Autism spectrum disorder

Autism spectrum disorder (henceforth autism) is a lifelong condition, which typically presents early in childhood. Autism is characterised by social interaction and communication challenges, repetitive behaviours and restricted interests, as well as sensory aberrations (American Psychiatric Association, 2013). Along with the core symptoms, there are often learning difficulties, intellectual disability, and other behavioural issues that may manifest ranging from aggression, self-injurious behaviour, impulsivity, irritability, hyperactivity, anxiety and mood symptoms (Levy et al. Reference Levy, Giarelli, Lee, Schieve, Kirby, Cunniff, Nicholas, Reaven and Rice2010; Rosen et al. Reference Rosen, Mazefsky, Vasa and Lerner2018; Hossain et al. Reference Hossain, Khan, Sultana, Ma, McKyer, Ahmed and Purohit2020). The severity of clinical symptoms and behavioural difficulties vary considerably from one autistic individual to the next, and may have either profound or mild impacts on day-to-day living. The heterogeneity in clinical phenotype implies the presence of ‘autisms’ or subgroups within autism. The discovery of effective treatments for the core symptoms has proved challenging due to the phenotypic and etiologic heterogeneity of autism. This is further complicated by changes in behaviours and symptoms throughout development and maturation, as well as the frequent presence of co-occurring conditions (van Steensel et al. Reference van Steensel, Bogels and Perrin2011; van Steensel & Heeman, Reference van Steensel and Heeman2017; Zheng et al. Reference Zheng, Zheng and Zou2018; Hollocks et al. Reference Hollocks, Lerh, Magiati, Meiser-Stedman and Brugha2019; Lai et al. Reference Lai, Kassee, Besney, Bonato, Hull, Mandy, Szatmari and Ameis2019). Genetics research has implicated multiple underlying biological pathways in the development of autism which means that a single treatment pathway for the core symptoms is not realistic (De Rubeis et al. Reference De Rubeis, He, Goldberg, Poultney, Samocha and Cicek2014; Ansel et al. Reference Ansel, Rosenzweig, Zisman, Melamed and Gesundheit2016; Satterstrom et al. Reference Satterstrom, Kosmicki, Wang, Breen, De Rubeis and An2020). It is possible that subgroups with similar underlying biology may respond to therapies targeting their specific pathways. Stratification biomarkers are objective measures used to define subgroups of individuals with shared biological characteristics. Identification of valid stratification biomarkers for autism may prove to be very beneficial, as it would not only help to elucidate subgroups, but could also better inform the development of therapeutics or drug treatment for specific subgroups by measuring effectiveness of a treatment across pre-defined subgroups rather than one heterogenous group in clinical trials. This would be clinically informative for understanding who may benefit most from different treatments.

Precision medicine, which is concerned with giving the right treatment to a patient based on their genetic, environmental and lifestyle factors (National Research Council, 2011), relies on a good understanding of the divergent underlying biology at the subgroup or individual level, further emphasising the need for stratification biomarkers. To understand the needs of autistic individuals and to provide more effective personal treatment plans we need to better characterise the relationships between clinical and biological heterogeneity in autism. Ideally stratification biomarkers would not only help to predict the best treatment option for an autistic individual based on their underlying biology and symptom presentation, but also at what stage intervention may be most beneficial.

Autism biomarker research

Biomarker research includes the investigation of molecular biomarkers such as genetics, proteomics, metabolomics, as well as neuroimaging and electrophysiological measures of brain structure and function, and eye-tracking measures of neurocognition (Loth et al. Reference Loth, Charman, Mason, Tillmann, Jones and Wooldridge2017; Ansel et al. Reference Ansel, Posen, Ellis, Deutsch, Zisman and Gesundheit2019). Important features of biomarkers include the ability to be measured consistently and objectively, and to be meaningful at the individual level (McPartland, Reference McPartland2016). However, there are still few research studies showing individual-level data for many of the brain structural and functional measures described above. It is also important to consider whether a biomarker is a diagnosis specific measure, or a symptom specific measure, such as social, sensory or restrictive and repetitive behaviours for autism.

The vast majority of autism biomarker research has focused on developing diagnostic biomarkers, which are measures that indicate whether someone does or does not have autism. Likelihood biomarkers, or early ‘risk’ biomarkers, which aim to predict who may be at an elevated likelihood of developing autism have also been examined (Ruggeri et al. Reference Ruggeri, Sarkans, Schumann and Persico2014; Jones et al. Reference Jones, Mason, Begum Ali, van den Boomen, Braukmann, Cauvet, Demurie, Hessels, Ward, Hunnius, Bolte, Tomalski, Kemner, Warreyn, Roeyers, Buitelaar, Falck-Ytter, Charman and Johnson2019). Likelihood biomarker research is based on the hypothesis that disruption to underlying biological mechanisms that lead to the development of autism may be present in infancy, and that early alterations to these biological mechanisms may be potentially addressed by early interventions to improve longer term outcome. Likelihood biomarker research is often conducted in younger siblings of autistic children who are approximately 20 times more likely to develop autism than those without an older sibling with autism (Ozonoff et al. Reference Ozonoff, Young, Carter, Messinger, Yirmiya and Zwaigenbaum2011). The identification of diagnostic and likelihood biomarkers to aid earlier diagnosis and facilitate earlier intervention would be invaluable as the average age of an autism diagnosis is still approximately 4 years of age, despite increased awareness and better approaches to diagnosis (Zuckerman et al. Reference Zuckerman, Lindly and Chavez2017).

Neuroimaging as a potential biomarker

Neuroimaging studies are an area of biomarker research that allows for in-vivo and non-invasive investigation of differences in brain structural and functional development. Structural neuroimaging studies showing differential developmental growth trajectory of brain volume indicated potential as an early diagnostic biomarker. These showed an increase in frontal and amygdala brain volume from 2 to 4 years of age in an autism group compared to typically developing controls, subsiding later in childhood at around 6–8 years of age (Courchesne, Reference Courchesne2002, Reference Courchesne2004; Courchesne et al. Reference Courchesne, Campbell and Solso2011a, Reference Courchesne, Mouton, Calhoun, Semendeferi, Ahrens-Barbeau, Hallet, Barnes and Pierce2011b). Further studies investigated whether early brain overgrowth was present even earlier in development, to assess the potential as an early diagnostic or likelihood biomarker. Brain overgrowth has been shown earlier than 2 years of age in infants with a high familial likelihood for developing autism. Increased surface area growth rate from 6 to 12 months of age, and increased total brain volume growth rate from 12 to 24 months were reported in infants with high familial likelihood that subsequently met criteria for autism diagnosis at 24 months compared to infants with both high and low familial likelihood that did not subsequently meet the criteria for diagnosis (Hazlett et al. Reference Hazlett, Gu, Munsell, Kim, Styner and Wolff2017). These findings indicated that early brain development differences can be detected with magnetic resonance imaging (MRI).

Brain functional connectivity patterns have also been explored as a potential brain biomarker for autism. Although many studies have focused on task-based fMRI, resting state fMRI is increasingly used as it more accessible, especially in younger children, due to the lack of cognitive demands of task-based fMRI (Yerys et al. Reference Yerys, Jankowski, Shook, Rosenberger, Barnes and Berl2009). Several resting state functional networks have been well characterised, in which functional activity in certain brain regions are temporally correlated at rest (Damoiseaux et al. Reference Damoiseaux, Rombouts, Barkhof, Scheltens, Stam, Smith and Beckmann2006; Biswal et al. Reference Biswal, Mennes, Zuo, Gohel, Kelly and Smith2010). While the literature has shown some mixed results, overall findings suggest patterns of local hyperconnectivity and long-range hypoconnectivity, with long-range connectivity also including alterations in inter-hemispheric connectivity (Yao et al. Reference Yao, Becker and Kendrick2021). Interestingly, some studies have reported underconnectivity between brain regions typically connected within a resting state network such as the default mode network, as well as overconnectivity between brain regions typically outside of a resting state network, suggesting perhaps a more diffuse pattern of functional connectivity is associated with autism (Hull et al. Reference Hull, Dokovna, Jacokes, Torgerson, Irimia and Van Horn2016). Age-related differences in hyper and hypo-connectivity patterns within autism also suggest that patterns of connectivity change throughout development in autism (Nomi & Uddin, Reference Nomi and Uddin2015a). Longitudinal studies of brain functional connectivity are necessary to examine developmental trajectories across the lifespan and associations with symptom severity to better understand and examine whether subtypes can be identified from functional connectivity patterns.

While many studies have used case-control group analyses to understand differences in brain structure and function in autism, this analytic approach does not assess whether certain brain structural or functional correlates have a high accuracy and specificity to predict whether an individual does or does not have a diagnosis or likelihood of developing autism. Pattern classification analytic approaches can be used to extract a pattern of biology or behaviour that is predictive of an autism diagnosis. A recent review reported that functional and structural MRI studies using pattern classification had autism diagnostic predictive accuracies of between 60% and 98%, whereas high likelihood infant studies were reported to have predictive accuracies of autism development of between 69% and 94% (Wolfers et al. Reference Wolfers, Floris, Dinga, van Rooij, Isakoglou and Kia2019). While diagnostic predictive abilities of up to 98% is very promising, the wide variability in predictive abilities of pattern classification studies is concerning. The discrepancies in predictive accuracies were attributed to differences in sample sizes and validation procedures, and to the heterogeneity or sampling bias within studies. This suggests that neuroimaging-based prediction may not be reliable enough to be a valid diagnostic or likelihood biomarker, or perhaps that more standardised analysis procedures are necessary. It also indicates that we need to consider the heterogeneity within autism and how this relates to brain structural and functional development. Currently, no diagnostic or likelihood biomarkers are used clinically for autism, and standard assessments are solely implemented for clinical diagnosis (Ansel et al. Reference Ansel, Posen, Ellis, Deutsch, Zisman and Gesundheit2019; Frye et al. Reference Frye, Vassall, Kaur, Lewis, Karim and Rossignol2019).

Social communication difficulties are core characteristics of autism. Research has focused on understanding differences in processing of social information and how it relates to brain functional development. Face processing is an important component of social perception that is impacted in autism (Golarai et al. Reference Golarai, Grill-Spector and Reiss2006; Nomi & Uddin, Reference Nomi and Uddin2015b). In typically developing individuals the event-related potential (ERP) component known as the N170 is an electrophysiological brain response detected in temporal-parietal electrodes that generally shows a greater amplitude and faster latency for faces compared to objects and is right lateralised in the brain (Bentin et al. Reference Bentin, Allison, Puce, Perez and McCarthy1996; Itier & Taylor, Reference Itier and Taylor2004). Studies have reported N170 amplitude and/or latency differences, or reduced lateralised responses in autism cohorts (McPartland et al. Reference McPartland, Dawson, Webb, Panagiotides and Carver2004; Webb et al. Reference Webb, Dawson, Bernier and Panagiotides2006; McPartland et al. Reference McPartland, Wu, Bailey, Mayes, Schultz and Klin2011; Tye et al. Reference Tye, Mercure, Ashwood, Azadi, Asherson, Johnson, Bolton and McLoughlin2013), although there are inconsistencies in findings across studies. Differences in electrophysiological responses to faces have also been reported in infants with high familial likelihood that subsequently go on to develop autism (Jones et al. Reference Jones, Venema, Earl, Lowy, Barnes, Estes, Dawson and Webb2016). A recent meta-analysis of child and adult studies suggested that N170 latency, but not amplitude, may be a potential biomarker of social communication differences in autism (Kang et al. Reference Kang, Keifer, Levy, Foss-Feig, McPartland and Lerner2018). Specifically, the N170 latency was prolonged in both hemispheres in autism groups compared to control groups, and this was not moderated by age, sex or IQ. Interestingly, this latency difference was specific to faces which may be suggestive of slower processing of faces, while longer latencies in both hemispheres may indicate reduced specialisation within the brain. This fits with the idea that brain development in autism is associated with a differential hemispheric functional specialisation and structural asymmetry. In contrast, Kang et al. (Reference Kang, Keifer, Levy, Foss-Feig, McPartland and Lerner2018) showed no significant difference in N170 amplitude between autism and control groups. However, differences across age groups were observed, such that child study samples generally showed increased amplitude and adults study samples showed reduced amplitude in autism groups compared to controls. This age-related change in amplitude was attributed to changes in processing with development, and also reduced lateralisation and specialisation of face processing. Although the N170 is a robust measure of face processing, whether it meets criteria to be considered a viable and valid biomarker of social communication is still uncertain. While the meta-analysis by Kang et al. (Reference Kang, Keifer, Levy, Foss-Feig, McPartland and Lerner2018) indicated some important findings observed across studies, the latency differences had a small effect size, and showed a considerable amount of heterogeneity. Again, this indicates the need to better examine the heterogeneity to assess whether the N170 can be leveraged to define subgroups within autism, which may better explain inconsistencies in study findings. Many factors are important to take into consideration when considering discrepancies in findings across studies, such as sample bias, sample sizes and age range, as well as differences paradigms, equipment, data collection quality or analysis (Key & Corbett, Reference Key and Corbett2020), and large samples and longitudinal studies will be beneficial in delineating the contribution of these factors to the heterogeneity.

Challenges of diagnostic biomarkers

Clinical complexity and biological heterogeneity within autism almost certainly contribute to challenges in identifying a valid diagnostic biomarker, and to the inconsistencies in diagnostic prediction accuracy across neuroimaging, or ERP study findings. Predefined subgroups may show similar differences in brain structure, brain function or specific cognitive processes such as face processing, which may be related to underlying molecular mechanisms and symptom profiles. To date, few studies have explored stratification of autism subtypes using measures such as neuroimaging, with subgroups mainly identified based on symptom or cognitive measures. Inter-individual differences in changes in neural development across the lifespan within the same individual are also relevant. This highlights the potential importance of the ability of a biological measure to track developmental trajectories longitudinally to assess deviations from typical development, at what stage in development they occur, and whether they continue to deviate, change trajectory or even improve with time for an individual.

Moving towards understanding the biological heterogeneity within autism

The aetiology of autism is still not well defined, although increasingly, evidence from genetic studies suggests the role of biological mechanisms involved in neuronal signalling pathways, synapse development, neurotransmission, gene regulation and immune regulation/inflammation (De Rubeis et al. Reference De Rubeis, He, Goldberg, Poultney, Samocha and Cicek2014; Ansel et al. Reference Ansel, Rosenzweig, Zisman, Melamed and Gesundheit2016; Kumar et al. Reference Kumar, Reynolds, Ji, Gu, Rai and Zhou2019; Baranova et al. Reference Baranova, Dragunas, Botellho, Ayub, Bueno-Alves, Alencar, Papaiz, Sogayar, Ulrich and Correa2020; Satterstrom et al. Reference Satterstrom, Kosmicki, Wang, Breen, De Rubeis and An2020). This divergence in aetiology makes it difficult to assess biological measures of autism as one cohesive group in order to establish valid biomarkers. Researchers are now seeking alternative ways to measure the heterogeneity within autism in order to develop valid stratification biomarkers (Loth et al. Reference Loth, Charman, Mason, Tillmann, Jones and Wooldridge2017; McPartland et al. Reference McPartland, Bernier, Jeste, Dawson, Nelson and Chawarska2020).

In line with this idea, analysis of large multi-site datasets and longitudinal studies, such as the European Longitudinal European Autism Project (LEAP; Charman et al. Reference Charman, Loth, Tillmann, Crawley, Wooldridge and Goyard2017; Loth et al. Reference Loth, Charman, Mason, Tillmann, Jones and Wooldridge2017), are fundamental for the identification of robust and replicable measures as large sample sizes increase statistical power for analysis approaches assessing neural and/or behavioural patterns that can delineate subgroups within autism. Analysis of one time point of the LEAP cohort, in which age ranges from between 6 and 30 years, has shown that brain structural development varies considerably within autism, and this heterogeneity is highly individualised in both children and adults (Zabihi et al. Reference Zabihi, Oldehinkel, Wolfers, Frouin, Goyard, Loth, Charman, Tillmann, Banaschewski, Dumas, Holt, Baron-Cohen, Durston, Bölte, Murphy, Ecker, Buitelaar, Beckmann and Marquand2019). Additionally, both leftward and rightward brain structural asymmetry is observed in autistic individuals, again pointing to differential brain lateralisation (Floris et al. Reference Floris, Wolfers, Zabihi, Holz, Zwiers and Charman2020). Of note, leftward structural asymmetry was largely explained by core symptom severity, and more rightward asymmetry was largely explained by language delay (Floris et al. Reference Floris, Wolfers, Zabihi, Holz, Zwiers and Charman2020). Although this shows heterogeneity in brain structural differences in autism, and how they may relate to clinical characteristics, the small effect sizes suggest that the deviances observed were subtle. Investigation of individual deviances in brain structure and longitudinal follow up will help to better understand differences in developmental trajectories from one autistic individual to another while encompassing both regression and improvement of symptoms. It is important that a stratification biomarker can capture individual developmental trajectories robustly and reliably as we know developmental trajectories of the brain, clinical symptoms and behavioural characteristics may differ from one autistic individual to another (Fountain et al. Reference Fountain, Winter and Bearman2012; Lord et al. Reference Lord, Bishop and Anderson2015). While biological measures may change throughout development and maturation, they may also change over time for subgroups or even at the individual level. For example, while two autistic individuals may present similarly in behaviour or some biological measure at one age, they may not follow the exact same developmental pathway. Along with differences attributed to age and developmental trajectory, it important to consider other factors that may contribute to the heterogeneity in autism, such influence of genetics and environmental factors, and the presence and onset of co-occurring conditions.

Incorporation of underlying biological mechanisms of autism

As genetic research studies have provided key information about biological mechanisms that underlie the various atypicalitys observed in autism, a genome first approach is another potential way of stratifying within autism. Examination of rare genetic variants or genetic syndromes associated with the development of autism can give insight into a cohort who, hypothetically, have a clearer biological mechanism underlying their clinical presentation. This provides a way to link disrupted underlying biological pathways to cognitive, behavioural, and clinical characteristics, and subsequently identify useful stratification biomarkers for genetically defined subgroups. Rare copy number variants (CNVs) are detected in around 7.1–7.8% and nucleotide variants in around 12.6% of autistic individuals (Kushima et al. Reference Kushima, Aleksic, Nakatochi, Shimamura, Okada and Uno2018; Husson et al. Reference Husson, Lecoquierre, Cassinari, Charbonnier, Quenez and Goldenberg2020). Many of the genes impacted by CNVs or single nucleotide variants associated with the development of autism converge to functional gene networks allowing us to examine how symptoms, behaviours and brain development relate to gene functions (Pinto et al. Reference Pinto, Delaby, Merico, Barbosa, Merikangas and Klei2014).

The genetics of autism is complex, and many factors are also important to consider such as common and rare genetic variation, incomplete penetrance, variable expressivity, multiple genetic hits, and gene-gene or gene-environment interactions. It has also been argued that we are currently lacking evidence to support the specificity of rare variants to autism alone (Myers et al. Reference Myers, Challman, Bernier, Bourgeron, Chung, Constantino, Eichler, Jacquemont, Miller, Mitchell, Zoghbi, Martin and Ledbetter2020). For example, some rare CNVs have been associated with both autism and schizophrenia, as well as intellectual disability (Kushima et al. Reference Kushima, Aleksic, Nakatochi, Shimamura, Okada and Uno2018). The shared genetics across neurodevelopmental and neuropsychiatric disorders (NDDs) indicates disruption to shared underlying biological pathways. However, we do not fully understand the factors that explain how or why one CNV carrier may confer greater likelihood of developing schizophrenia, while another may develop autism. Given the overlap in genetics, biological pathways and symptom domains with other NDDs, there is the question of whether a stratification biomarker should be specific to underlying biology and linked symptoms or behavioural domains rather than being specific to diagnosis. The benefit is that this could also take into account comorbidities, which are often overlooked in research studies. Regardless, clinical research may benefit immensely from cross-disorder approaches to biomarker assessment, as it is important to ensure that biomarkers are either disorder specific or symptom specific. Importantly, this is not to replace standard clinical diagnosis, but rather that stratification biomarkers would aid decisions about best care and treatment plans after diagnosis. However, this challenges the conventional, classification-based approaches in psychiatry where treatment is diagnosis based.

The Research Domain Criteria (RDoC) is a research framework for understanding variability across behavioural dimensions and neural systems in mental health disorders. This framework may be a useful approach to understanding and developing biomarkers for symptoms and behaviours related to underlying neurobiology observed in autism, and NDDs more generally. Although specific genes or genetic conditions, are not currently part of the RDoC framework, it may be useful for genetic conditions associated with neurodevelopmental or psychiatric clinical symptoms (Salles et al. Reference Salles, Lacassagne, Benvegnu, Berthoumieu, Franchitto and Tauber2020), as it may progress research and our understanding of how neurobiological mechanisms are linked to behaviour. In the context of stratification of autism subtypes, the strong links between genetics, underlying biology, neural development and clinical and behavioural outcomes, suggest that the RDoC framework could be used to explore the vast heterogeneity within autism using a mechanism-based approach to elucidate subgroups. Although clinical assessments diagnose symptoms and conditions, they do not offer information about the underlying mechanisms that may be altered. The advantage of integrating a mechanistic approach is that biological measures incorporate the divergent aetiology observed in autism, which could be important for clinical decisions about effective treatment plans for an individual.

A multi-biomarker approach

The clinical and genetic heterogeneity within autism also poses the question as to whether one single stratification biomarker can account for the wide range in variability in observed in autism, or whether a set of measures may be more realistic to help stratify and subgroup autistic individuals (Bridgemohan et al. Reference Bridgemohan, Cochran, Howe, Pawlowski, Zimmerman, Anderson, Choueiri, Sices, Miller, Ultmann, Helt, Forbes, Farfel, Brewster, Frazier and Neumeyer2019; Vargason et al. Reference Vargason, Grivas, Hollowood-Jones and Hahn2020). Defining a set of stratification biomarkers may be a plausible solution for a clinically and aetiologically heterogeneous condition. Deep phenotyping studies using multi-methodological approaches are currently being applied to assess the validity of multiple measures as potential stratification biomarkers, such as genetic and proteomic measures, eye-tracking measures of neurocognition, MRI and EEG (Charman et al. Reference Charman, Loth, Tillmann, Crawley, Wooldridge and Goyard2017; Loth et al. Reference Loth, Charman, Mason, Tillmann, Jones and Wooldridge2017; McPartland et al. Reference McPartland, Bernier, Jeste, Dawson, Nelson and Chawarska2020). Stratification biomarker research remains in the preliminary stages and many potential measures are yet to be replicated across samples before validation as a robust biomarker for autism. It is also yet to be determined whether some potential stratification biomarkers are specific to autism, or to underlying biology and related behaviour or symptoms that may also be impacted in other NDDs. This will only be confirmed by assessing biomarkers across different NDDs.

A multi-biomarker approach may enhance the ability to describe the heterogeneity and define subgroups using an ‘autism stratification biomarker profile’ (Fig 1). This theoretical diagram illustrates how a stratification biomarker profile based on multiple measures may be a more appropriate approach for such a complex genetic, biological and clinical condition, and how it is possible that one biomarker may be more relevant for one subgroup compared to another. This approach would ideally identify and encompass different patterns of clinical symptoms, cognitive, emotional, behavioural and neurodevelopmental characteristics, with links to genetics or underlying neurobiology. However, before this could be implemented, clinical research needs to first identify and validate which measures can be used to elucidate subgroups, assess if they can measure efficacy of treatments within subgroups, and finally then help to refine individual treatment plans to improve mental health outcomes for individuals as part of a combined clinical assessment and multi-biomarker profile. This would rely on clinics having access to genetic testing and EEG and/or MRI facilities, and standardised analysis procedures. The outcome of this biomarker profile, along with clinical assessments would then facilitate the choice of treatment plan for that individual. It must be noted that there are some potential limitations to using some of the proposed measures. For example, MRI may be challenging for some individuals, such as those with claustrophobia, auditory sensory issues, or even those who find it difficult to stay still. However, the advancements in techniques such as silence sequence scans or the use of sleep scan procedures (either during natural sleep or with the use of melatonin) may be possible solutions. Often in research, strategies are used to help familiarise individuals with the MRI environment before their scan, such as playing recordings of the sounds of the scans, which can improve their experience and data quality. Another potential barrier is the expense and access to MRI facilities for some clinics. As for EEG, some individuals with sensory issues may also find this less tolerable. In research use of storybooks and practising at home are often used as strategies to familiarise and prepare participants, which often improve feasibility. Stratification biomarker research has some way to go before being validated for clinical use in this context. However, if MRI or EEG measures prove to be a valid and useful biomarker that can improve outcomes for individuals, this may be a plausible argument for its routine use. The use of the multi-biomarker approach will require the development of machine algorithms to a range of biological, neural, behavioural and clinical measures to improve diagnostics and inform best treatment pathways (Uddin et al. Reference Uddin, Wang and Woodbury-Smith2019). The theoretical diagram in Fig 1 illustrates how combination of clinical assessments with an in-depth biomarker profile may help to facilitate the choice of treatment plan and improve outcomes and quality of life for those autistic individuals who are severely impacted.

Fig. 1.

This theoretical diagram illustrates how a biomarker profile based on multiple measures such as molecular or genetics, brain structure and function, may help to define subgroups within autism. The combination of in-depth phenotyping from clinical assessments and biomarker profile may help to improve treatment plans and outcomes for pre-defined subgroups within autism. (This figure was created with BioRender.com).