety, developmental coordination disorder) extends to the general population, where there is strong evidence for overlap in the respective genetic causes. An ASD “comorbidity” can have several fundamentally-distinct causal origins: it can arise due to shared genetic risk between ASD and non-ASD phenotypes (e.g., ASD and microcephaly in the context of the MECP2 mutation), as a “secondary symptom” of ASD when engendered by the same causal influence (e.g., epilepsy in channelopathies associated with ASD), due to chance co-occurrence of ASD with a causally-independent liability (e.g., ASD and diabetes), or as the late manifestation of an independent causal influence on ASD (e.g., Attention-Deficit/Hyperactivity Disorder). Here, we review evidence for the latter, i.e., the role of non-specific causal influences on the development of ASD itself. The notion that non-specific insults to neural development, either inherited or acquired, might augment the impact of ASD-specific genetic susceptibilities in contributing to its cause has not been appreciated in the literature on comorbidity and has significant implications for both personalized intervention and future research. Prior biomarker studies of ASD have typically not accounted for variation in such traits. The statistical power of future studies, particularly in autism genetics and neuroimaging, can be enhanced by more comprehensive attention to the measurement of comorbid behavioral traits that index causal influences on the disorder, among not only cases but (importantly) controls.

1. Introduction

Behavioral comorbidity is the rule rather than the exception in autism spectrum disorder (ASD). Within individual subjects, clinical studies estimate that 70-96% of ASD diagnoses are complicated by at least one co-occurring psychiatric disorder ^1,2, and the phenomenon of comorbidity between autism and other psychiatric syndromes generalizes to observations of overlap in subclinical traits of these syndromes in the general population ^3-5. An ASD “comorbidity” can have several fundamentally-distinct origins (Figure 1): it can arise (A) due to shared genetic risk between ASD and non-ASD liability (e.g., ASD and microcephaly in the context of the MECP2 mutation), (B) as a “secondary symptom” of ASD when engendered by the same causal influence (e.g., epilepsy in channelopathies associated with ASD), (C, i) due to chance co-occurrence of ASD with a causally-independent liability (e.g., ASD and diabetes), or (C, ii-iii) as the late manifestation of an independent causal influence on ASD (e.g., Attention-Deficit/Hyperactivity Disorder). This review explores the latter possibility (C, ii-iii), i.e., that some of the psychiatric comorbidities which aggregate in patients with ASD do so because they contribute to its causation, and that trait correlations with comorbidities of any origin (A-C) may emerge from developmental interactions between autistic impairment and these co-occurring conditions ^6,7. Our goals in this review are threefold. First, we will review evidence for the role of non-specific causal influences on the development of ASD. Then, we will discuss methods for elucidating whether an aggregating comorbidity may represent a causal influence on the development of autism. Finally, we will detail the practical and theoretical implications of reconceptualizing some ASD comorbidities as potential contributing causea of the syndrome.

Figure 1: Potential Pathways to Autism Comorbidity.

Note: Schematic depicting causal pathways for inherited comorbidities. Black circles represent ASD-specific genetic risk, white circles represent risk for non-ASD conditions, and gray represents genetic overlap. Squares represent phenotypic traits. In (A), comorbidity arises due to shared genetic risk between ASD and non-ASD liability. In (B), comorbidity arises as a “secondary symptom” of ASD. In (C), comorbidity may arise due to chance co-occurrence of ASD with a causally-independent liability (solid line, i), and/or as a function of non-ASD-specific liability amplifying risk for ASD (dotted lines); the latter could occur in utero (ii) or post-natally (iii) and would result in the independent co-occurring condition contributing to cumulative risk that results in total liability for ASD to exceed the threshold for diagnosis.

2. Non-specific causal influences on ASD

Some of the earliest evidence for the role of non-specific causal influences on the development of ASD arose from observations of correlations between autistic traits and quantitative traits of general psychopathology in large epidemiologic samples ^8,9. General psychopathologic traits that have been associated with sub clinical manifestations of the autistic syndrome in the general population include, but are not limited to, measurements of anxiety ¹⁰, symptoms of attention-deficit/hyperactivity disorder (ADHD) ⁹, and deficits in motor coordination ¹¹. When observations of autistic-psychopathologic trait overlap extend to the pathologic extremes of the respective population distributions, this by definition constitutes diagnostic comorbidity ¹². As a clinical entity, ASD is highly comorbid with conditions such as epilepsy ¹³, anxiety disorder ², ADHD ¹⁴, and developmental coordination disorder ^15,16.

Trait co-aggregations that emerge in a genetic epidemiologic context, such as in twin and family studies, allow the degree of genetic and environmental causal overlap to be estimated from latent parcels of influence and their associated patterns of trait aggregation in family members who are genetically related to differing extents. There have been highly replicated observations of overlap in additive genetic influence between autistic traits and other inherited neuropsychiatric traits ¹⁷, suggesting that some of the causal influence on autistic traits may be highly pleiotropic ¹⁸. This is beginning to be verified in large molecular genetic studies that are adequately powered to detect the additive effects of common variants ¹⁹. Recently, it was observed that non-specific neurodevelopmental impairments (principally, motor coordination impairments and attention-deficit/hyperactivity symptoms) accounted for 60% of the variation in the recurrence of autism in multiplex families ²⁰. This occurred in the absence of aggregation of either trait among the unaffected siblings in the families, suggesting that motor coordination impairments and attention-deficit/hyperactivity symptoms may—when super-imposed upon a background of sufficient ASD-specific liability—exert causal influence on the development of ASD.

Corroborating this observation, recent work in a large population-based study confirmed that siblings of children with ADHD have a four-fold increase in the incidence of autism ²¹. The notion that non-specific insults to neural development, either inherited or acquired, might augment the impact of highly specific background genetic susceptibilities (i.e. to autism) was additionally demonstrated in a landmark study ²² detailing effects of the highly pleiotropic 16p11.2 deletion. In this study, it was shown that the effect of a de novo 16p11.2 deletion, rather than being “absolute,” incurred a predictable, deleterious “shift” (on the order of 2 standard deviations) in the degree of autistic impairment in offspring relative to what was expected on the basis of parental traits (indexing genetic background) for the condition within the normal range. Thus, children with de novo deletions of 16p11.2 whose parents fell in the below average range for social relatedness (yet still entirely within normal limits) were significantly more likely to be affected by clinical-level ASD than the average 16p11.2 deletion patient. The notion of a pleiotropic (non-specific) chromosomal rearrangement inducing a predictable shift against an ASD-specific genetic background, indexed by variation in the sub clinical ASD trait burden of parents, represents an important and novel way of conceptualizing the manner in which independent genetic liabilities can interact to accentuate risk for a given diagnostic outcome.

3. Distinguishing features of comorbidities that contribute to causation

Figure 1 schematically depicts hypothetical distinctions between competing mechanisms that underlie the phenomenon of comorbidity. Clues to whether a given comorbidity might contribute to causation in autism include the following, many of which encompass the principles by which endophenotypes are defined: does the co-morbid phenotype (1) occur more commonly in patients with ASD than in the general population; (2) exhibit heritability (since ASD is highly heritable); (3) predict recurrence in ASD-multiplex families; (4) occur more commonly in the unaffected relatives of individuals with ASD than in the general population (i.e. has the characteristic of an endophenotype); (5) manifest levels of severity that are correlated with autistic symptom severity in patients with ASD; and manifest levels of severity that are correlated with autistic trait severity in the general population (6) during infancy and/or (7) during childhood and adulthood. This matrix of considerations is depicted by the column headings of Figure 2. For many common neuropsychiatric “traits” (i.e., enduring, quantitatively distributed phenotypes) that have been associated with ASD or considered common comorbidities, this fundamental set of questions remains partially or fully unanswered.

Figure 2: Comorbidity Matrix.

Note: Neuropsychiatric traits that commonly complicate ASD diagnosis and severity (rows) characterized with respect to quantitative axes of impairment (columns). This list reflects heterogeneity among potential endophenotypes of ASD, but it is not meant to be exhaustive. Further research aimed at deconstructing autism along quantitative axes may help disambiguate ASD specific, non-ASD-specific, and independent liabilities.

Here we refer to causal influences that are specific to autism as ASD-specific liabilities. They would be expected to be highly heritable because autism is highly heritable. They would be expected to manifest during infancy because that is when the onset of autism occurs, and to predict recurrence in multiplex ASD families. At levels of aggregation whose phenotypic expression does not reach a clinical threshold, they would be expected to exhibit the characteristics of endophenotypes. Social communicative impairment is one such ASD specific liability ^23-25.

Non-ASD-specific liabilities that exert causal influences on autism would be expected to engender ASD only when superimposed upon backgrounds of inherited ASD specific liability ^18,20. Motor coordination deficits, attention-deficit/hyperactivity (ADHD) symptoms, and low levels of visual social engagement (VSE) are a few recently-established candidates for this type of causal comorbidity (Figure 2). Anxiety-related traits also may fall within this category. They constitute highly pleiotropic behavioral influences and, although non-specific to autism, exhibit endophenotypic characteristics in some family studies ^4,10,26-28. Like ASD-specific liabilities, non-ASD-specific causal liabilities predict familial recurrence and exhibit significant causal overlap with autistic traits in the general population ^9,11,18. They may ²⁹ or may not ^20,30 aggregate in the unaffected siblings of individuals clinically affected by ASD. Future research is warranted to extend the list of co-occurring phenotypes which can be specifically interrogated as potential contributors to causation in autism.

Because autism is a disorder that first arises early in development, it is important to consider the timing of associations between autism and common comorbidities. To date, this issue has not been carefully considered for most behavioral comorbidities that complicate ASD. Unlike ASD-specific liabilities, which are expected to manifest during infancy, non-ASD-specific causal liabilities (e.g. ADHD symptoms) would be expected to exhibit substantial independence from ASD-specific liabilities prior to ASD diagnosis, manifest as “comorbidities” upon ASD diagnosis, and bear significant trait correlations with ASD following diagnosis if, over time, either co-occurring trait exacerbates the severity of the other (Figure 1C, ii-iii). Cross-sectional ascertainment methods cannot fully account for these reciprocal interactions between autistic and neuropsychiatric traits that may occur over the course of development.

A recent longitudinal study demonstrated the mechanistic implications of longitudinal ascertainment by assessing autistic traits and early manifestations of general psychopathology—e.g., internalizing traits, externalizing traits, and dysregulation behaviors—in a cohort of epidemiologically-ascertained toddler twins ⁶. These psychiatric traits consistently exhibit moderate correlations with autistic traits in school-aged children ^8,9. During the developmental period from 18-36 months, a pattern of independence between autistic and psychiatric liabilities was observed, evidenced in part by their distinct genetic and environmental structures. This indicated that some traits which are commonly correlated in school age children may be essentially uncorrelated during infancy and early childhood ⁶. Moreover, it suggested that commonly-observed trait associations in school age children may arise from interactions between autistic liability and independent susceptibilities to other psychiatric liabilities that occur after the period when autism first manifests ^6,18. Consistent with this notion, abnormalities in social adaptation during early childhood, indexed by the Competence subscale of the Brief Infant Toddler Social Emotional Assessment, have been found to predict psychiatric symptoms in later childhood ³¹.

It is important to consider how population-level features that characterize the autistic syndrome, such as sex differences ³² and inter-individual heterogeneity ^33,34, may influence patterns of comorbidity. In autism, sex profoundly differentiates phenotypic expression of inherited liability such that the population prevalence and family recurrence ratio are both 3:1::male:female ^35,36. Comorbidities arising from overlapping genetic influence would be expected to recapitulate a higher male prevalence (ADHD is a possible example), those which are secondary to ASD may only do so when comorbid with ASD, and those which arise independently from ASD would not be expected to exhibit sex effects unless influenced by the same or other mechanisms that incur sexual dimorphisms in that particular trait. With respect to heterogeneity, research has implicated a multitude of rare and common genetic variants in ASD ^33,34. These variants may converge on a discreet set of early developmental endophenotypes ¹⁸, different quantities and combinations of which could (1) account for behavioral and psychiatric heterogeneity in ASD; and (2) include phenotypic contributors that manifest as comorbidities in patients with ASD.

4. Implications

The studies briefly reviewed herein indicate that autism may be predicted from an array of neurobehavioral susceptibilities, many appreciable before the syndrome is diagnosed, and each potentially traceable to specific sets of genetic influence ¹⁸. Some of these liabilities are not necessarily specific to ASD and manifest as “comorbidities” later in development. This view of comorbidity has several important implications.

First, invoking a causal role for genetically-independent, non-ASD-specific liabilities can potentially reconcile as-yet-unresolved discrepancies between genetic epidemiologic studies and case-control molecular genetic studies regarding the degree to which the genetic causes of autism and those of other psychiatric syndromes overlap. In genetic epidemiologic studies, all causation is accounted for and estimation of overlap has been relatively high ¹⁷; in contrast, in case-control molecular genetic studies of distinct disorders, variant sets that distinguish autism cases from controls appear largely distinct from variant sets that distinguish general psychopathology cases from controls, and estimation of overlap has been relatively low ³⁷. This discrepancy would be expected if a non-ASD-specific genetic liability (e.g. ADHD susceptibility) both contributes to autism and is common (but unmeasured) in research participants without autism who serve as controls. In that scenario, genetic variants responsible for the comorbid (ADHD) traits would be excluded from the set of variants which distinguishes ASD cases from controls. Molecular genetic studies in which all phenotypes are ascertained in all subjects (cases and controls) avoid this problem and, not surprisingly, are yielding results that increasingly align with results of the original genetic epidemiologic studies—that is, results that suggest substantial additive genetic (causal) overlap between the disorders ¹⁹.

Second, this leads to the more general point that because non-specific liabilities often manifest at levels of severity that fall below accepted thresholds for a comorbid diagnosis, case-control studies of autism are confounded by definition when variation in such liabilities is ascertained only among individuals with clinical diagnoses of ASD ¹⁸. Under a model of causation that includes the contributing influences of both ASD specific and non-specific factors, failure to deconstruct (and measure) autism with respect to its contributing quantitative liabilities may severely erode researchers’ ability to derive a biological signal from group comparisons. This is particularly true when (1) sub clinical autism liability is present in control subjects—this is common ^12,38; and (2) non-specific traits that contribute substantially to the causation of autism are present in both cases and controls—there is now reason to believe that this, too, is common ^17,20,39. As described above for molecular genetic research, measurement of causal comorbidities among cases and controls should be incorporated into all biomarker studies that employ case-control study designs ^18,40, including neuroimaging and molecular genetic studies.

Third, biological studies of autism should attempt to map risk to phenotypes that index contributing heritable elements of ASD (both specific and non-specific) rather than to categorical diagnosis per se. If autism can arise from diverse combinations of inherited susceptibilities—each with its own genetic structure, and each of which yields a phenocopy of the condition—then it will be very difficult to derive meaningful genotype-phenotype correlations using categorical diagnosis alone. Critically, causal relationships can only be resolved in genetically informative studies to the extent that measurements of respective phenotypic traits are precise and maintain specificity in relation to disease-specific genetic liabilities.

Fourth, it will be necessary to understand the manner in which contributing susceptibilities affect one another over the course of development. In essence, there is a need to construct relational maps depicting core elements of behavioral development (i.e. domains of cognition, attention, communicative capacity, reciprocal social behavior, emotion regulation) in order to norm the effect of variation in one domain on the trajectory of another. This is analogous to the height-versus-weight tables that physicians use to determine whether a baby’s weight is normal or abnormal; likewise, it is akin to the discrepancy score criterion (indexing the difference between a child’s academic achievement and intelligence quotient) that school psychologists use to determine whether a learning disorder should be diagnosed. In all cases, the goal is to specify predictable relationships among various neurodevelopmental traits in order to detect when expected relationships are violated by pathogenic mechanisms ^40,41.

We conclude that some ASD comorbidities index neuropsychiatric liabilities that may contribute to the causation of ASD itself, particularly when superimposed upon backgrounds of inherited, ASD-specific liability. Identifying factors that exert causal influences on the development of ASD, whether specific to autism or highly pleiotropic in their effect, informs opportunity for targeted prevention of deleterious interactions that give rise to the autistic syndrome. Thorough exploration of the common “comorbidities” of ASD—in the manner depicted in Figure 2—may provide critical new clues to the developmental origins of the condition, may yield new insights into the various permutations and combinations of susceptibility that can result in clinical ASD, and may identify markers of resiliency among children who carry such susceptibilities but avert the impairments of ASD. Linking biomarkers and molecular genetic variants to the underlying traits that contribute to the causation of autism—rather than linking these variants to a diagnosis of “autism” per se—may enhance the statistical power of biological studies, improve the signal-to-noise ratio, and accelerate the development of personalized approaches to intervention, particularly if phenocopies of the autistic syndrome represent a unitary downstream consequence of various combinations of earlier-interacting liabilities.