Using Connectivity to Explain Autism

The case of cerebral connectivity research in ASD contains themes that are commonly seen throughout cognitive neuroscience research. The end goal of the enterprise is to evaluate and validate causal models that demonstrate how physiological changes—in this case, alterations in cerebral connectivity—cause a behavioral phenotype. Method development, validation and usage are important, but are explicitly in the service of more rigorously and specifically evaluating causal models. It is alternative causal models, and not methodological read-outs, which should be the language by which progress in the field is discussed. Cognitive neuroscience, as a relatively new field with many disciplinary forbearers, needs greater validation of constructs that serve as elements within cognitive models. Bridging an imaging level of analysis and a computational/cognitive level, we have attempted to open a conversation about differentiable computational constructs that could add important nuance to what we mean when we report on “connectivity” and bring us closer to understanding how semantic information is transferred and computed upon in the brain. Within ASD-connectivity research, we need to sharpen our correlational knowledge by specifying and testing both the topography (where) and developmental (when) parameters of causal models. We then need methods to perturb the system at various points in the model to firmly establish causality. We also need multi-modal studies to eliminate confounds (such as artifacts to which one method but not the other is sensitive), to index non-connectivity-related biological elements within a connectivity model, and to directly test non-connectivity theories against connectivity theories of ASD. Theories also need to be clear about theoretical scope. Is the study making causal claims about “all of ASD,” about a core symptom (social communication, restricted/repetitive interest/behaviors) or about a peripheral symptom (altered executive function, language, motor function, perception)? Is it making claims about a symptom that is seen only in ASD or one that is seen in multiple neuropsychiatric conditions? Consideration of scope is important for two practical reasons: for dealing with the heterogeneity of ASD and for dealing with the fact that there is overlap among classical neuropsychiatric diagnoses both in terms of connectivity differences (compared with controls) and symptoms. Data-driven approaches may help us “re-slice the pie,” but even if that venture is successful, causal inference will still be needed to help us understand how brain changes cause behavioral consequences. And only in doing so will we best be able to develop biomarkers and interventions that help affected individuals and families reach their life goals.