Immune dysregulation in early life and risk of autism

Abstract

Background: The notion of immune perturbations in autism is in part based on numerous observations of an association between maternal infections in general during pregnancy and autism in the child, and on observations of immune abnormalities in individuals diagnosed with autism. Maternal infections during pregnancy are linked to autism in several epidemiological studies, although genetic/familial confounding has rarely been considered. It therefore remains to be established if these associations represent a true causal effect. Moreover, biomarker studies of the maternal and neonatal immune status indicate the presence of atypical levels of immune markers among neonates later diagnosed with autism, as well as their pregnant mothers, compared to general population controls. However, the results from these studies are disparate. Most of the studies focus on rapidly fluctuating markers of inflammation, making comparisons across studies difficult, and none consider a more integrated view of the immune status during early development to evaluate its potential clinical utility as predictor of autism. Nor have any of the biomarker studies taken the possibility for genetic confounding into account, by restricting the comparisons to related individuals.

Methods: We used the register-linkage Stockholm Youth Cohort (SYC) to identify cases of autism in Stockholm County born between 1996-2000, and a random selection of population controls (Study I-Study III). From these individuals, archived Neonatal Dried Blood Spots (NDBSs) and the corresponding first trimester maternal serum samples were collected, as well as NDBSs from unaffected siblings to the individuals with autism. Concentrations of 26 maternal immune markers (9 acute phase proteins [APPs] and 17 cytokines) and 9 neonatal markers (acute phase proteins) were measured. The associations of individual immune markers with autism, stratified by the presence of co-occurring conditions (intellectual disability [ID], and attention deficit/hyperactivity disorder [ADHD]), were estimated using logistic regression models, as were the associations between grouped components of maternal immune markers and autism, representing patterns of unique variance extracted by principal component analysis (PCA). Finally, the predictive performance of the maternal immune markers collectively was evaluated by using Random Forest machine learning algorithms. To assess the etiologic relevance of maternal infections during pregnancy for autism (Study IV), we included children born 1987-2010 in the SYC, and employed a cohort study design using Cox regression models to estimate the associations between specialized care for any infection during pregnancy and a later diagnosis of autism in the child, stratified by co-occurring ID. To account for potential residual confounding by familial factors, we additionally employed negative control (using maternal infections prior to pregnancy as a negative control exposure) and matched sibling analyses.

Results: Of the maternal APPs (Study I), low levels of maternal CRP, and both low and high levels of maternal ferritin, were associated with autism in the child. The strength and direction of specific associations varied by the presence of co-occurring ID and ADHD. Only the association between ferritin and autism with co-occurring ID survived correction for multiple comparisons. Several weak to moderate associations between individual maternal cytokines (Study II) and autism were observed, including positive associations with the pro-inflammatory cytokines Interleukin 1β and tumor necrosis factor α. As with the maternal APPs, the specific patterns of associations varied by diagnostic sub-group, and none survived correction for multiple comparisons. Using PCA and machine learning algorithms, considering the maternal immune markers collectively as a group (using both cytokines [from Study II] and APPs [from Study I]), we found no evidence to suggest that adding the maternal immune markers to other maternal/obstetric factors (e.g., sex of fetus, maternal age, and maternal psychiatric history) improves prediction of autism. Further, when examining the associations between autism and the extracted principal components, only the second largest component (PC2), mostly influenced by the maternal APPs, was associated with odds of autism in the child. This indicates that maternal APPs as a class are more strongly associated with autism compared to cytokines. Among the neonatal APPs (Study III), only C-reactive protein (CRP) displayed an (u-shaped) association (p<0.05) with odds of autism in the case-cohort comparison, although the relationship was attenuated (p>0.05) when the comparison was restricted to autism-discordant siblings. In contrast, both ferritin and α-2-macroglobulin were inversely associated with autism in the discordant sibling comparison, but not in the case-cohort comparison of unrelated individuals. The levels of neonatal immune markers and the associations between several of the markers and autism varied by the presence of perinatal/maternal factors, including maternal anemia, pregnancy infection, and psychiatric history. In the full cohort comparison (Study IV), maternal infection during pregnancy was associated with a later diagnosis of both autism and ID in the child. Maternal infection the year before pregnancy (negative control exposure) was also associated with autism, but not ID. In the discordant sibling analyses, the relationships with maternal infections during pregnancy were completely attenuated for autism, but not to the same extent for ID.

Conclusions: In conclusion, we found no strong consistent pattern of maternal or neonatal immune activation when measuring 26 immune markers (cytokines and acute phase proteins) in archived samples from pregnant mothers or nine acute phase proteins in neonatal samples. Adding the maternal immune markers to other registered maternal/obstetric variables did not improve the predictive performance of machine learning models, indicating that the immune markers have limited utility as tools for early detection in clinical practice. Among the strongest associations were maternal FER and autism, particularly with co-occurring ID, and neonatal FER and autism in the discordant sibling comparison, indicating a potential role for maternal and neonatal iron metabolism in brain development. In line with previous register studies, we observed an association in the general population between maternal infection during pregnancy and autism. However, the association disappeared in in the sibling comparison, indicating the presence of familial confounding factors operating at the population level, and that the previously reported associations do not likely reflect a causal relationship. Furthermore, mothers to children with autism had a higher incidence of infections the year prior to pregnancy, suggesting an increased propensity for infections in mothers with liability for autism. Taken together, we found no strong support of maternal immune activation as a causal factor for autism, across all studies in the thesis. Future studies ought to consider the complex interaction of the genetic background/liabilities in addressing questions of the role of the immune system in early life and the etiology of autism.