i
y
ga
15
mp
ild
distinctions between ‘hot’ and ‘cool’ executive function
measures. We propose an integrative model that
types to current candidate genes, all of which have minor
Review TRENDS in Cognitive Sciences Vol.10 No.3 March 2006
throughout the lifespan. However, the ADHD literature is
yet to be identified, several genes related to monoaminer-
gic neuromodulation are confirmed as minor contributors
to the overall phenotypic variance [4]. Moreover, there is
robust evidence of structural, functional and neurochemi-
cal brain differences in ADHD, in regions that support
vital cognitive functions [5]. Thus, a coherent and
comprehensive model of the cognitive substrates of
ADHD would be a highly desirable means of linking
genetic, neurobiological and phenotypic levels of analysis,
with the objective of improving diagnostic approaches and
therapeutic options.
Such a model should eventually encompass ADHD
sustained), investigators failed to observe specific diag-
nostic deficits [12]. Based on parallels between ADHD
symptoms and presumed cognitive deficits and those of
patients with frontal lobe disorders, researchers expanded
the scope of inquiry to higher-order cognitive processes
thought to be sub-served by the frontal lobes such as
inhibitory control, attentional regulation and working
memory – constructs grouped under the rubric of
executive function (EF) [13] (see Box 1). In 1997, Barkley
proposed a comprehensive theory of ADHD with deficient
inhibitory control as the core deficit that secondarily
disrupts other EF processes [14]. The explicit testable
prediction that inhibitory deficits and broad EF dysfunc-
The heritability of ADHD is, however, both substantial
and solidly established. Although major risk genes have
‘behavioral condition’ that is diagnosed subjectively [3]. continuous performance task [11]. However, when specific
attentional processes were targeted (i.e. divided, selective,
incorporates new neuroanatomical findings and empha-
sizes the interactions between parallel processing path-
ways as potential loci for dysfunction. Such a
reconceptualization provides a means to transcend the
limits of current models of executive dysfunction in
ADHD and suggests a plan for future research on
cognition grounded in neurophysiological and develop-
mental considerations.
Introduction
Attention-deficit/hyperactivity disorder (ADHD) is
characterized by pervasive behavioral symptoms of
hyperactivity, impulsivity and inattention, beginning in
childhood [1]. Despite its high prevalence and associated
lifelong impairment [2], ADHD remains controversial
because of the use of psychostimulants for treatment of a
Characterizing cogn
beyond executive d
F. Xavier Castellanos1, Edmund J.S. Sonu
and Rosemary Tannock3
1Institute for Pediatric Neuroscience, NYU Child Study Center, 2
2Department of Psychology, University of Southampton, Southa
3Brain and Behavior Research Program, The Hospital for Sick Ch
The hypothesis that Attention-Deficit/Hyperactivity Dis-
order (ADHD) reflects a primary inhibitory executive
function deficit has spurred a substantial literature.
However, empirical findings and methodological issues
challenge the etiologic primacy of inhibitory and
executive deficits in ADHD. Based on accumulating
evidence of increased intra-individual variability in
ADHD, we reconsider executive dysfunction in light of
effects at best [4]. Similarly, active areas of potentially
related research, such as psychopharmacology [7] or
neuroimaging [8,9], which have been recently reviewed
elsewhere, are not covered, but will certainly contribute to
fully integrated models. Our purposes here are to reassess
currently dominant cognitive models of ADHD in light of
empirical and conceptual challenges and to highlight an
integrative model that incorporates emerging neuroana-
tomical perspectives. We believe that this approach will
provide a more robust framework for the translational
multidisciplinary efforts that are already underway in
laboratories throughout the world.
Evolving cognitive models of ADHD
The explicit criteria for ADHD that were first codified in
1980 (see historical review in [10]) emphasized inattention
as much as hyperactivity, based on robust objective
evidence of behavioral inattention and performance
deficits on a laboratory measure of attention, the
tion in ADHD:
sfunction
-Barke1,2, Michael P. Milham1
Lexington Avenue, New York, NY 10016, USA
ton SO17 1BJ, UK
ren, 555 University Avenue, Toronto, Ontario M5G 1X8, Canada
preponderantly based on children of elementary school
age. As neuropsychological differences are most detectible
at this age [6], this brief review will generally exclude
studies on preschool-age-children or adults, despite
remarkable consistency between cognitive deficits across
these wide age ranges. We defer attempts to synthesize
initial efforts at linking cognitive and neuronal pheno-
taking into account measurement error, established the
EF deficit model as the dominant paradigm over the past
Corresponding author: Castellanos, F.X. (castef01@med.nyu.edu).
Available online 7 February 2006
www.sciencedirect.com 1364-6613/$ - see front matter Q 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.tics.2006.01.011
tion should be observable in all children with ADHD,
useful intermediate phenotype (endophenotype) for deli-
Box 1. Executive function: is there a central executive?
There is no consensus definition of ‘executive function’ (EF). The
term is generally used to describe a broad range of ‘top-down’
cognitive processes and abilities that enable flexible, goal-directed
behavior. Examples of such processes include planning and
implementing strategies for performance, initiation and discontinu-
ation of actions, inhibiting habitual or prepotent responses or task-
irrelevant information, performance monitoring, vigilant attention
and set switching.
Researchers have struggled to understand whether the broad
range of ‘executive’ functions are supported by a single unitary
process or a diverse array of cognitive processes. Although some
have argued for the existence of a unitary process based in the
frontal lobes, neuroimaging and focal lesion studies fail to support
this notion. The frontal lobes have proven to be heterogeneous, with
numerous anatomically and functionally distinct subregions, each
associated with a different subset of executive functions [20]. As
such, current models favor a conception of EF as a collection of
higher-order cognitive control processes [21]. (See http://www.
aboutkidshealth.ca/ofhc/news/SREF/4144.asp for further develop-
mental perspectives on EF.)
Review TRENDS in Cognitive Sciences Vol.10 No.3 March 2006118
neating risk genes corresponding to a neuropsychologi-
cally distinct subtype of ADHD.
The usefulness of the Stop task is also apparently
supported by a meta-analysis of 17 studies (nearly 1200
decade and catalyzed a burgeoning literature [15–19],
much of it focused on inhibition as the core deficit
in ADHD.
Testing inhibition as the primary executive deficit in
ADHD
Of several types of inhibitory processes, only executive
motor inhibition has clear replicated evidence in ADHD
[24]. The bulk of this support derives from Go/No-Go tasks
[25] and particularly, the Stop task [26] (see Box 2).
Converging lesion and imaging studies pinpoint the right
inferior prefrontal cortex as a crucial region for effective
Stop task performance [27,28]. Additionally, evidence of
familiality [29–31] suggests that the Stop task could be a
Box 2. Measuring executive inhibition: the Stop Task
The Stop Signal Task examines an individual’s ability to stop a
prepotent motor response and is unique among tasks used to
measure inhibitory control in that it permits an estimation of the
latency of the inhibitory process. It involves two tasks that differ with
respect to frequency, predictability, and stimulus parameters
(modality, intensity). The Go-Task, presented on every trial, is a
speeded forced-choice reaction time task that requires participants
to respond to a stimulus referred to as the Go-Signal. The Stop-Task,
occurring randomly and infrequently (e.g. 25% of Go trials), involves
presentation of a Stop Signal (SS; often an auditory tone) that
countermands the Go response requirements (i.e. participants are to
immediately inhibit the response to the Go-Signal).
The task assumes that ability to inhibit the Go response depends
on the outcome of a race between two independent processes
(response generation/execution, response inhibition) [22]. If the
inhibitory process wins, the planned action is stopped; otherwise,
the Go response continues to completion. Thus, inhibitory control
depends on the Go response time, the within-subject variability of
the Go response time, and the reaction time to the SS (SSRT). SSRT
is the primary performance variable, indicating the speed of the
inhibitory process; faster SSRT’s reflect efficient inhibitory
control [23].
www.sciencedirect.com
children) reporting significant longer Stop Signal reaction
times (SSRT) in ADHD (Cohen’s effect size, dZ0.58) [18].
However, several potential confounds complicate the
interpretation of this difference. First, children with
ADHD also exhibit significantly slower RTs to Go stimuli
(dZ0.52) which may disproportionately influence the
calculation of the SSRT. Second, they demonstrate even
greater Go stimulus RT variability (dZ0.72) which
challenges the Stop task’s assumption of SSRT invariance
and undermines the feasibility of using a tracking
procedure to establish appropriate interval between Stop
and Go signals to produce equal proportions of successful
and failed inhibitions [18]. More generally, the Stop task
imposes subtle but continuous demands on stimulus
anticipation, response preparation, speed of stimulus
processing, and the ability to hold task instructions on-
line [27] and children with ADHD may be impaired in each
of these processes [19,32,33]. The alternative of examining
motor inhibition with the simpler Go/No-Go task may also
be problematic. Electrophysiological studies of Go/No-Go
tasks in ADHD, which allow a more precise dissection of
effects, find nonspecific deficits that are not limited to No-
Go trials [34–37]. Also casting doubt on the centrality of
inhibitory deficits is the largest study of stimulant naı¨ve
boys (nZ75) with severe pervasive ADHD (Hyperkinetic
Disorder) and matched controls (nZ70) which found
strikingly similar rates of No-Go errors and mean Go RT
on a computerized Go/No-Go task [38]. Thus, these
counterfactuals fail to support the strong hypothesis of
deficient inhibition as the primary cognitive deficit
in ADHD.
Is executive dysfunction intrinsic to ADHD?
When meta-analyses are extended to the broader domain
of executive function, a similar pattern of significant albeit
moderate associations emerges. A meta-analysis compris-
ing 83 studies and over 6700 subjects found associations
between ADHD and executive dysfunction across all
domains tested (i.e. planning, vigilance, set shifting, and
verbal and spatial working memory) ranging in effect from
dZ0.4–0.7 [19]. A meta-analysis focused on working
memory examined a somewhat different subset of studies
and detected stronger effects (dZ0.85–1.14) when spatial
working memory manipulation was distinguished from
simple storage [33]. Thus, manipulation of spatial working
memory appears to offer the strongest evidence in ADHD,
but direct comparison studies have not yet been con-
ducted. As in the case of response inhibition, EF effects are
difficult to interpret because most executive function tasks
fail to control for potential confounds ascribable to more
‘primitive’ cognitive or physiologic processes [39]. Two
recent studies highlight the seriousness of this issue by
demonstrating that when non-executive abilities are
accounted for by using appropriate control tasks, little
evidence of executive dysfunction remains [38,40].
The modest pattern of associations between a range of
EF deficits and ADHD is frequently interpreted as
providing evidence for a broader/weaker variant of the
Barkley EF hypothesis. This is the result of a failure to
understand the prediction of pervasive deficits, which has
not been adequately tested until recently. An
Review TRENDS in Cognitive Sciences Vol.10 No.3 March 2006 119
approximation of such a test was performed by pooling
data from three different sites [41]. EF impairment was
defined in terms of performance exceeding the 90th
percentile cutoff (based on the control samples) on five
EF measures. On any individual measure, between 16%
and 51% of children with ADHD were classified as
impaired. When multiple deficits were compared, only
31% of children with ADHD, versus 9% of controls,
displayed pervasive impairment (deficits on 3 or more
measures) [41]. Only 10% of ADHD children showed
deficits across all five domains. By contrast, 21% of
children with ADHD (and 53% of controls) were unim-
paired on all five measures. This combined analysis
recapitulates the conclusion that ‘EF weaknesses are
neither necessary nor sufficient to cause all cases of
ADHD’ ([19], p. 1336). Although not yet examined in meta-
analyses, growing evidence links EF deficits to the
inattention dimension of ADHD, rather than to hyper-
activity/impulsivity [42,43]. Clearly the strong predictions
of the EF deficit hypothesis of ADHD are not supported.
Cognitive heterogeneity of ADHD: multiple pathway
models
The cognitive literature is thus incompatible with the
assumption of pathophysiological homogeneity – of a
single core deficit. Researchers have responded by
building models that accept heterogeneity [41,44,45] – in
which ADHD is regarded as an umbrella construct with
clinical value that subsumes multiple potentially dissoci-
able but overlapping cognitive profiles. So far, too few
studies include measures from different functional
domains (EF, reinforcement/motivational, sensory/percep-
tual and motoric), so heterogeneity is largely inferred from
independent studies each working within a particular
functional domain.
One notable exception, designed by proponents of both
camps, contrasted EF and motivational models using a
Stop task and a Choice Delay task in which children chose
between small immediate and large delayed rewards [46].
This collaborative study produced two crucial findings:
first, choices of the small immediate reward (Delay
Aversion) were uncorrelated with SSRT – suggesting
that inhibitory deficits and Delay Aversion in ADHD
were dissociable processes. Second, performance on either
task was only moderately associated with ADHD but
together correctly classified nearly 90% of children with
ADHD. Similar results were found in preschool children
with ADHD [47] and in a study of the mediating pathways
between hydrocephalus and hyperactivity [48], highlight-
ing that neither EF nor Delay Aversion models are
individually sufficient to account for neuropsychological
findings in ADHD.
Towards an integration of multiple pathway models
‘Hot’ and ‘cool’ executive functions
In considering potential synthetic linkages between EF
and processes related to motivation such as underlie Delay
Aversion, a line of work on the development of EF is
particularly pertinent. Noting the functional differen-
tiations within frontal cortices, Zelazo and Muller
distinguish between more purely cognitive aspects of EF
www.sciencedirect.com
associated with dorsolateral prefrontal cortex (DLPFC)
which are characterized as ‘cool’ as contrasted with
relatively ‘hot’ affective aspects of EF associated with
orbital and medial prefrontal cortex (OMPFC) [49,50]. In
this schema, ‘cool’ EF is elicited by relatively abstract,
decontextualized problems, such as most of the EF tasks
tested so far in ADHD (e.g. [15,16,19,24,33]), including
Stroop, flanker, Go/No-Go, Stop, continuous performance
and working memory tasks, which focus on the ability to
suppress automatic processes or prepotent responses and/
or maintain task instructions or representations in
working memory. ‘Hot’ EF ‘is required for problems that
are characterized by high affective involvement or
demand flexible appraisals of the affective significance of
stimuli’ ([49], p. 455). Based on the inarguable association
of moderately impaired performance on ‘cool’ EF tasks and
ADHD [19], Zelazo and Muller proposed that ADHD
should be considered a disorder of ‘cool’ EF [49]. We believe
this may be premature, reflecting the preponderance of
studies which have used just such a conclusion as a
starting point. Rather, we propose that inattention
symptoms may be associated with deficits in ‘cool’ EF,
whereas hyperactivity/impulsivity symptoms will be
found to reflect ‘hot’ EF deficits. This gives rise to the
possibility that some individuals with ADHD will mani-
fest primarily ‘hot’ EF dysfunction, whereas others will
show mainly ‘cool’ EF deficits and others will have both
types. For example, risky decision making in the Iowa
gambling task – an index of ‘hot’ EF – was associated with
hyperactivity/impulsivity symptoms but not symptoms of
inattention, or ‘cool’ EF measures such as working
memory or IQ [51].
The increasing literature on the impact of reinforce-
ment contingencies on ADHD comprising 22 studies with
nearly 1200 children was recently reviewed [52]. Although
the range of approaches used in these studies is too
disparate to allow quantitative meta-analyses, the
authors highlight clear evidence of delay aversion, and
some support for greater behavioral and lower psycho-
physiological sensitivity to reinforcements in ADHD [52].
Abnormalities or inconsistencies in maintaining instruc-
tional set (‘cool’) or motivational state (‘hot’) may also
account for the substantial variability in responding that
emerges in many cognitive tasks and in behavioral
descriptions of ADHD [39] (see Box 3). In summary,
whereas the literature surrounding executive dysfunction
in ADHD has mostly focused on ‘cool’ EF with modest
success, the presence of impairments in incentive,
motivational and reward-related processing suggests
that both ‘hot’ and ‘cool’ EF deficits should be assessed,
with particular attention to developmental aspects and
symptom subtypes [53–55].
Spiraling cortico-striato-thalamo-cortical circuits
A recent synthesis highlights the neuroanatomical sub-
strates for interactions between motivational and cognitive
processes as potential loci of dysfunction in various forms of
psychopathology, including ADHD [56] (see Figure 1).
Dysfunction of distributed cortico-striato-thalamo-cortical
loops is implicated in all neurophysiological models of
psychopathology. However, models of parallel reciprocal
Review TRENDS in Cognitive Sciences Vol.10 No.3 March 2006120
Box 3. Intra-individual variability and ADHD
Ironically, one of the most consistent manifestations of ADHD is the
high prevalence of ‘moment-to-moment variability and inconsistency
loops have failed to address ‘how information can be
transformed across functional regions to help implement
the learning and adaptability that is necessary in the
development of goal-directed behaviors’ ([50], p. 322). The
delineation of complex non-reciprocal pathways consisting
in performance. Such response variability is the one ubiquitous
finding in ADHD research across a variety of speeded-reaction-time
(RT) tasks, laboratories and cultures’ ([45], p. 624). Such variability can
be quantified by decomposing RT responses into the sum of a
normally distributed random variable, m (Figure Ia), and an indepen-
dent exponentially distributed variable, t (Figure Ib), which accounts
for the positive skew of ‘ex-Gaussian’ RT distributions (Figure Ic).
Children with ADHD who had almost identical m values to those of
controls differed markedly from controls in the skew of the RT
distributions and in the derived values of t (Figure Id), yielding a
remarkable diagnostic efficiency of 96% [60].
The implication that c