Autism Innovative Medicines Studies-2-Trials (AIMS-2-TRIALS) is a public–private partnership that brings together academia with industry, charities, regulators and the autism community, to improve outcomes for autistic people by matching mechanism-based therapies to individual needs and biological profiles. AIMS-2-TRIALS extends EU-AIMS (2012–2018) — the first Europe-wide collaborative autism biomarker discovery initiative.

A global infrastructure for biomarker research

EU-AIMS and AIMS-2-TRIALS have established a unique multidisciplinary longitudinal research platform of linked cohorts, already spanning >1,500 individuals from infancy to adulthood. These studies aim to parse the heterogeneity of autism, moving beyond case–control design (implicitly targeted at diagnostic biomarker discovery) to include prospective/longitudinal and cross-condition cohorts (including autism, attention-deficit hyperactivity disorder and intellectual disability), gene-first approaches (investigating genes linked to synaptic differences in autism), non-Western community samples (of children with increased environmental risk for social-emotional/cognitive difficulties in South Africa) and pharmaco-challenge studies. This platform supports biomarker discovery across different contexts of use, including validation of prognostic markers to predict individual developmental trajectories and outcome, and predictive markers of the efficacy of an intervention for a particular person (Supplementary Fig. 1).

All cohort studies assess participants in six bio-behavioural domains thought to be involved in autism core/associated features (social, emotional, reward/motivation, cognitive, un/predictability and sensory processing) across different levels — from behaviour to cognition, brain structure/function, and contributing molecular processes, genomics and environmental factors. This work is complemented by a parallel preclinical workstream to promote forward/back translation and identify plausible molecular targets for trials. For example, the same social habituation paradigm used in cohort studies to identify a subgroup of individuals with specific difficulties in social processing has been applied in Nlgn3-knockout mouse models of autism, where we have recently demonstrated rescue of oxytocin response and social behaviour with a novel, highly specific, brain-penetrant inhibitor of MAP kinase-interacting kinases1.

Biomarker discovery in AIMS-2-TRIALS follows a rigorous pipeline to ensure candidate markers meet evidentiary criteria, from pre-analytical validation (for example test–retest reliability) to analytic validation and utility in clinical trials (Supplementary Fig. 2). Industry partners have provided substantial expertise on defining precise context-of-use statements. This is a pivotal step for any biomarker development pipeline, but one with which many academic experts in psychiatry may be inexperienced.

Industry partners have also facilitated consortium engagement with regulators, beginning during EU-AIMS. For instance, in 2015 and for the first time for a neurodevelopmental condition, the European Medicines Agency (EMA) provided qualification advice and five letters of support for the design and methodologies of the EU-AIMS and AIMS-2-TRIALS Longitudinal European Autism Project (LEAP). This increases the likelihood that biomarkers identified in this study will be qualified for clinical trials.

The autism community are further contributing lived experience and expertise in evaluating the desirability and feasibility of candidate biomarkers and novel intervention approaches, discussing ethical considerations of the project, advising on task and trial design, and defining meaningful clinical endpoints that intervention targets map onto. In addition to proximal clinical outcomes such as social-communication difficulties, this includes greater emphasis on sensory processing, co-occurring mental health (for example, anxiety) and medical (for example, epilepsy) conditions, and more distal outcomes such as quality of life.

New leads for autism biomarkers

Using the research platform described above, we have replicated resting-state functional connectivity differences in autism across four large cohort studies comprising >2,000 individuals2. Results from this work indicated that, on average, autistic people showed hypoconnectivity in sensory-motor regions and hyperconnectivity in prefrontal/parietal cortices — the latter related to increased communication difficulties and reduced adaptive functioning.

These findings are key first steps toward analytical and clinical validation; however, mean-group differences are limited in demonstrating biomarker utility in a heterogeneous condition such as autism. In fact, we have identified that across the most influential areas of autism research (for example, theory of mind), significant mean group differences typically indicate that only a subgroup of autistic people are likely ‘biomarker-positive’ on the relevant measure. We have therefore developed ‘big data’ analytic approaches to capture individual profiles (rather than mean-group differences) for biomarker discovery. For example, we developed ‘normative modelling’ approaches for neuroimaging analyses to assess how each individual in a given sample varies against expected trajectories of brain development, social or cognitive function according to age and sex. These individual scores provide a basis for the identification of potential subgroups (for example, via clustering), and bringing forward cognitive and neurobiological indices as broadly generalizable outcome measures.

Recent work applying this approach identified that individual-level change in adaptive functioning in autism was underpinned by the degree of individual divergence from neurotypical neuroanatomical profiles (of cortical thickness/volume and surface area), as well as the genetic processes associated with these profiles and broader polygenic likelihood for autism3. This evidence highlights progress in identifying putative systems underpinning heterogeneous neurobiological and phenotypic trajectories of autism and related conditions. However, the feasibility of deploying the biomarker in clinical practice is also a major consideration, given that some methods such as magnetic resonance imaging may be more expensive, invasive or burdensome than others.

Electroencephalography (EEG) is already widely used in clinical practice, including epilepsy diagnosis. In joint work with the US FNIH Autism Biomarker Consortium for Clinical Trials (ABC-CT), we have replicated significant differences in an EEG face processing signal (N170 latency) in autism — chosen a priori, based on the existing literature showing that face processing differences are linked to social-communication difficulties. We demonstrated that N170 latency has convergent validity (with functional imaging of brain response to faces and autism polygenic scores), prognostic value in predicting changes in social adaptive behaviour ~18 months later and is scalable for real-world clinical settings (>80% successful acquisition in autistic people, including children)4. N170 latency has become the first autism biomarker to be accepted into the US Food and Drug Administration (FDA) biomarker qualification program and has support from the EMA.

The crucial next step is to also test pharmacodynamic biomarkers for therapeutic candidates in proof-of-concept studies and predictive biomarkers in clinical trials. So far, we have shown proof-of-concept that the selective GABA-B receptor antagonist arbaclofen abolishes specific visual sensory processing difficulties in some autistic people5. Arbaclofen is also being used in a European clinical trial targeting social functioning in autistic young people (NCT03682978; and in a ‘sister’ trial led by the Canadian Province of Ontario Neurodevelopmental Network) that incorporates N170 latency to assess its utility as a baseline covariate to improve the efficiency of detecting intervention effects.

Conclusions

Through international translational collaborations and working with regulators, significant progress has been made in validating autism candidate biomarkers, such as EEG N170 latency, and in characterizing the neuroanatomical basis of changes in clinical features. These advances are being bolstered by multi-stakeholder collaboration to ensure outputs correspond with community priorities and deliver real-world impact. The experiences from our joint teams could provide guidance for future efforts to advance biomarker discovery and precision medicine across psychiatry.

Acknowledgements

We acknowledge the AIMS-2-TRIALS consortium/collaborators, including experts contributing to this article: Evdokia Anagnostou, Celso Arango, Tobias Banaschewski, Simon Baron-Cohen, Christian Beckmann, Thomas Bourgeron, Linda S. Brady, Jan K. Buitelaar, Florence Campana, Tony Charman, Christine Ecker, Richard Delorme, Louise Gallagher, Lindsay M. Ham, Rosemary Holt, Mark H. Johnson, Irena Kadiu, Luke Mason, Grainne McAlonan, James C. McPartland, Hein Odendaal, Gahan Pandina, Mark P. Richardson, Herbert Roeyers, Maria Tome, Peter Scheiffele, Emily Simonoff, Priscilla Springer, Julian Tillmann and Terje Falck-Ytter. This work received support from the Innovative Medicines Initiative 2 Joint Undertaking (grant agreement no. 777394) for AIMS-2-TRIALS. This Joint Undertaking receives support from the EU’s Horizon 2020 programme, EFPIA, Autism Speaks, Autistica and SFARI.

Disclaimer

The views expressed are those of the author(s) and not necessarily those of the IMI 2JU, NHS, NIHR, Department of Health and Social Care, National Institutes of Health, the Department of Health and Human Services, nor the United States government.

Competing Interests

C.H.C. is an employee of F-Hoffmann La Roche. The other authors declare no competing interests.