Molecular Psychiatry (2013) 18, 435 -- 442

& 2013 Macmillan Publishers Limited All rights reserved 1359-4184/13

www.nature.com/mp

EXPERT REVIEW
Translational approaches to the biology of Autism:
false dawn or a new era?
C Ecker,1, W Spooren2 and DGM Murphy1

Discovering novel treatments for Autism Spectrum Disorders (ASD) is a challenge. Its etiology and pathology remain largely
unknown, the condition shows wide clinical diversity, and case identification is still solely based on symptomatology. Hence
clinical trials typically include samples of biologically and clinically heterogeneous individuals. ‘Core deficits’, that is, deficits
common to all individuals with ASD, are thus inherently difficult to find. Nevertheless, recent reports suggest that new
opportunities are emerging, which may help develop new treatments and biomarkers for the condition. Most important, several
risk gene variants have now been identified that significantly contribute to ASD susceptibility, many linked to synaptic
functioning, excitation--inhibition balance, and brain connectivity. Second, neuroimaging studies have advanced our
understanding of the ‘wider’ neural systems underlying ASD; and significantly contributed to our knowledge of the complex
neurobiology associated with the condition. Last, the recent development of powerful multivariate analytical techniques now
enable us to use multi-modal information in order to develop complex ‘biomarker systems’, which may in the future be used to
assist the behavioral diagnosis, aid patient stratification and predict response to treatment/intervention. The aim of this review
is, therefore, to summarize some of these important new findings and highlight their potential significant translational value to
the future of ASD research.

Molecular Psychiatry (2013) 18, 435--442; doi:10.1038/mp.2012.102; published online 17 July 2012
Keywords: Autism spectrum disorder; biomarkers; drug development; genetics; neuroimaging

INTRODUCTION between identical twins reported at nearly 90% in some studies.3,4

Autism Spectrum Disorders (ASD) are a group of conditions in The high heritability of the condition has been linked to several
which heterogeneity, both causative and phenotypic, is emerging risk gene variants that significantly contribute to the genetic
as a dominant theme.1 This poses a complex challenge to architecture of ASD (reviewed in Abrahams et al.1). These include
developing new treatments, and biomarkers. Also, case identifica- common as well as rare genetic variants, which may plausibly help
tion is still solely based on symptomatology (that is, impaired guide the development of new pharmacological targets and
social communication, social reciprocity and repetitive/stereotypic interventions in the future.
behavior2). Hence, clinical studies typically include samples of Despite the high heritability of the condition, the number of
biologically heterogeneous cases, and this has hampered studies common genetic variants that have been reliably associated with
of clinical outcomes as well as research into the specific molecular ASD is surprisingly small. Such common variants are generally
and genetic underpinnings of the condition. However, several identified by genome-wide association studies (GWAS), which
lines of evidence are now emerging that may unite findings from examine the association of a large number of common
separate disciplines (for example, genetics, neuropathology and polymorphisms with a trait or groups of individuals (for example,
neuroimaging). These seem to converge in suggesting a common ASD vs ‘neurotypicals’). For example, Wang et al.5 found a
etiological pathway for ASD, namely: defective synaptic function- significant association between ASD and cadherin 10 (CDH10 and
ing, excitation--inhibition balance and brain connectivity. Further, CDH9) genes on region 5p14.1, which code for neuronal cell-
they present new (and tractable) opportunities to identify adhesion molecules. Another genome-wide linkage study has also
biomarkers that can be used to fractionate the disorder and identified a potential role for semophorin 5A (SEMA5A), a gene
underpin the discovery of new drug targets. Here, we examine the implicated in axonal guidance and reported to be downregulated
genetic and neurobiological ‘building blocks’ of ASD, and suggest in ASD.6 It has generally been assumed that the relative paucity of
a pathway from abnormalities in the synapse to the formation of significant genome-wide results may be a consequence of the
large-scale cortical systems (Figure 1). small effect sizes attributable to any particular gene, and
conservative statistical thresholds resulting from multiple compar-
ison corrections.1 More recent studies, however, also highlight the
importance of a number of distinct and individually rare genetic
GENETIC MODELS OF ASD AND THEIR IMPORTANCE FOR variants for ASD, which significantly contribute to the genetic
DRUG DEVELOPMENT architecture of the condition.
It has long been known that ASD is among the most genetically Rare copy number variants (CNVs), which are insertions or deletions
determined neuro-psychiatric conditions, with concordance rates of relatively large DNA fragments, are often not transmitted from the

1
Department of Forensic and Neurodevelopmental Sciences, Institute of Psychiatry, King’s College, London, UK and 2F. Hoffmann-La Roche Ltd., Psychiatry Disease Area, Basel,
Switzerland. Correspondence: Dr C Ecker, Department of Forensic and Neurodevelopmental Sciences, PO50, Institute of Psychiatry, King’s College, De Crespigny Park, London,
SE5 8AF, UK.
E-mail: [email protected]
Received 22 February 2012; revised 29 May 2012; accepted 31 May 2012; published online 17 July 2012
 Translational approaches to Autism
C Ecker et al

436

Figure 1. Synaptic functioning and brain connectivity in Autism Spectrum Disorders (ASD)---from the molecular level to the neural systems level.

parent but instead occur de novo. Such rare CNVs are currently laminar organization, indicating a role for this CNVs being the
thought to account for B10% of cases with idiopathic autism, that construction of neural circuits.14 The cortical pattern of CNTNAP2
is, those with no known cause (unlike ASD occurring in Fragile X or expression particularly implicates frontal lobe connectivity,
Rett syndrome). To date, the rare mutations have predominantly which has been reported to be atypical not only in ASD but
been observed in genes encoding for synaptic cell adhesion also in a variety of ASD-related conditions, such as attention
molecules, which have a crucial role in synaptogenesis and neuronal deficit-hyperactivity disorder15 (also discussed below). This finding
differentiation. For example, well-replicated findings implicate CNVs is of particular importance as the genetic mediation of disconnec-
in genes encoding for a family of proteins known as neuroligins, tion across multiple phenotypes suggests that certain genetic
neurexins and SHANK3. variants, and their associated brain phenotypes, may not be
specific to ASD. Instead, they may increase the overall risk of
ASD by affecting isolated neural systems that mediate specific
Neuroligins and neurexins autistic symptoms.
One of the first CNVs groups found to be associated with ASD
included neuroligin (NLGN) genes such as neuroligin 3 (NLGN3) and
neuroligin 4 X-linked (NLGN4X).7 Neuroligins are a family of cell SHANK3, glutamate and GABAergic signaling in ASD
adhesion proteins whose members are localized postsynaptically, Another promising gene, which has been linked to ASD, is
and are believed to be involved in the formation and consolidation SHANK3.16 SHANK3 codes for a key postsynaptic density protein at
of synaptic contacts.8 Neuroligins form a transsynaptic contact with glutamatergic synapses, which is believed to function as a scaffold
presynaptic neurexins, which is thought to trigger synaptogenesis, for the assembly of the postsynaptic signaling complex. Mice with
and abnormalities in this process have been linked to a deletion of the SHANK3 gene have been reported to exhibit
ASD.9 Neuroligins (and neurexins) can induce the formation of autistic-like symptoms such as deficits in social interaction and
both excitatory and inhibitory synapses, depending on their repetitive behavior.17 Further, SHANK3 modulates striatal size
specific subtypes (reviewed in Dalva et al.10). Furthermore, it has (larger in knockout mouse) as well as the anatomy of cortico-
been demonstrated in neuroligin-knockout animals that striatal circuits known to be affected in ASD18---and for which
depletion of the gene leads to a shift in the balance between there is preliminary evidence to suggest that glutamate metabo-
excitation and inhibition.11 This is of importance because increas- lism is abnormal in individuals with ASD.19 In humans, it has also
ing evidence points to a role for disturbances to the excitation-- recently been demonstrated that mutations in SHANK3 strongly
inhibition balance being implicated in ASD (Rubenstein and affects the development and morphology of dendritic spines and
Merzenich12). reduces synaptic transmission in mature neurons.20 Hence it has
ASD has also been linked to another member of the neurexin been suggested21 that variation in SHANK3 may directly modulate
family, contactin associated protein-line 2 (CNTNAP2). CNTNAP2 some aspects of brain phenotype in ASD.
has been shown to modulate language function in ASD (and other In addition, the direct effects of SHANK3 may also be combined
conditions), and may modulate long-range brain connectivity. with other genetically (and/or environmentally) determined risk
CNTNAP2 messenger RNA is significantly enriched in the factors affecting the balance of excitation--inhibition---for example,
developing human brain in the frontal and temporal lobes, as the balance of glutamate and g-aminobutyric acid (GABA)).22
well as in striatal circuits and in the frontal cortex of the adult GABA is an inhibitory amino acid neurotransmitter synthesized by
brain.13 In humans, these regions support speech and language decarboxylation of glutamate by the enzyme glutamic acid
learning, which are frequently impaired in ASD. On the molecular decarboxylase (GAD). In the adult brain, GAD exists in two major
level, CNTNAP2 is thought to have a major role in assisting isoforms, GAD65 and GAD67, which are products of two
interactions important for cellular migration and subsequent independently regulated genes located on chromosomes 2 and

Molecular Psychiatry (2013), 435 -- 442 & 2013 Macmillan Publishers Limited
 Translational approaches to Autism
C Ecker et al

437
10, respectively.23 In ASD, chromosome 10 has been shown to ASD---A MULTI-ETIOLOGICAL DISORDER OF DEFECTIVE
reach genome-wide significance in linkage studies, and particu- SYNAPTIC STRUCTURE AND FUNCTIONING
larly in the region 10p12.1, which also encodes GAD65.24 It has Although several CNVs are now known to mediate the risk of ASD,
further been shown that individuals with ASD have a B50% it is likely that each CNVs does not act in isolation. Instead, several
reduction in protein levels of the enzymes GAD65 and GAD67 in significant ‘gene clusters’ have now been identified, which can be
parietal (BA40) and cerebellar cortices,20 which are both grouped together according to their functional involvement in
regions where individuals with ASD show differences in brain ASD. For example, Gilman et al.27 identified several large
anatomy from controls. During development, GABA can also act as functional network of genes affected by rare de novo CNVs in
a trophic factor and influence neuronal proliferation, migration, ASD using a network-based analysis of genetic associations
differentiation, synapse maturation and cell death.21 Hence, (NETBAG). These networks included many of the CNVs discussed
SHANK3 (alone or in combination with other factors) may affect above, and primarily include genes coding for cell adhesion (for
synaptic functioning and brain connectivity, and increase the risk example, NRXN1, NLGN3) and scaffolding proteins (for example,
for ASD. SHANK2/3). Atypical cell adhesion and scaffolding affect the
formation and consolidation of synaptic contacts, and hence have
Implications for research and drug development an impact on the formation of local neuronal connections on
the microscopic level. Synaptogenesis has therefore become a
These findings coming from genetic investigations have important ‘common theme’--- linking CNVs associated with ASD function-
implications for ASD. First, it seems that very few genetic variants ally---and directly impacts on the structural architecture of
are ‘causal’ for ASD. Instead, they affect a wide variety of synaptic terminal on the microscropic level. The analysis of the
phenotypes with variable symptom expressivity, which may or functional impact across genes and/or CNVs supports the
may not meet the clinical cutoffs for ASD (that is, the ‘broader hypothesis that perturbed synaptogenesis may be a ‘core deficit’
autism phenotype’).25 Thus, there is a strong need for detailed
of ASD, that is, common to all individuals with autistic symptoms
phenotypic studies not only of patients with autism but also of
and traits (see also28). In addition, the findings of genetic studies
unaffected individuals with more or less autistic traits who harbor
imply that synaptic structure and functioning---mediated by
such rare potentially causal mutations. For example, a recent study
synaptic transmission---is affected in ASD.
has also shown that a distinct autism-related CNTNAP2 ‘risk allele’
Evidence for perturbed synaptic functioning come from several
is associated with reduced functional connectivity in frontal lobe
proton magnetic resonance spectroscopy studies. For example,
networks, regardless of whether participants were autistic or
individuals with ASD show differences in N-acetyl-aspartate (a
‘neurotypical’.26 The findings of this study are important as they
measure of neuronal density and mitochondrial function)
imply that some genetic variants may predispose individuals to
concentration,29 and in choline-containing compounds (Cho)---a
ASD (that is, increase the risk of ASD) without necessarily being
measure of membrane turnover. However, recent studies have
‘causal’ for the condition (that is, lead to clinical diagnosis of ASD).
Such noncausal genetic variants may also affect the functional and also highlighted the importance of abnormalities in glutamatergic
structural makeup of particular brain regions (or neural systems) neurotransmission in ASD, which can now be measured using
that contribute to the wider brain phenotype of ASD. Investiga- proton magnetic resonance spectroscopy. For instance, in adults
tions of a similar study design, which investigate the association with ASD, glutamate concentrations were found to be increased in
between genetic variants and autistic traits across various the amygdala--hippocampal regions in adults,19 but to be
conditions---and in neurotypicals---are hence of high importance decreased in many brain regions in children with ASD.30 Thus,
for the future, as this will allow us to link specific aspects of brain synaptic defects in ASD may not be confined to their structural
structure, function and connectivity to specific CNVs that increase architecture but also to synaptic inhibition and excitation
autism susceptibility. mediated by glutamate and GABA. The balanced interaction
The existence of specific ASD genotypes also implies that these between glutamate and GABA transmission is essential for
could be rescued (that is, ‘manipulated’) by specific, targeted regulating cognition (for example, learning and memory) and
molecular treatment. There is, therefore, now an opportunity to emotional behaviors, and an imbalance between gluatamate
exploit these new findings and make progress on the develop- excitation and GABA inhibition, leading to hyperexcitation, has
ment of new therapies for ASD, including both children and been linked to ASD.31--34 There is recent in vivo evidence from a
adults. For example, by specifically targeting CNVs of known Positron Emission Tomography study demonstrating proof of
developmental genes, efforts can be focused onto precisely those concept that individuals with ASD have significant differences
elements where these genes would be predicted to make the from controls in GABA a-5 receptor binding.35 Last, there is
most impact. So far, drug discovery in ASD has also been preliminary evidence implicating abnormalities in serotonergic
hampered by the availability of valid human cellular assays that system to ASD. For example, a significant proportion of ASD
recapitulate normal and diseased neural function in vitro. individuals may have hyperserotonemia (Hranilovic et al.36); and
Emerging data from both genetic and clinical studies in addition there are significant associations between ASD and genetic
to rodent models of disease might, however, soon be used to polymorphisms for serotonin synthesis,37 transporters38 and
configure cellular assays, which model the etiology of ASD. One receptors.39 Additionally, neuroimaging studies report that
approach being pioneered as part of a large-scale EU Innovative individuals with ASD have significant differences in serotonin
Medicines Initiative on Autism (EU-AIMS; http://www.eu-aims.eu) synthesis and reductions in serotonin2A receptor binding, and the
is to use induced human pluripotential stem cells from specific 5HT transporter, in brain regions involved in social communication
cohorts of patients to generate (in a culture dish) human neurons (for example, cingulate cortices).40--42 Furthermore, very recent
carrying the specific genotype of ASD individuals, and to examine evidence shows that abnormal brain activity during facial emotion
the effect of these in mouse models. If successful, this approach processing in individuals with ASD is modulated by cortical
will allow us to target specific aspects of neurobiology that are serotonin levels, which may underpin some of the impairments in
most atypical in ASD, rapidly ‘assay’ potential new treatments, and social functioning.43 Although it has previously been suggested
translate the findings from ‘bench to bedside’ (and back). Also, this that selective serotonin reuptake inhibitors may not be effective in
approach may be used to find clusters of genotypes with similar the treatment of all individuals with ASD,44 it remains to be
neuronal phenotypes, which could then be used for case established if such small effect sizes are due to the large degree of
stratification in clinical studies and to evaluate the effect(s) of phenotypic heterogeneity within the ASD population. Thus,
specific, targeted molecular treatment. individual’s response to modulation of 5HT may potentially be

& 2013 Macmillan Publishers Limited Molecular Psychiatry (2013), 435 -- 442
 Translational approaches to Autism
C Ecker et al

438
used as a possible stratification tool in the future and to elucidate fronto-temporal regions59,60 and the caudate nuclei.18,62 It has also
the various biological phenotypes of the condition. been suggested that neuroanatomical differences in specific brain
structures mediate specific clinical symptoms. For instance,
abnormalities in; (1) Broca’s and Wernicke’s area have been
CORTICAL NEUROPATHOLOGY OF ASD---FROM SYNAPSES associated with social communication and language deficits;63 (2)
TO LOCAL CIRCUITS fronto-temporal regions and the amygdala have been related to
Synaptic defects mediated by genetic variation in ASD not only abnormalities in socio-emotional processing;64,65 and 3) the
affect their structural architecture, but also affect the formation of orbitofrontal cortex and the caudate nucleus (that is, fronto-striatal
local neuronal circuitry. These, in turn, constitute the cyto- system) may mediate repetitive and stereotyped behaviors.66 These
architectural and microstructural makeup of neocortical regions. studies were important first steps that add weight to the suggestion
For example, GABA---mediated by genes encoding for enzymes that differences in brain anatomy observed in ASD underpin
GAD65/67, which have both been associated with ASD (discussed specific clinical symptoms and traits.
above)---has a crucial role as ‘trophic’ factor during neurodevelop- However, despite growing evidence for the involvement of
ment.45 However, GABA not only affects synaptic structure and specific brain regions in ASD, many previous findings are
functioning during neurodevelopment but also has an important nonreplicated. For example, cerebellum and amygdala have been
role in the formation and functioning of local microcircuitry in the variously reported to be normal, smaller and increased in size. For
mature brain. Here, GABAergic interneurons are an integral part of a long time, it was believed that this variability arose because
cortical ‘minicolumns’, the basic architectonic and physiological most studies were of relatively small samples that differed in
elements of the neocortex,46 and are located mostly in the so- several key aspects within and across subject groups (for example,
called peripheral neuropil space surrounding the column ‘core’. In diagnostic criteria, intelligence quotient and age). However, recent
ASD, minicolumns have been reported to be more numerous and work demonstrates that the effect sizes for neuroanatomical
narrower than in neurotypicals, which is also associated with differences between individuals with ASD and neurotypicals
reduced peripheral neuropile space.47,48 Thus, ASD may be linked remain moderate even when samples are large and well-
to a reduction in GABAergic inhibitory activity and an imbalance characterized. For instance, Ecker et al.55 investigated neuroana-
of inhibition and excitation, which in turn may explain some tomical differences in a large sample of male adults with ASD and
clinical symptoms of ASD (for example, increased incidence of matched ‘neurotypicals’ (N ¼ 176) using a multi-center acquisition
seizures and hypersensitivity to visual/auditory stimulation).12 paradigm. Despite the large sample size, however, relatively few
This example demonstrates how genetic variation associated with clusters survived correction for multiple comparisons (for example,
ASD can cause atypical development and morphology of synaptic significant increases in volume of the anterior temporal and
terminals on the microstructural level, which then leads to altered frontal lobe in ASD relative to controls). This may indicate that (1)
neuronal organization and synaptic transmission in local neuronal the clinical diversity of the behaviorally defined autistic phenotype
circuits (for example, within minicolumns). In turn, altered is also reflected on the level of brain anatomy (that is, multiple
neuronal organization will affect the cytoarchitectural makeup, brain phenotypes) so that increasing sample sizes will not
gross anatomy and function of individual brain regions, which has necessarily improve effect sizes. Alternatively, it may suggest that
been investigated in ASD using various neuroimaging techniques. (2) individuals with ASD have differences in brain anatomy that are
difficult to describe using conventional analytical approaches.
More recent theoretical models suggest the need to consider
THE ANATOMY OF THE BRAIN IN ASD ASD as a disorder of several large-scale neurocognitive networks.
It is well established that ASD is accompanied by differences in Regional or voxel-level analytical methods, which rely on
brain anatomy (also reviewed in49). During early childhood, these conservative statistical thresholds mandated by the large number
differences are most prominent on the global level and include of voxels compared between groups, may not be optimal for
differences in overall brain growth trajectory of total gray- orwhite- detecting differences that are theoretically expected to be subtle
matter volume (Courchesne et al.50), head circumferences (Lainhart and spatially distributed (that is, standard voxel-level approaches
et al.51) and/or surface area (Hazlett et al. 52). During later childhood, powerfully detect regional differences, but they are not are not
adolescence and adulthood, however, ASD are associated with suitable for systems-level questions). Multivariate or multi-voxel
abnormalities in specific spatially distributed neural systems that approaches, which are statistically more powerful and hence offer
may mediate specific symptoms and traits.49 During early postnatal higher exploratory power, are therefore becoming increasingly
life, the brain in ASD may undergo a period of precocious popular to examine the brain in ASD. For example, the recent
accelerated growth, which may then be followed by an atypically introduction of pattern classification techniques (discussed below)
slow or arrested growth during childhood, so that no global has proven invaluable in detecting brain regions that distinguish
differences are generally observed in adulthood.53 In childhood, individuals with ASD from ‘neurotypicals’ on the basis of several
increased brain size in ASD affects both gray and white matter,54 neuroanatomical features (for example, cortical thickness, regional
and may be mainly driven by an increase in surface area---rather brain volumes and so on.). Taken together, these findings suggest
than cortical thickness.52 Normal intelligence adults with ASD, that instead of asking the question ‘what can ASD tell us about its
however, do not differ significantly from ‘neurotypicals’ in whole neuroanatomy?’, we ought to be examining what neuroanatomy
brain gray/white matter volume, but do in spatially distributed can tell us about ASD. In this undertaking, the development of
patterns of significant gray- or white-matter differences, indicating novel analytical approaches to identify the potentially multiple
the wider neural systems implicated in ASD.55 It has also been brain phenotypes of ASD as well as their genetic and molecular
suggested that the overgrowth is idiosyncratic for different lobes of underpinnings will be essential. It is imperative that future
the brain, with frontal and temporal lobes being more affected than investigations do not only target the genetic underpinnings of
occipital and parietal lobes (Carper et al.56). Both the temporal and differences in brain volume in ASD, but also associate genetic
the frontal lobe are also brain areas that mature relatively late markers with measures of brain connectivity in several large-scale
during normal development,57 which further strengthens the ‘neural systems’.
hypothesis that the altered time course of brain maturation may
lead to neuroanatomical differences in multiple neural systems
rather than isolated regions. Across studies, the most replicated ASD---A ‘NEURAL SYSTEMS’ CONDITION
structural differences associated with ASD have been described It has been noted that the altered neurodevelopmental trajectory
in the cerebellum,58 the amygdala--hippocampal complex,59--61 is likely to interfere with brain connectivity in ASD (recently

Molecular Psychiatry (2013), 435 -- 442 & 2013 Macmillan Publishers Limited
 Translational approaches to Autism
C Ecker et al

439
reviewed in Vissers et al.67), as the time window of overgrowth unlikely that ASD can be linked to a single biomarker (that is, a
coincides with the period when synaptogenesis, dendritic growth single gene or brain region) across the neurodevelopmental time
and myelination are at their peak (reviewed in Courchesne et al.68 course. Instead, ASD biomarkers are most likely to be multivariate
and Rippon et al.69). Moreover, the maturation of higher-order and complex, encompassing data from different aspects of biology
cortical systems, such as the frontal and temporal lobes, rely on as well as genetics. However, no single analytical framework has so
the earlier maturation of lower-order and phylogenetically older far been powerful enough to establish such complex ‘biomarker
cortical systems (for example, somatosensory and visual cortices57) systems’.
so that any developmental dysregulation(s) during this critical Recent advances in analytical techniques now make it possible
time will not only affect the neural architecture of isolated brain to utilize such complex, multivariate data in order to make a
regions (and their local connectivity), but also the formation of prediction. In the context of brain imaging, these techniques have
their global circuitry.68 Thus, ASD is most likely a ‘neural systems’ been described as ‘brain-reading’ or ‘brain-decoding’ methods,80
condition that is mediated by abnormalities in spatially distrib- and belong to a broad group of techniques collectively known as
uted, large-scale cortical networks rather than isolated brain ‘machine learning’. The basic idea of machine learning is to train a
regions. ASD has therefore also been referred to as a ‘develop- computer algorithm to identify a complex pattern of data that can
mental disconnection syndrome’.70 then be applied in new individuals to make a prediction. Training
Evidence for atypical structural connectivity in ASD comes from usually occurs in a well-characterized sample by finding a
an increasing number of neuroimaging studies measuring white- boundary or ‘hyperplane’ that best discriminates between
matter anatomy. For example, prior studies found that individuals different classes (for example, patients and controls). Once the
with ASD have significant differences in white-matter volume,60,65 classifier is ‘trained’, it can subsequently be used to predict group
and microstructural integrity---as measured by diffusion tensor membership of a new test example (for example, new individual
magnetic resonance imaging.71--75 Furthermore, it has been with unknown group membership) (see Figure 2). A key feature of
reported that individuals with ASD undergo abnormal postnatal pattern classification is their potential to detect global, complex
white-matter development. Such prior reports mostly highlight and potentially multimodal patterns of abnormalities that cannot
significant increases in white matter during early childhood, which be efficiently identified with univariate methods (for example,
may precede the abnormal pattern of growth in gray matter.54 In general linear model). Machine-learning approaches are therefore
adults, however, ASD is associated with a pattern of regional particularly suited to explore biomarkers for ASD.
reductions in white matter, which suggests that white-matter A growing number of recent publications have therefore started
differences may also result from differences in neurodevelop- to explore the diagnostic (that is, predictive) value of various
mental trajectories. Several core deficits seen in ASD have also measures of brain anatomy, functioning and connectivity for ASD.
been associated with atypical connectivity of specific white-matter Most of the initial studies were based on measures of brain
fiber tracts. For example, (1) diffusivity measures of the corpus anatomy. Using a common variant of machine-learning, the so-
callosum are significantly correlated with reduced processing called support vector machine, Ecker et al.81 explored the
speed in performance intelligence quotient tests in ASD;74 (2) the diagnostic value of whole-brain structural MRI scans measuring
severity of social impairments in ASD are related to abnormal regional gray- and white-matter volume. In this sample, support
diffusion anisotropy in fibers of the superior cerebellar peduncle73 vector machine correctly identified individuals with ASD based on
or (3) that individuals with ASD have significant differences in the their brain anatomy with B90% specificity and sensitivity. In
anatomy and maturation of limbic tracts, which predominantly addition to the overall binary classification (that is, being autistic or
relay information underpinning socio-emotional processing.72 not), support vector machine provided a ‘test margin’ for each
Thus, although it is difficult to link specific cognitive functions to subject indicating how ‘prototypical’ the test example is of each
differences in white-matter anatomy, altered brain connectivity in class, which is related to the confidence with which a new
addition to structural gray-matter differences may explain some of participant can be classified. These test margins were positively
the behavioral features typically observed in ASD. correlated with the severity of current autistic symptoms---
Altered connectivity in ASD has also been reported by functional suggesting that support vector machine may be able to measure
magnetic resonance imaging (MRI) studies, which have led to one ASD along a continuum based on neuroanatomy. The develop-
of the dominant theories regarding brain connectivity in ASD; ment of a quantitative ‘dimensional’ approach (rather than a
namely, that there is long distance under-connectivity and local categorical classification) is of importance, as any usable biomarker
over-connectivity of the frontal cortex.68,76 Reduced long-range needs to provide a quantitative measure of a pathogenic process
cortical functional connectivity has been reported predominantly in rather than simply testing for the existence of a particular
frontal regions using a variety of fMRI paradigms including pathological phenotype. These original reports, which provided a
executive functioning,77 working memory for faces78 and facial- proof of concept, are now supported by several other studies using
affect processing.79 Less evidence has been provided for the different imaging modalities; and which also examined different
concept of local under-connectivity, which may be mainly owing to age groups. Using structural MRI data, high classification accuracies
the fact that it is difficult to measure. However, these MRI findings have been replicated in children and adolescents with ASD,82,83 as
complement genetic investigations suggesting that atypical con- well as females with the condition.84 It has also been demonstrated
nectivity on the cellular level (that is, defect synaptic functioning) that ASD may be detected using measures of brain functioning,85
may affect interregional connectivity on the ‘systems level’ in ASD. as well as using functional and structural connectivity indices.86--88
Notably, a recent study, which also incorporated genetic informa-
tion into the classifier, proved highly accurate in distinguishing
AUTISM BIOMARKERS individuals with Asperger syndrome from high-functioning aut-
Owing to the new insights described above, there is now a search ism.89 In summary, initial work on the ability of biomarkers to
for biomarkers that can be used to assist the behavioral diagnosis, classify people having autism looks promising, but several crucial
aid patient stratification and predict response to treatment/ questions need to be addressed first before these novel methods
intervention. find their way into clinical practice. One of these crucial issues is
So far, the discovery of biomarkers has been hindered by the the clinical specificity.
complexity of the condition, as ASD has multiple causes (see Although the established methods seem highly successful at
above), comorbid conditions and varies in the type and severity of distinguishing individuals with ASD from ‘neurotypicals’, it is currently
symptoms expressed. In addition, ASD is a neurodevelopmental unknown how well the biomarkers are able to separate ASD from
condition and phenotypes are likely to vary with age. Therefore, it is related comorbid conditions (for example, deficit-hyperactivity

& 2013 Macmillan Publishers Limited Molecular Psychiatry (2013), 435 -- 442
 Translational approaches to Autism
C Ecker et al

440

Figure 2. Pattern classification using support vector machine based on structural magnetic resonance images in Autism Spectrum Disorders
(ASD). The classifier is initially ‘trained’ on a well-characterized sample of individuals with ASD and controls. Training results in a ‘discrimination
map’ indicating regions that can be used to discriminate between ASD and controls. This map can then be used to predict group membership
of a new test example (for example, ASD or control).

disorder, social anxiety or obsessive--compulsive disorders). specific ASD genotypes implies that these may be rescued (that
Preliminary evidence suggest that the proposed methods are is, ‘manipulated’) by specific, targeted molecular treatment. There
indeed specific to ASD rather than neurodevelopmental condi- is therefore now an opportunity to exploit these new findings
tions in general,90 but conclusive evidence is still required. Also, it and make progress on the development of new therapies for ASD,
is currently unknown if the proposed ASD biomarkers will be including both children and adults. There is no doubt that
able to deal with the clinical heterogeneity of the condition. So advances in genetic and imaging technologies offer promising
far, only one study has looked at a specific subtype of ASD--- new ways of finding biomarkers and/or treatment targets for ASD.
low-functioning children with ASD---and demonstrated that these If successful, these new approaches may one day prove invaluable
can be differentiated from children with a general intellectual in diagnosis, treating and characterizing ASD.
disability using perfusion (Positron Emission Tomography) data.91
This study was an important step forward, as classifier accuracy
was investigated across different autism subtypes.2 CONFLICT OF INTEREST
Thus, although the search for ASD biomarkers is still in its
WS is employed by F. Hoffmann-La Roche. None of the remaining authors have
infancy, the availability of new analytical techniques with high declared any conflict of interest of financial interest, which may arise from being
exploratory power and predictive value offers promising new names as an author on the manuscript.
avenues into finding a biomarker whose complexity equals the
etiology and phenotype of the condition. If successful, such a
biomarker (or a set of biomarkers) might one day prove invaluable
in helping diagnose, treat and characterize ASD. ACKNOWLEDGEMENTS
This work was supported by the AIMS Consortium (Autism Imaging Multicentre
Study) funded by the Medical Research Council UK (G0400061), IMI funded EU AIMS
CONCLUSIONS (http://www.eu-aims.eu), and the NIHR Biomedical Research Centre for Mental Health
at King’s College London, Institute of Psychiatry and South London & Maudsley NHS
Over the last decade, the behavioral diagnosis of ASD has been Foundation Trust.
invaluable in the clinical setting, as it can accommodate all
variations of the ASD spectrum, regardless of their etiology.
However, for any biologist (geneticist or neuroscientist) trying to
solve what is often referred to as the ‘puzzle of neurobiology’ in REFERENCES
ASD, starting with the behavioral phenotype, represents a 1 Abrahams BS, Geschwind DH. Advances in autism genetics: on the threshold of a
heuristic challenge, which can only be compared with solving new neurobiology. Nat Rev Genet 2008; 9: 341--355.
an inverse problem. However, new approaches to genotype 2 Wing L, Gould J, Gillberg C. Autism spectrum disorders in the DSM-V: better or
worse than the DSM-IV? Res Dev Disabil 2011; 32: 768--773.
clustering and the identification of their common functional
3 Muhle R, Trentacoste SV, Rapin I. The genetics of autism. Pediatrics 2004; 113:
pathways may enable us in the future to find comparable (that is, e472--e486.
homogenous) groups of individuals with ASD, which can be 4 Rosenberg RE, Law JK, Yenokyan G, McGready J, Kaufmann WE, Law PA.
combined in order to elucidate their common underlying Characteristics and concordance of autism spectrum disorders among 277 twin
neurobiology. Given the genetic and phenotypic heterogeneity pairs. Arch Pediatr Adolesc Med 2009; 163: 907--914.
of ASD, neither ‘top-down’ clinical and translational studies, nor 5 Wang K, Zhang H, Ma D, Bucan M, Glessner JT, Abrahams BS et al. Common
‘bottom-up’ model system analysis are therefore likely to impact genetic variants on 5p14.1 associate with autism spectrum disorders. Nature 2009;
on ASD alone. Rather, we need to integrate proven technologies 459: 528--533.
around animal models, cellular assays, together with new 6 Weiss LA, Arking DE, , Daly MJ, Chakravarti A. A genome-wide linkage and
association scan reveals novel loci for autism. Nature 2009; 461: 802--808.
analytical approaches in clinical populations to elucidate the
7 Jamain S, Quach H, Betancur C, Råstam M, Colineaux C, Gillberg IC et al. Mutations
complex etiology and phenotype of the condition. Recent animal of the X-linked genes encoding neuroligins NLGN3 and NLGN4 are associated
and postmortem studies are important steps forward in this quest. with autism. Nat Genet 2003; 34: 27--29.
For example, recent work17 demonstrate how causal interactions 8 Scheiffele P, Fan J, Choih J, Fetter R, Serafini T. Neuroligin expressed in
between geno- and phenotype can be established by means of nonneuronal cells triggers presynaptic development in contacting axons. Cell
active genetic manipulation. Furthermore, the discovery of multiple, 2000; 101: 657--669.

Molecular Psychiatry (2013), 435 -- 442 & 2013 Macmillan Publishers Limited
 Translational approaches to Autism
C Ecker et al

441
9 Autism Genome Project Consortium Szatmari P, Paterson AD, Zwaigenbaum L, 36 Hranilovic D, Bujas-Petkovic Z, Vragovic R, Vuk T, Hock K, Jernej B. Hyperser-
Roberts W, Brian J et al. Mapping autism risk loci using genetic linkage and otonemia in adults with autistic disorder. J Autism Dev Disord 2007; 37: 1934--1940.
chromosomal rearrangements. Nat Genet 2007; 39: 319--328. 37 Nabi R, Serajee FJ, Chugani DC, Zhong H, Huq AHMM. Association of tryptophan
10 Dalva MB, McClelland AC, Kayser MS. Cell adhesion molecules: signalling 2,3 dioxygenase gene polymorphism with autism. Am J Med Genet B
functions at the synapse. Nat Rev Neurosci 2007; 8: 206--220. Neuropsychiatr Genet 2004; 125B: 63--68.
11 Varoqueaux F, Aramuni G, Rawson RL, Mohrmann R, Missler M, Gottmann K 38 Devlin B, Cook EH, Coon H, Dawson G, Grigorenko EL, McMahon W et al. Autism
et al. Neuroligins determine synapse maturation and function. Neuron 2006; 51: and the serotonin transporter: the long and short of it. Mol Psychiatry 2005; 10:
741--754. 1110--1116.
12 Rubenstein JLR, Merzenich MM. Model of autism: increased ratio of excitation/ 39 Anderson BM, Schnetz-Boutaud NC, Bartlett J, Wotawa AM, Wright HH, Abramson
inhibition in key neural systems. Genes Brain Behav 2003; 2: 255--267. RK et al. Examination of association of genes in the serotonin system to autism.
13 Abrahams BS, Tentler D, Perederiy JV, Oldham MC, Coppola G, Geschwind DH. Neurogenetics 2009; 10: 209--216.
Genome-wide analyses of human perisylvian cerebral cortical patterning. Proc 40 Chugani DC, Muzik O, Behen M, Rothermel R, Janisse JJ, Lee J et al.
Natl Acad Sci USA 2007; 104: 17849--17854. Developmental changes in brain serotonin synthesis capacity in autistic and
14 Strauss KA, Puffenberger EG, Huentelman MJ, Gottlieb S, Dobrin SE, Parod JM nonautistic children. Ann Neurol 1999; 45: 287--295.
et al. Recessive symptomatic focal epilepsy and mutant contactin-associated 41 Murphy DGM, Daly E, Schmitz N, Toal F, Murphy K, Curran S et al. Cortical
protein-like 2. N Engl J Med 2006; 354: 1370--1377. serotonin 5-HT2A receptor binding and social communication in adults with
15 Casey BJ, Nigg JT, Durston S. New potential leads in the biology and treatment of Asperger’s syndrome: an in vivo SPECT study. Am J Psychiatry 2006; 163: 934--936.
attention deficit-hyperactivity disorder. Curr Opin Neurol 2007; 20: 119--124. 42 Nakamura K, Sekine Y, Ouchi Y, Tsujii M, Yoshikawa E, Futatsubashi M et al. Brain
16 Durand CM, Betancur C, Boeckers TM, Bockmann J, Chaste P, Fauchereau F et al. serotonin and dopamine transporter bindings in adults with high-functioning
Mutations in the gene encoding the synaptic scaffolding protein SHANK3 are autism. Arch Gen Psychiatry 2010; 67: 59--68.
associated with autism spectrum disorders. Nat Genet 2007; 39: 25--27. 43 Daly E, Deeley Q, Ecker C, Craig M, Hallahan B, Murphy C et al. Serotonin and the
17 Pec¸a J, Feliciano C, Ting JT, Wang W, Wells MF, Venkatraman TN et al. Shank3 neural processing of facial emotions in autism; an fMRI study using acute
mutant mice display autistic-like behaviours and striatal dysfunction. Nature 2011; tryptophan depletion. Arch Gen Psychiatry 2012; 4: 1--11.
472: 437--442. 44 Williams K, Wheeler DM, Silove N, Hazell P. Selective serotonin reuptake inhibitors
18 Langen M, Durston S, Staal WG, Palmen SJMC, van Engeland H. Caudate nucleus is for the treatment of autism spectrum disorders. Cochrane database of systematic
enlarged in high-functioning medication-naive subjects with autism. Biol reviews. Advance online publication, 8 September 2010.
Psychiatry 2007; 62: 262--266. 45 Represa A, Ben-Ari Y. Trophic actions of GABA on neuronal development. Trends
19 Page LA, Daly E, Schmitz N, Simmons A, Toal F, Deeley Q et al. In vivo 1H-magnetic Neurosci 2005; 28: 278--283.
resonance spectroscopy study of amygdala-hippocampal and parietal regions in 46 Buxhoeveden DP, Casanova MF. The minicolumn and evolution of the brain. Brain
autism. Am J Psychiatry 2006; 163: 2189--2192. Behav Evol 2002; 60: 125--151.
20 Durand CM, Perroy J, Loll F, Perrais D, Fagni L, Bourgeron T et al. SHANK3 47 Casanova MF, Buxhoeveden DP, Brown C. Clinical and macroscopic correlates of
mutations identified in autism lead to modification of dendritic spine morphology minicolumnar pathology in autism. J Child Neurol 2002; 17: 692--695.
via an actin-dependent mechanism. Mol Psychiatry 2012; 17: 71--84. 48 Casanova MF, Buxhoeveden DP, Switala AE, Roy E. Neuronal density and
21 Herbert MR. SHANK3, the synapse, and autism. N Engl J Med 2011; 365: 173--175. architecture (Gray Level Index) in the brains of autistic patients. J Child Neurol
22 Pizzarelli R, Cherubini E. Alterations of GABAergic signaling in autism spectrum 2002; 17: 515--521.
disorders. Neural Plast 2011; 2011: 297153. 49 Amaral DG, Schumann CM, Nordahl CW. Neuroanatomy of autism. Trends Neurosci
23 Erlander MG, Tillakaratne NJ, Feldblum S, Patel N, Tobin AJ. Two genes encode 2008; 31: 137--145.
distinct glutamate decarboxylases. Neuron 1991; 7: 91--100. 50 Courchesne E, Karns CM, Davis HR, Ziccardi R, Carper RA, Tigue ZD et al. Unusual
24 Trikalinos TA, Karvouni A, Zintzaras E, Ylisaukko-oja T, Peltonen L, Järvelä I et al. A brain growth patterns in early life in patients with autistic disorder: an MRI study.
heterogeneity-based genome search meta-analysis for autism-spectrum disor- Neurology 2001; 57: 245--254.
ders. Mol Psychiatry 2006; 11: 29--36. 51 Lainhart JE, Bigler ED, Bocian M, Coon H, Dinh E, Dawson G et al. Head
25 Geschwind DH. Genetics of autism spectrum disorders. Trends Cogn Sci 2011; 15: circumference and height in autism: a study by the Collaborative Program of
409--416. Excellence in Autism. Am J Med Genet A 2006; 140: 2257--2274.
26 Scott-Van Zeeland AA, Abrahams BS, Alvarez-Retuerto AI, Sonnenblick LI, Rudie 52 Hazlett HC, Poe MD, Gerig G, Styner M, Chappell C, Smith RG et al. Early brain
JD, Ghahremani D et al. Altered functional connectivity in frontal lobe circuits is overgrowth in autism associated with an increase in cortical surface area before
associated with variation in the autism risk gene CNTNAP2. Sci Transl Med 2010; 2: age 2 years. Arch Gen Psychiatry 2011; 68: 467--476.
56ra80. 53 Courchesne E. Abnormal early brain development in autism. Mol Psychiatry 2002;
27 Gilman SR, Iossifov I, Levy D, Ronemus M, Wigler M, Vitkup D. Rare de novo 7(Suppl 2): S21--S23.
variants associated with autism implicate a large functional network of genes 54 Herbert MR, Ziegler DA, Deutsch CK, O’Brien LM, Lange N, Bakardjiev A et al.
involved in formation and function of synapses. Neuron 2011; 70: 898--907. Dissociations of cerebral cortex, subcortical and cerebral white matter volumes in
28 Pinto D, Pagnamenta AT, Klei L, Anney R, Merico D, Regan R et al. Functional autistic boys. Brain 2003; 126: 1182--1192.
impact of global rare copy number variation in autism spectrum disorders. Nature 55 Ecker C, Suckling J, Deoni SC, Lombardo MV, Bullmore ET, Baron-Cohen S et al.
2010; 466: 368--372. Brain anatomy and its relationship to behavior in adults with autism spectrum
29 Murphy DGM, Critchley HD, Schmitz N, McAlonan G, van Amelsvoort T, Robertson disorder: a multicenter magnetic resonance imaging study. Arch Gen Psychiatry
D et al. Asperger syndrome: a proton magnetic resonance spectroscopy study of 2012; 69: 195--209.
brain. Arch Gen Psychiatry 2002; 59: 885--891. 56 Carper RA, Moses P, Tigue ZD, Courchesne E. Cerebral lobes in autism: early
30 DeVito TJ, Drost DJ, Neufeld RWJ, Rajakumar N, Pavlosky W, Williamson P et al. hyperplasia and abnormal age effects. Neuroimage 2002; 16: 1038--1051.
Evidence for cortical dysfunction in autism: a proton magnetic resonance 57 Gogtay N, Thompson PM. Mapping gray matter development: implications for typical
spectroscopic imaging study. Biol Psychiatry 2007; 61: 465--473. development and vulnerability to psychopathology. Brain Cogn 2010; 72: 6--15.
31 Chao H-T, Chen H, Samaco RC, Xue M, Chahrour M, Yoo J et al. Dysfunction in 58 Courchesne E, Yeung-Courchesne R, Press GA, Hesselink JR, Jernigan TL. Hypoplasia
GABA signalling mediates autism-like stereotypies and Rett syndrome pheno- of cerebellar vermal lobules VI and VII in autism. N Engl J Med 1988; 318: 1349--1354.
types. Nature 2010; 468: 263--269. 59 Abell F, Krams M, Ashburner J, Passingham R, Friston K, Frackowiak R et al. The
32 Delahanty RJ, Kang JQ, Brune CW, Kistner EO, Courchesne E, Cox NJ et al. Maternal neuroanatomy of autism: a voxel-based whole brain analysis of structural scans.
transmission of a rare GABRB3 signal peptide variant is associated with autism. Neuroreport 1999; 10: 1647--1651.
Mol Psychiatry 2011; 16: 86--96. 60 McAlonan GM, Cheung V, Cheung C, Suckling J, Lam GY, Tai KS et al. Mapping the
33 Gogolla N, Leblanc JJ, Quast KB, Südhof TC, Fagiolini M, Hensch TK. Common brain in autism. A voxel-based MRI study of volumetric differences and
circuit defect of excitatory-inhibitory balance in mouse models of autism. intercorrelations in autism. Brain 2005; 128: 268--276.
J Neurodev Disord 2009; 1: 172--181. 61 Saitoh O, Karns CM, Courchesne E. Development of the hippocampal formation
34 Hampson DR, Adusei DC, Pacey LKK. The neurochemical basis for the treatment of from 2 to 42 years: MRI evidence of smaller area dentata in autism. Brain 2001;
autism spectrum disorders and Fragile X Syndrome. Biochem Pharmacol 2011; 81: 124: 1317--1324.
1078--1086. 62 Sears LL, Vest C, Mohamed S, Bailey J, Ranson BJ, Piven J. An MRI study of the basal
35 Mendez MA, Horder J, Myers J, Coghlan S, Stokes P, Erritzoe D et al. The brain ganglia in autism. Prog Neuropsychopharmacol Biol Psychiatry 1999; 23: 613--624.
GABA-benzodiazepine receptor alpha-5 subtype in autism spectrum disorder: 63 Redcay E, Courchesne E. Deviant functional magnetic resonance imaging patterns
A pilot [(11)C]Ro15-4513 positron emission tomography study. Neuropharmacol- of brain activity to speech in 2-3-year-old children with autism spectrum disorder.
ogy 2012, PMID: 22546616 (in press). Biol Psychiatry 2008; 64: 589--598.

& 2013 Macmillan Publishers Limited Molecular Psychiatry (2013), 435 -- 442
 Translational approaches to Autism
C Ecker et al

442
64 Rojas DC, Peterson E, Winterrowd E, Reite ML, Rogers SJ, Tregellas JR. Regional 79 Welchew DE, Ashwin C, Berkouk K, Salvador R, Suckling J, Baron-Cohen S et al.
gray matter volumetric changes in autism associated with social and repetitive Functional disconnectivity of the medial temporal lobe in Asperger’s syndrome.
behavior symptoms. BMC Psychiatry 2006; 6: 56. Biol Psychiatry 2005; 57: 991--998.
65 Boddaert N, Chabane N, Gervais H, Good CD, Bourgeois M, Plumet M-H et al. 80 Cox DD, Savoy RL. Functional magnetic resonance imaging (fMRI) ‘brain reading’:
Superior temporal sulcus anatomical abnormalities in childhood autism: a voxel- detecting and classifying distributed patterns of fMRI activity in human visual
based morphometry MRI study. Neuroimage 2004; 23: 364--369. cortex. Neuroimage 2003; 19: 261--270.
66 McAlonan GM, Daly E, Kumari V, Critchley HD, van Amelsvoort T, Suckling J et al. 81 Ecker C, Rocha-Rego V, Johnston P, Mourão-Miranda J, Marquand A, Daly EM et al.
Brain anatomy and sensorimotor gating in Asperger’s syndrome. Brain 2002; 125: Investigating the predictive value of whole-brain structural MR scans in autism: a
1594--1606. pattern classification approach. Neuroimage 2010; 49: 44--56.
67 Vissers ME, Cohen MX, Geurts HM. Brain connectivity and high functioning autism: a 82 Jiao Y, Chen R, Ke X, Chu K, Lu Z, Herskovits EH. Predictive models of autism
promising path of research that needs refined models, methodological conver- spectrum disorder based on brain regional cortical thickness. Neuroimage 2010;
gence, and stronger behavioral links. Neurosci Biobehav Rev 2012; 36: 604--625. 50: 589--599.
68 Courchesne E, Pierce K. Brain overgrowth in autism during a critical time in 83 Uddin LQ, Menon V, Young CB, Ryali S, Chen T, Khouzam A et al. Multivariate
development: implications for frontal pyramidal neuron and interneuron searchlight classification of structural magnetic resonance imaging in children
development and connectivity. Int J Dev Neurosci 2005; 23: 153--170. and adolescents with autism. Biol Psychiatry 2011; 70: 833--841.
69 Rippon G, Brock J, Brown C, Boucher J. Disordered connectivity in the autistic 84 Calderoni S, Retico A, Biagi L, Tancredi R, Muratori F, Tosetti M. Female children
brain: challenges for the ‘new psychophysiology’. Int J Psychophysiol 2007; 63: with autism spectrum disorder: an insight from mass-univariate and pattern
164--172. classification analyses. Neuroimage 2012; 59: 1013--1022.
70 Geschwind DH, Levitt P. Autism spectrum disorders: developmental disconnec- 85 Coutanche MN, Thompson-Schill SL, Schultz RT. Multi-voxel pattern
tion syndromes. Curr Opin Neurobiol 2007; 17: 103--111. analysis of fMRI data predicts clinical symptom severity. Neuroimage 2011; 57:
71 Cheung C, Chua SE, Cheung V, Khong PL, Tai KS, Wong TKW et al. White matter 113--123.
fractional anisotrophy differences and correlates of diagnostic symptoms in 86 Lange N, Dubray MB, Lee JE, Froimowitz MP, Froehlich A, Adluru N et al. Atypical
autism. J Child Psychol Psychiatry 2009; 50: 1102--1112. diffusion tensor hemispheric asymmetry in autism. Autism Res 2010; 3: 350--358.
72 Pugliese L, Catani M, Ameis S, Dell’acqua F, Thiebaut de Schotten M, Murphy C et al. 87 Ingalhalikar M, Parker D, Bloy L, Roberts TPL, Verma R. Diffusion based
The anatomy of extended limbic pathways in Asperger syndrome: a preliminary abnormality markers of pathology: toward learned diagnostic prediction of
diffusion tensor imaging tractography study. Neuroimage 2009; 47: 427--434. ASD. Neuroimage 2011; 57: 918--927.
73 Catani M, Jones DK, Daly E, Embiricos N, Deeley Q, Pugliese L et al. Altered 88 Anderson JS, Nielsen JA, Froehlich AL, Dubray MB, Druzgal TJ, Cariello AN et al.
cerebellar feedback projections in Asperger syndrome. Neuroimage 2008; 41: Functional connectivity magnetic resonance imaging classification of autism.
1184--1191. Brain 2011; 134: 3739--3751.
74 Alexander AL, Lee JE, Lazar M, Boudos R, Dubray MB, Oakes TR et al. Diffusion 89 Jiao Y, Chen R, Ke X, Cheng L, Chu K, Lu Z et al. Predictive models for subtypes of
tensor imaging of the corpus callosum in Autism. Neuroimage 2007; 34: 61--73. autism spectrum disorder based on single-nucleotide polymorphisms and
75 Bloemen OJN, Deeley Q, Sundram F, Daly EM, Barker GJ, Jones DK et al. White magnetic resonance imaging. Adv Med Sci 2011; 56: 334--342.
matter integrity in Asperger syndrome: a preliminary diffusion tensor magnetic 90 Ecker C, Marquand A, Mourão-Miranda J, Johnston P, Daly EM, Brammer MJ et al.
resonance imaging study in adults. Autism Res 2010; 3: 203--213. Describing the brain in autism in five dimensions---magnetic resonance imaging-
76 Belmonte MK, Allen G, Beckel-Mitchener A, Boulanger LM, Carper RA, Webb SJ. assisted diagnosis of autism spectrum disorder using a multiparameter
Autism and abnormal development of brain connectivity. J Neurosci 2004; 24: classification approach. J Neurosci 2010; 30: 10612--10623.
9228--9231. 91 Duchesnay E, Cachia A, Boddaert N, Chabane N, Mangin J-F, Martinot J-L et al.
77 Just MA, Cherkassky VL, Keller TA, Kana RK, Minshew NJ. Functional and Feature selection and classification of imbalanced datasets: application to
anatomical cortical underconnectivity in autism: evidence from an FMRI study of PET images of children with autistic spectrum disorders. Neuroimage 2011; 57:
an executive function task and corpus callosum morphometry. Cereb Cortex 2007; 1003--1014.
17: 951--961.
78 Koshino H, Kana RK, Keller TA, Cherkassky VL, Minshew NJ, Just MA. fMRI This work is licensed under the Creative Commons Attribution-
investigation of working memory for faces in autism: visual coding and NonCommercial-No Derivative Works 3.0 Unported License. To view a
underconnectivity with frontal areas. Cereb Cortex 2008; 18: 289--300. copy of this license, visit http://creativecommons.org/licenses/by-nc-nd/3.0/

Molecular Psychiatry (2013), 435 -- 442 & 2013 Macmillan Publishers Limited