Clinical phenotype

We begin by briefly describing the phenotype to be explained. ADHD is an early-emerging behavioral syndrome, usually apparent by age 6 or 7 years, but thought to have developmental precursors much earlier than that. It is characterized by developmentally excessive activity, impulsivity, and disorganized, off-task behaviors. Hallmark behaviors include going off task with speech or activity, losing things, and having trouble staying organized, not seeming to pay attention to what is being said or done, excess activity (in children, running around “as if driven by a motor”), loud, reckless play, and impulsive, even dangerous behavior (e.g., blurting out something that should not be said; darting out into traffic without looking). As a result of these behaviors, children with ADHD are far more likely than nondisordered children to fall behind in school, to be suspended or expelled, to be rejected by peers and have few friends, and to be at risk for accidents and physical injuries (Hinshaw, 2002). The syndrome is more common in boys, at least in childhood, and relatively persistent, with a majority continuing to have some impairment into adolescence and a substantial percentage continuing to do so into adulthood (Mannuzza & Klein, 2000). These later impairing problems can include excess traffic accidents, marital problems, employment difficulties, injuries, as well as antisocial behaviors, alcohol and substance abuse and dependence, and aggression (Barkley, Murphy, & Kwasnik, 1996a, 1996b).

Groups of individuals diagnosed with ADHD are heterogeneous. DSM-IV specifies three subtypes: primarily inattentive, primarily hyperactive–impulsive, and combined (both inattentive and hyperactive–impulsive). Our focus in this paper is primarily on the ADHD-combined type.¹

Despite the problem of heterogeneity, the literature admits of several clear group trends that can guide theorizing and application of neuroscience to the syndrome. Any such account may, in the end, account for a particular subset of the population with this behavioral phenotype. Yet, such accounts need to be developed so that those subgroups can eventually be properly identified and the diagnostic nosology refined in relation to etiological mechanisms.

Neurocognitive phenotype

The behavioral, cognitive, and neuropsychological phenotype of ADHD is relatively well characterized at this point. Thus, we know that, as a group, children with ADHD exhibit a range of context-dependent characteristic responses that cannot be fully explained by comorbid psychopathology (Nigg, Hinshaw, Carte, & Treuting, 1998; Nigg, Stavro, Ettenhofer, Hambrick, Miller, & Henderson, in press): (a) in slow, careful contexts, responses are hasty and inaccurate or inappropriate (clinically defined “impulsivity”); (b) in fast, accurate contexts, responses are slow and inaccurate (the “underaroused” hallmark; Oosterlaan, Logan, & Sergeant, 1998); (c) in fast decision contexts, the ability to interrupt a prepotent or prepared response is impaired (faulty “executive response inhibition”; Barkley, 1997; Logan, 1994; Nigg, 1999, 2001); (d) when complex organization of response is required, executive functioning is impaired, in the form of faulty planning, working memory, and integration (Barkley, 1997; Pennington & Ozonoff, 1996; Willcutt, Doyle, Nigg, Farone, & Pennington, in press); (e) in the presence of cues for reward or incentive, delay of reward is difficult, and immediate rewards are given disproportionate weight over larger but later rewards (Luman et al., 2005); and (f) temporal information processing is affected, with a tendency to overestimate time intervals (Barkley, Murphy, & Bush, 2001; Toplak, Rucklidge, Hetherington, John, & Tannock, 2003), a historical hallmark of impulsivity now being integrated with neuroscience theories (Castellanos & Tannock, 2002).

Several neurocognitive process theories have been developed to try to account both for the symptomatic behavior of individuals with ADHD, and for the preceding profile of cognitive and affective weaknesses, nearly all addressing aspects of self regulation in some form (Barkley, 1997; MacCoon, Wallace, & Newman, 2004; Sagvolden, Johansen, Aase, & Russell, 2005; Sergeant, Oosterlaan, & van der Meere, 1999). On the one hand, then, neuroscience has influenced ADHD theory substantially already (Barkley, 1997; Berger & Posner, 2000; Castellanos & Tannock, 2002; Pennington & Ozonoff, 1996; Sagvolden et al., 2005). On the other hand, theories remain relatively unintegrated from a cognitive and affective neuroscience point of view (Berger & Posner, 2000; Castellanos & Tannock, 2002). Furthermore, the developmental account in most theories is limited with regard to the interdependence of respective neural systems and cognitive abilities. We focus on how cognitive and affective neuroscience can help define the core computational mechanisms that may cut across some of these diverse theoretical accounts.

Neurodevelopment and ADHD

Development of the neural systems that support self-regulation (and are involved in ADHD), is expected to feature at least two key principals. The first is the mutual influence of components of regulatory control on one another during development, including the scaffolding of later abilities onto earlier abilities, assisted by the socialization context. Because of the importance of social interactions in this development, we often choose social interaction examples to illustrate our points later. Second, neural organization is expected to reflect increasing specialization during development in childhood as higher order cognitive abilities take hold (Johnson, 2001). Indeed, neural development of the cortical and subcortical pathways, at issue here, continues as late as early adulthood in the form of continued myelination, pruning, and specialization of circuitry (Benes, 2001; Casey, Tottenham, Liston, & Durston, 2005; Gogtay et al., 2004; Sowell, Thompson, Holmes, Jernigan, & Toga, 1999; Sowell, Thompson, Leonard, Welcome, Kan, & Toga, 2004). These processes mature at different rates in different cortical and subcortical areas, with the regions of prefrontal cortex most important for cognitive control completing synaptic specialization the latest (Bourgeois, Goldman–Rakic, & Rakic, 1994; Huttonlocher & Dabholkar, 1997).

Among the core computational mechanisms at issue here, the ability to predict temporal and contextual structure in the environment is central. This ability develops as early as infancy, as evidenced by infant looking times to events that violate their implicit predictions (Saffran, Aslin, & Newport, 1996). However, the ancillary ability to deliberately alter behavior in response to cues or unexpected events may not emerge until toddlerhood (Rothbart & Bates, 1998), as evidenced by toddler's ability to voluntarily override a cue to calm their own mood. This ability becomes refined with age in the cognitive domain as shown recently by Rueda, Fan, McCandliss, Halparin, Gruber, Lercari, and Posner (2004), who found that by age 7 years, attentional responses could be redirected from competing sources to the relevant one.

With development, these cognitive and neural systems become more differentiated as they are modulated both by experience and by the maturation of top-down cortical projections from prefrontal cortex that help the child alter behavior to achieve a goal when these predictions are violated. This increasing differentiation is observed, for example, when a simple ability to wait a short time for a reward (the toddler or preschooler who manages to wait quietly in anticipation of dessert) develops into the ability to plan and carry out several self-directed steps to realize a goal (as when a 10-year-old child assembles a model airplane or plans an elaborate game of house with her friends).

Early infant and toddler abilities to orient attention, respond to cues, and control their behavior may be framed as precursors to the mature development of (a) reflexive orienting (“where” orienting), guided by a parietal–thalamic circuit; (b) alerting and vigilance (“when” orienting), guided by noradrenergic activation of right-lateralized circuitry including the cerebellar–cortical circuits; (c) reinforcement learning and reward response, guided by frontolimbic circuitry; and (d) effortful regulation and voluntary planning of behavior and cognition, which involves frontostriatal circuitry.²

Of these basic circuits, three appear to be involved relatively more in the case of ADHD. Evidence for the involvement of the posterior reflexive orienting system has generally been negative (see meta-analysis by Huang–Pollock & Nigg, 2003). Thus children with ADHD generally do not have difficulty orienting to “where” a stimulus is expected. We therefore do not discuss that circuit further. However, more evidence suggests they have difficulty efficiently activating voluntary or effortful control systems (Nigg, 2001).

Their difficulty in doing so may emanate from difficulty in altering behavior in light of information contrary to predictions, perhaps in turn due to failure to form accurate predictions of two sorts. First, they may have difficulty adjusting to what is presented when it violates what was expected (or when predictions were ineffective). Second, they may have difficulty adjusting behavior to when (timing) something happened that was unexpected (or ineffectively predicted). Because these fundamental, somewhat mechanistic prediction abilities are presumed to underlie higher order behavioral control, and because that deliberate control is affected in ADHD, we focus most heavily on a detailed account of those fundamental “what” and “when” prediction abilities. If the myriad cognitive deficits associated with ADHD could be traced to these fundamental prediction mechanisms, it would enable a simplification and consolidation of neurobiological theories of the disorder.

However, we note also that the development of these cognitive control systems (experienced as voluntary or deliberate control) emerges in conjunction with reactive (relatively involuntary) affective response systems. ADHD co-occurs at above chance levels with affective disorders such as anxiety disorders and major depression (Pliszka, Carlson, & Swanson, 1999) and may be a precursor to severe unsocialized aggressive behavior and psychopathy, which may be accompanied by failure of normal affective reactions such as empathy. Moreover, the normal development of cognitive control may be thought of as presuming intact affect response systems and vice versa, because of the interdependent nature of these circuits functionally and their cross-talk during development. Therefore, we argue that a complete understanding of ADHD along with its comorbid and related disorders mandates integrating an account of these cognitive control systems with affect-related neural systems involved in detecting the emotional significance of events. We first summarize neural correlates of ADHD and then provide a more detailed account of the circuitry that appears to be involved and its functional significance. We conclude by summarizing potential distinct pathways to the disorder of ADHD.

Neural systems implicated in ADHD

Frontostriatal and frontocerebellar circuitry in relation to cognitive control

The frontostriatal and frontocerebellar loops share a number of characteristics that lead us to discuss them in concert. For example, both the cerebellum and the basal ganglia project to the prefrontal cortex via the thalamus. The primary neurotransmitter in both the basal ganglia and cerebellum is γ-aminobutyric acid (GABA), an inhibitory neurotransmitter. Glutamate, an excitatory neurotransmitter, is found in the prefrontal cortex and thalamus. Dopamine is a critical neuromodulator of both circuits (Braver & Cohen, 2001; Cohen & Servan–Schreiber, 1992; Montague, Dayan, & Sejnowski, 1996) that is expressed preferentially in portions of the prefrontal cortex, basal ganglia, and in the neocerebellum including the dentate nucleus and vermis, which are regions implicated in ADHD, namely lobules 9–10 (Hurley, Mash, & Jenner, 2003). Both circuits also receive noradrenergic projections (Berridge, 2001), which again are notable in cerebellar vermis (Powers, O'Connor, & Price, 1989) and both show changes in activation levels in response to methylphenidate administration (Anderson, Polcari, Lowen, Renshaw, & Teicher, 2002; Taylor & Jentsch, 2001).

These two circuits differ, however, in their specific functional computations. The detection of unpredicted rewarding or novel events (what) has been linked to frontostriatal functioning (Berns, Cohen, & Mintun, 1997; Schultz, Dayan, & Montague, 1997), whereas alerting, monitoring and detecting violations in the timing of events (when) has been linked to frontocerebellar function (Ivry, Spencer, Zelaznik, & Diedrichsen, 2002; Spencer, Zelaznik, Diedrichsen, & Ivry, 2003).

Casey (in press; Casey, Amso, & Davidson, in press) proposed that striatal and cerebellar neural systems may have discrete roles in learning from environmental inputs based solely on event frequencies and timing, even as they work together in monitoring the environment and altering behavior in the appropriate context. We propose that these brain regions are able to alter behavior via afferent and efferent connections within prefrontal cortex, which are fine-tuned with experience over the course of development in childhood. This experientially dependent fine-tuning would proceed more slowly if these systems started out being inefficient. In ADHD, we suggest that there is diminished ability to alter behavior when predictions about the environment are violated, that can be due to a diminished ability to (a) make accurate predictions (subcortical) or (b) adequately override a predicted action with the appropriate alternative one when predicted events do not match with the actual ones (cortical). This type of deficit could emerge from altered catecholinergic modulation of the circuitry, altered prefrontal projections, or poor prediction-related functions presumably related to altered development of striatal and/or cerebellar neuronal regions or connections.

The frontostriatal (i.e., basal ganglia thalamocortical) circuit is well described (Figure 1), and at least five different loops are identified (Alexander, Crutcher, & DeLong, 1991; Middleton & Strick, 2002). These circuits include oculomotor, motor (bypassed herein, see footnote 2), prefrontal (dorsolateral and lateral orbital, which we emphasize here), and limbic circuits (which we describe in the next section). Each parallel circuit involves the same projection regions (basal ganglia, thalamus, and cortex), but they differ in the exact projection zone within each region and in the set of thoughts and actions they support.

The basal ganglia and cerebellar thalamocortical loops.

The general organization of these circuits involves prefrontal projections to different areas of the striatum (i.e., putamen or caudate nuclei, depending on whether in motor or cognitive loops, respectively), and then projections to a direct or indirect pathway. The direct pathway involves an inhibitory projection to the internal capsule of the globus pallidus and substantia nigra, resulting in the dampening of an inhibitory projection to the thalamus (i.e., disinhibition). The behavioral result is that activation in this pathway is thought to facilitate top-down goal-oriented behavior and thereby simultaneously suppress alternative nongoal-oriented behaviors. The indirect pathway consists of an inhibitory GABA projection to the external capsule of the globus pallidus that dampens the inhibitory projection to the subthalamic nuclei, resulting in excitation of the internal capsule of the globus pallidus and substantia nigra and inhibition of the thalamus. The behavioral result is that activation in this pathway leads to interruption of on-going behavior, enabling the individual to orient to new or unexpected information and thus setting the stage for appropriate adjustment of behavior to the new information. That pathway may also be important for the interaction of affective and cognitive control systems, as described subsequently.

Because the prefrontal cortex projects directly to the basal ganglia and cerebellum and both project back to the prefrontal cortex via the thalamus, an account of suppression of competing actions may be described here in somewhat mechanistic terms. That is, the basal ganglia and cerebellum are assumed to monitor the relative frequency and/or timing of events, respectively (Hayes, Davidson, Keele, & Rafal, 1998; McClure, Berns, & Montague, 2003; Spencer et al., 2003). This allows the system to predict what and when an event will occur, essential components of cognitive control, a process presumably disrupted during development in individuals with ADHD. Top-down prefrontal regions are assumed to “kick-in” when these predictions are violated (as signaled by the subcortical circuits), so are critical in enabling effective adjustment of behavior when a situation changes unexpectedly. Thus, frontocerebellar and frontostriatal circuits may provide neural mechanisms for learning about structure in the environment and detecting violations in these predictions, which are fed to the prefrontal cortex for integration with goals represented in working memory. The result is facilitation of top-down control of behavior as well as altering of behavior in response to relevant changes in context.

In the case of the child with ADHD, the system may break down in a couple of ways. First, it may break down in failure to identify an expected event (e.g., if the child has not learned that the bell at the beginning of school is a signal to settle down and begin working within 1 or 2 min, instead judging that there are 5 or 10 min of cushion, then she may continue clowning around, which will now be disruptive). Second, without having accurately predicted (learned) aspects of this temporal association, the ability to detect a violation in expected outcome is hindered (others are no longer joining in the clowning around, but the significance of this is missed). Third, even after the violated prediction is identified, the deliberate control mechanisms that should be recruited to support adjusting of behavior may be slow to come into play. What is important here is that these simple learning and prediction mechanisms, when compromised, could result in poor learning of predicted events and thus poor detection of unexpected events that should signal the child to alter her behavior. This in turn could drive continued maladaptive behavior, even when a child “knows” the correct behavior in a hypothetical situation.

In this view, then, the basal ganglia and cerebellum learn to predict and detect violations of prediction in the (a) timing (cerebellum) and (b) nature or frequency of events (basal ganglia), providing the system with a way to shift out a behavioral set to a more adaptive prefrontally guided behavior when salient, or unexpected, events occur. Otherwise, when a goal-oriented behavior program is running, the basal ganglia and cerebellum act broadly to inhibit or interrupt competing thoughts and behaviors that would otherwise interfere with the prefrontally driven goal or behavior (Casey, Durston, & Fossella, 2001; Mink, 1996). Thus, this model of cognitive control (Casey, 2000, in press) suggests that the role of the basal ganglia and cerebellum is, in part, supported by inhibitory (GABAergic) projections that could serve to interrupt the ongoing behavior when expectations are violated, thus enabling the switching or shifting of attention elsewhere to accommodate that important new information. The prefrontal cortex, which consists primarily of excitatory (glutamergic) projections, is thus involved in guiding and maintaining relevant information for action.

In this view, disruption of the prefrontal cortex results in weakened ability to carry out actions that must be represented in working memory, that require monitoring of new information, and that require holding multiple goals in mind. This model is consistent with our hypothesis of the way in which the prefrontal cortex, basal ganglia, and cerebellum are involved in the suppression of inappropriate thoughts and behaviors. This view of cognitive control also is consistent with some others that specify the role of the basal ganglia in interrupting behavior (Kimberg & Farrah, 1998), but it contrasts with still other cognitive theories that emphasize the role of ventral and orbital prefrontal cortex in inhibitory functions (Diamond, 1990; Iversen & Mishkin, 1970).

In turn, if the basal ganglia are injured, we would expect the child to either have difficulty in (a) suppressing competing behaviors during goal-directed activity, and/or (b) accurately predicting the likely results of their actions in a real situation (as opposed to a hypothetical situation) and thus in choosing an appropriate action in response to these cues from peers, teachers, and parents. Finally, if the cerebellum is affected, we would expect difficulty in (c) accurately predicting when behaviors should change. For example, the child in the classroom may continue to clown around telling jokes even though the bell has rung to indicate that it is time to settle down and begin school work, because she has misjudged the implicit temporal window allowed by the teacher for children to settle immediately after the bell.³

To summarize and recapitulate: how might breakdowns in frontostriatal and/or frontocerebellar circuitry contribute to the symptoms and behaviors observed in a disorder such as ADHD? Assuming that frontostriatal and frontocerebellar pathways serve to recruit top-down control when detecting changes in the environment, then their disruption may result in inefficient prefrontal recruitment in shifting and maintaining a behavioral set in different contexts. This might appear as poor sustained attention to multistep or complex tasks (because the expected changes in the context as the task unfolds would not be computed efficiently), slow response in rapid decision contexts (because the system is inefficient at computing new information), difficulty efficiently shifting from one response set to another, and inefficient response to changing reward or learning contexts, all of which are consistent with the behavioral profile of ADHD. More speculatively, early disruption of these mechanisms may lead to inefficiencies in learning and socialization during development, hampering the normal consolidation of self-regulation. For example, to benefit from parental attunement, reflection, explanation, or correction, the young child has to accurately link the response to her own behavior. This linkage then enables the child to further strengthen their learning and self-regulation abilities. Inefficiency in this link may lead to failure to benefit from the “expectable” amount of parental correction, guidance, or discipline.⁴

The proposed model implies an important mutual-influence process in which intact cerebellar and basal ganglia activation depends on “expectable” prefrontal cortical responsivity, and vice versa. Thus, breakdowns in a prefrontal cortex system could lead to maladaptation of behaviors that depend on the exquisite recognition and timing of prediction violations in the environment, the latter depending on cerebellar/basal ganglia operations. In this instance, we might expect overactivity of these subcortical structures in a maladaptive response. Conversely, atypical development of cerebellar and/or basal ganglia neural pathways could disrupt adaptive operation of prefrontally, goal-directed behavior (i.e., cognitive control).

Frontoamygdala circuitry in affect regulation

Although affect-related frontoamygdala circuits have received comparatively less attention to date in biological theories of ADHD, affect regulation, motivation, and reactive response are quite important to the disorder. For example, there is increasing recognition that in addition to their difficulties with impulsive behavior, children with ADHD (and other impulse-control disorders) also have difficulty with anger, mood, and affect regulation (Pliszka, Carlson, & Swanson, 1999), an observation that is completely congruent with the integrated neural circuitry we are describing. Indeed, emotional context affects the recruitment of cognitive control, and a substantial literature investigates problems with approach or avoidance related processes in ADHD (Luman, Oosterlaan, & Sergeant, 2005). Therefore, an account of affective neural circuitry is essential to a complete portrayal of the neuroscience of ADHD and its clinical and comorbid features.

At the most basic level, positive emotional valence (or expectation of reward) is related to approach, whereas negative valence (or expectation of nonreward) is associated with avoidance and withdrawal (Davidson, Ekman, Saron, Senulis, & Friesen, 1990; Hare, Tottenham, Davidson, Glover, & Casey, 2005; O'Doherty, Winston, Critchley, Perrett, Burt, & Dolan, 2003). Thus, interactions between cognitive and emotional processes play an important role in regulation of behavior in different incentive contexts. Although the neural basis of these interactions is not yet understood, reciprocal neural connections exist between prefrontal control regions and subcortical regions involved in affective response (Amaral, 1986; Groenewegen, Berendse, & Haber, 1993; Ghashghaei & Barbas, 2002; Haber, Groenewegen, Grove, & Nauta, 1985; Haber, Kunishio, Mizobuchi, & Lynd–Balta, 1995). Animal studies have shown the importance of the ventral striatum in appetitive (approach) conditioning (Garcia, Vouimba, Baudry, & Thompson, 1999; LeDoux, 1995; Parkinson, Cardinal, & Everitt, 2000; Parkinson, Dalley, Cardinal, Bamford, Fehnert, Lachenal, Rudarakanchana, Halkerston, Robbins, & Everitt, 2002) and of the amygdala in aversive (avoidance) conditioning. The ventral amygdala in particular has been implicated in fear conditioning and evaluating the emotional significance of a stimulus (LeDoux, 1998).

Human neuroimaging studies of affective control parallel the animal findings and show a role for top-down prefrontal modulation of subcortical regions (Hariri, Bookheimer, & Mazziotta, 2000; Hariri, Mattay, Tessitore, Fera, & Weinberger, 2003; Keightley, Winocur, Graham, Mayberg, Hevenor, & Grady, 2003; Nomura, Ohira, Haneda, Iidaka, Sadato, Okada, & Yonekura, 2004; Ochsner, Ray, Cooper, Robertson, Chopra, Gabrieli, & Gross, 2004). These studies show greater activation in the ventral amygdala when approaching negative information (i.e., fearful faces; Hare et al., 2005; Thomas, Hunt, Vizueta, Sommer, Durston, Yang, & Worden, 2004). The greater the amygdala activity, the slower the reaction times in detecting (approaching) negative information (e.g., fearful faces; Hare et al., 2005). Positive emotional stimuli influence behavior and neural activity differently (Hare et al., 2005). Compared to fearful expressions, happy expressions elicit greater activity in the nucleus accumbens, a region associated with reward and appetitive behavior (Nestler, 2004; Schultz, Tremblay, & Hollerman, 2000). Therefore, activation of the amygdala is associated with avoidance of potentially unpleasant events, while activation of the nucleus accumbens is related to approaching a potentially positive event. Previous imaging studies have shown an inverse correlation between prefrontal cortical activation and amygdala activity, which can be interpreted as reflecting top-down modulation of emotional processes (Drevets & Raichle, 1998). More recently, Hare and colleagues (2005) reported an inverse correlation between ventral striatum and amygdala activation, such that the activity of one inhibits or suppresses the other (Hare & Casey, 2005).

We now extend an affective component to our model of prefrontal modulation of subcortical systems in behavioral regulation. Signals to the prefrontal cortex from the ventral striatum enhance the tendency to approach positive (reward related or reinforcing) stimuli. Signaling from the ventral amygdala, in contrast, would invoke avoidance behavior (cue for nonreward cues). These signals then must be integrated with the existing behavior (goal directed or otherwise) of the child. Thus, when emotional information is irrelevant or contrary to the current goal, top-down connections from the prefrontal cortex to subcortical regions are critical for regulating the effects of affective information on behavior. Their ability to do so obviously depends both on the integrity of those top-down systems and on the intensity or strength of the affective response.

It is straightforward to imagine that impulsivity could result from unregulated approach to almost any potential reward, as well as from failure to avoid potential punishment. Historically, theories of impulsivity have often pointed to such possibilities, especially in conjunction with impaired frontal cortical function (Gray, 1971). For example, excess reward response easily translates, conceptually, into ill-conceived impulsive acts. If a child is working on his homework in class, but sees his friend leave his seat to sharpen a pencil, he may see the potential immediate reward in leaving his own seat to go and interact with his friend. If prefrontal modulation is weak, he may fail to hold in mind the goal of finishing his test preparation so that he can pass the test and earn the parental reward of spending the entire day with his friend this weekend. Impulsive behavior in defeat of the long-range goal results, as he leaves his seat and ultimately loses his weekend privilege.

Thus, it is rather straightforward to begin to imagine how these affective systems could contribute to ADHD, at least to its impulsive features. Others have developed this idea before, and the role of excessive approach responding in particular (i.e., overvaluing of immediate reward cues) is widely discussed in the ADHD literature (Douglas, 1989; Haenlein & Caul, 1987; Sagvolden et al., 1998, 2005). Clearly, a tendency to overweight immediate rewards could lead to failure to consider long-term consequences sufficiently, emanating in impulsive behaviors (the child leaves his seat, gets held after school, is unable to see his friend that weekend).

Likewise, failure of avoidance learning can, in principal, result in continuing to engage in behavior even when it is about to lead to negative consequences, which could have been predicted based on past event frequencies and timing, but this time because the negative valence is insufficiently weighted (i.e., the child fails to experience adequate anxiety when the teacher gives a warning, or is unaffected emotionally by the punishment or expulsion, so the behavior continues). Many impulsive and socially ineffective behaviors of individuals with ADHD could conceivably fit in one of these categories.

However, the empirical data in support of these “single vector” affective explanations of impulsivity in ADHD are rather weak. We comment briefly on each of these points. First, how likely is it that ADHD can be understood in light of an excess approach response or inadequate modulation of approach tendencies? One source of support for this view would come from studies that suggest that temperament or personality traits associated with approach (e.g., Extraversion or positive affectivity) are associated with ADHD. Studies on this question are few, and results are admittedly mixed (Nigg, John, Blaskey, Huang–Pollock, Willcutt, Hinshaw, & Pennington, 2002). Nigg, Blaskey, Huang–Pollock, and John (2002) found that ADHD was associated in most cases with low scores on the personality trait of constraint (associated conceptually with the frontostriatal circuitry we described earlier), but that only a small subgroup was associated with very high extraversion scores.

Efforts to assess reaction time and accuracy responses in relation to cues for reward in ADHD have yielded mixed findings as well (Iaboni, Douglas, & Baker, 1995), perhaps in part due to failure to control for conduct disorder and in part due to varying ages of samples. Physiological measures (e.g., heart rate response to cues for reward) have also shown an inconsistent picture (Iaboni, Douglas, & Ditto, 1997; Pliszka, Hatch, Borcherding, & Rogeness, 1993), although some support can be found. On balance, despite some supportive evidence, it seems likely that excess approach tendencies cannot by themselves account for ADHD (for a review, see Luman et al., 2005).

Second, how readily can ADHD be viewed as insufficient avoidance response (Quay, 1988)? Most performance-based research has failed to support the idea that children with ADHD are insensitive to cues for nonreward (Iaboni et al., 1995; for a review, see Nigg, 2001), although some physiological studies appear supportive. In contrast, an impaired avoidance response may be a feature of impulsive behavior in some cases of unsocialized conduct disorder and/or psychopathy (Newman & Wallace, 1993; Raine, 2002). This, in turn, would implicate a subset of children with ADHD with low anxiety and high aggression, who go on to develop these more serious problems characterized by primary disturbance of this affective response systems.

In short, it appears that support is incomplete at best for these two “single vector” attempts to explain ADHD via a breakdown in affective response neural systems. A more promising approach may be the recognition that ADHD can arise from discrete routes, but given the “crosstalk” among the neural circuits (Haber, 2003), may reflect the joint operation of the systems. The goal-directed behaviors that we have described in this paper and that appear disrupted in ADHD require continual processing of ongoing and complex chains of events. This updating of goal states in the context of the frequency, timing, and emotional valence of information can be reflected in the neural circuitry. Information can thus be channeled from limbic to cognitive and to motor circuits at the subcortical level, allowing for appropriate responses to environmental cues. Parallel circuits must work together, so that the coordinated behaviors are maintained, but also can be modified and changed according to the particular context and whether it was expected or not. The inability to protect specific actions related to later goals, or to adapt effectively to new environment cues, are core features of ADHD that can be reflected in the previously described neural circuitry that supports aspects of motor control, cognition, and motivation.

Developmental pathways to ADHD

Collectively, this review of key clinical symptoms and the implicated neuroanatomical circuits indicates that prefrontal projections are a critical aspect of the basic elements of cognitive and affective control that are involved in ADHD. Developmental changes in prefrontal circuitry contribute to the emergence of mature cognitive control as well as well-regulated affective response. Therefore, development must be considered in delineating the contributions of different brain regions (e.g., prefrontal cortex, striatum, cerebellum, amygdala) to cognitive and affective control as relevant to impulsive or dysregulatory psychopathology generally and ADHD in particular. Given the protracted maturation of prefrontal cortex relative to subcortical regions, one would expect to see a shift from greater activity in subcortical regions like the amygdala and nucleus accumbens in childhood to increased prefrontally mediated processing in adults. A developmental delay view of ADHD may implicate such further protracted development of these systems.

An integrated cognitive and affective neuroscience understanding of ADHD would ideally consider components of cognitive control, affect regulation, and their mutual influence on one another in behavioral regulation and in development. An important principal that we noted earlier is that these systems support one another (or fail to do so) during early development. Because temperament can be thought of as implicating affective regulation, early temperamental predispositions may represent precursors to difficulty with cognitive control or affective (reactive) intensity (Rothbart & Bates, 1998). For example, very early in life infants can be reliably coded as tending to strongly approach novel stimuli (associated with weak physiological reactivity to those situations), versus tending to strongly withdraw (associated with strong physiological reactivity); the same features can be reliably coded in toddlers and preschoolers placed in a room with unfamiliar children (Kagan, 2003; Kagan, Reznick, & Snidman, 1987).

Similarly, toddlers can be reliably assessed in relation to their temperamental ability to deliberately redirect attention (Rothbart & Bates, 1998), a possible precursor to top-down cognitive control that may be indicative of successful prediction mechanisms. By contrast, these predispositions likely influence one another and are influenced by socialization during the toddler and preschool years, culminating in relatively stable patterns of behavior by the early elementary school years that enable characterization of behavioral syndromes such as ADHD.

Different types of temperamental tendencies may lead to ADHD, and ADHD itself then may represent the common final outcome of a discrete handful of developmental pathways, when viewed from the perspective of cognitive and affective neuroscience. Drawing upon principals of developmental psychopathology, this situation could exemplify equifinality (Cicchetti & Rogosch, 1996) in that somewhat distinct breakdowns in neural processing systems lead to a common behavioral phenotype. We summarize this state of affairs in Figure 2a. It illustrates that several types of breakdown in the prediction and learning circuits that underlie cognitive control might nonetheless result in similar surface symptoms at the level of ratings or observations of children's molar behavior.

A schematic illustration of the pathways to ADHD symptoms and syndrome: (a) equifinality in ADHD and neuroscience-based hypothesized computational mechanism breakdowns and (b) multiple pathways to related but partially distinct etiological phenotypes in ADHD.

Alternatively, and what we think is more likely, is that in addition to some equifinality, ADHD also exemplifies multiple pathways with somewhat different phenotypic outcomes that are still in the process of being validated as distinct etiological subgroups. We summarize these various pathways in Figure 2b. As it suggests, one route to ADHD might be via (a) breakdown in fundamental prediction mechanisms as we described earlier, leading secondarily to problems in cognitive control, “executive functioning,” with secondary problems in arousal regulation and affect regulation due to weakness in the “top-down” regulatory system. If this route was accompanied by relatively intact affect response systems, then the clinical picture might be relatively benign: it might yield ADHD characterized by ineffective, disorganized behavior, but without the severe comorbid picture that emerges when affective systems also break down.

Another route might be via primary dysfunction in the affective (b) approach or (c) avoidance systems. These routes may lead to somewhat different clinical phenotypes. For example, as noted, we suspect that faulty avoidance learning (characterized by lack of skin conductance responses to cues for nonreward or punishment), relatively well characterized in severe psychopathy or unsocialized conduct disorder, is a feature of impulsive/antisocial behavior (bottom route, Figure 2b). This route represents another subgroup of children with impulsive behavior problems, most of whom may initially be viewed clinically as having ADHD. However, they will go on to aggressive/or antisocial behaviors in many instances (for simplicity, we are omitting here discussion of the extensively described socialization experiences that moderate, mediate, and maintain that developmental pathway; see Snyder, Reid, & Patterson, 2003). The result would be that this pathways often yields a relatively severely disturbed group of youngsters with problems in impulsivity, aggression, social adaptation, and affective response, who are at risk for severe antisocial outcomes or psychopathy (Pliszka et al., 1999; Simonoff, Elander, Holmshaw, Pickles, Murray, & Rutter, 2004).

Still another possibility is that ADHD could emerge simply via extreme approach tendencies (i.e., extreme temperamental approach, probably related neurobiologically to reactivity of nucleus accumbens and associated circuitry as described earlier). This route might correspond to problems in reinforcement learning as described by Sagvolden et al. (2005). This pathway is illustrated in the middle route in Figure 2b. This group may have minimal cognitive problems and may not exhibit difficulties in “executive functioning,” cognitive control, or related measures. Nigg, Willcutt, Doyle, and Sonuga–Barke (2005) provide a discussion of the possibility that only some children with ADHD are characterized by neuropsychological executive function problems. Thus, this route as well may, in many instances, be relatively benign in its clinical course despite leading to diagnosis and to impairment at least early in development. Early emerging temperament traits thus may indicate precursors to eventual development of ADHD and map onto particular neural systems that may be involved in the disorder.

However, a further and more nuanced possibility bears consideration. That is that different components of the ADHD phenotype may be related to breakdowns in different neural circuits, but which then tend to co-occur in their breakdown more often than chance, leading to the observed syndrome of ADHD. One model (e.g., Sonuga–Barke, 2002) is that the inattention disorganization symptoms of ADHD are related to breakdown in cognitive control (response inhibition, or executive functions), whereas the impulsive symptoms of ADHD are related to breakdowns in affective responding (primarily in a poorly regulated approach system). This state of affairs can be visualized by considering the first two pathways in Figure 2b, not as distinct syndrome outcomes, but as dimensional behavioral domains.

Empirical tests of this possibility have been few. However, Nigg et al. (in press) created a latent “executive” or “control” variable using several clinical executive function tasks (including the stopping task, the Trailmaking Test, the Tower of London, and the Wisconsin Card Sort). They then regressed this score on latent variables for symptoms of inattention–disorganization (pooled from multiple informants) and impulsivity–hyperactivity. Only inattention–disorganization was uniquely related to the executive score, roughly congruent with our broad domain of cognitive control and consistent with our two-pathway model (Figure 2b).

Consistent with that finding, but approaching the question from a personality/temperament perspective, Nigg, John, et al. (2002) found that symptoms of inattention–disorganization were related to low trait Constraint after partialing hyperactivity–implusivity from scores for inattention–disorganization. Conversely, when inattention was partialed from impulsivity, the latter symptoms were related to high trait hostility but not to Constraint. This finding also is broadly consistent with the idea that the hyperactivity–impulsivity dimension may be related to breakdown in affective response systems, whereas the inattention–disorganization behavioral dimension may be related to breakdowns in cognitive control (including “what” and “when” prediction systems). Because these neural systems are needed in support of one another in day to day behavioral regulation, these problems tend to co-occur, and the integrated syndrome of ADHD is the result. However, experimentally it may be possible to partially dissociate these effects as illustrated in the cited studies.

Key developmental issues in ADHD neuroscience research

Self-regulation develops through mutual-influence processes both at the level of interdependent neural systems, and at the level of socialization (learning experiences) interacting with neural development. Social interactions help the developing nervous system consolidate into adaptive and functional regulatory capability that is increasingly at the command of the child as she develops. At the same time, neural development and maturation continue all the way into late adolescence in the neural structures involved in ADHD. At the neural level of analysis, evidence is converging on frontal–striatal/frontal–cerebellar circuitry as involved in ADHD.

These systems are involved in myriad self-regulatory behaviors and functions that emerge and consolidate throughout childhood and into adolescence. They include some of the latest maturing structures in the brain, and they include networks heavily dependent on catecholamines as well as GABAergic systems. As a result, several related but distinct cognitive process models of ADHD have been proposed, addressing different elements of the integrated self-regulatory system (effortful control, response inhibition, arousal regulation, activation, reinforcement response, temporal information processing). These different theories indicate that ADHD theory has already begun to be influenced by cognitive and affective neuroscience.

At the same time, these theories may be able to be simplified by considering that these two main circuits (frontal–striatal and frontal–cerebellar) share an integrated role in “cognitive control,” which requires expectation matching as to when and what is going to occur in response to a behavior. We have noted the evidence that children with ADHD have normally developed spatial orienting (“where;” Huang–Pollock & Nigg, 2003), but poor alerting (“when”; see Brandeis, van Leeuwen, Rubia, Vitacco, Steger, Pascual–Marqui, & Steinhausen, 1998; Nigg, Swanson, & Hinshaw, 1997) and poor ability to adjust to new information in rapid-response contexts (Nigg, 1999; Schachar, Tannock, & Logan, 1993), consistent with involvement of these core functional operations. Although this evidence is thus beginning to converge in ways that are quite promising, we close by noting two major issues that confront these theories and the scientific field going forward in neuroscience research on ADHD.

One issue is the need for continued elaboration of developmental dynamics at two key developmental periods. The first period is the rapid increasing specialization of neural networks during early to middle childhood, when ADHD is usually emerging and diagnosed. Failure of the normal process of specialization (perhaps via experience-dependent arborization and pruning in prefrontal–basal ganglia and cerebellar–cortical loops) may be related to the difficulty in consolidating cognitive control that we think is contributing to at least one major pathway ADHD. This is also a period when cognitive control is rapidly emerging and consolidating. For example, effortful control in the form of strategic attentional and impulse control continues to develop throughout childhood, yielding steady progress in the ability to ignore competing stimulus-driven responses over the second to fourth grade years (Huang–Pollock, Carr, & Nigg, 2002) and the ability to inhibit primary responses over the age period from about ages 5 to 7 and beyond (Carver, Livesey, & Charles, 2001). Likewise, Rueda et al. (2004) found that between ages 6 and 9 years the attentional selection network showed development to about age 7, whereas the alerting response continued to mature at least through age 10.

ADHD probably emerges in relation to differences or breakdowns in this developmental progression during the early childhood years. We are suggesting here that this breakdown may be related to breakdown in the basic mechanism of detecting what and when violations of expectations occur. A potential approach for examining these mechanisms in parallel is similar to an attention network task described by Rueda et al. (2004), which measures alerting, orienting, and conflict resolution within the same task. We developed a visual detection task (Davidson, Horvitz, Tottenham, Durston, Fossella, & Casey, 2003) wherein the predictability of events is manipulated in two ways. Either an expected or unexpected stimulus is presented at either an expected or unexpected time. On the majority of trials an expected event occurs at the expected time for predictions to occur, but for the conditions of interest either an unexpected event occurred or an event occurred at an unexpected time. Behavioral results using this paradigm showed increased reaction times and reduced accuracy for unexpected events, consistent with the notion that control processes may be needed to perform these trials. Developmental data from 36 subjects show that children perform worse than adolescents and adults. Imaging data using this paradigm suggested that the caudate nucleus appears to be more sensitive to stimulus violations (what), whereas the cerebellum is more sensitive to temporal violations (when) during performance of this task. These results provide further evidence for a functional distinction between frontostriatal and frontocerebellar circuitry in the human brain and suggest that this paradigm may serve as an exemplar task for use with children with ADHD to test our theory that basic stimulus and temporal prediction mechanisms are affected in ADHD.

To test the potential relevance of detecting the emotional significance of predicted events, the stimuli used in the paradigm could be emotional in nature (e.g., happy and fearful faces). This additional component to the task could provide important information on how competing limbic (amygdala) and reward neural systems (nucleus accumbens) implicated in approach and avoidance behavior play a role in the clinical phenotype of this disorder.

Another prediction that flows from our argument is that ADHD should be associated with failures in implicit learning (this has been examined in only a limited number of studies, usually involving patients with tic disorders or obsessive–compulsive disorder; Channon, Pratt, & Robertson, 2003). Several paradigms are available for such investigation. In all, we are suggesting that modified experimental paradigms may be useful in elucidating particular mechanistic breakdowns in ADHD, but that these must be integrated with a developmental account of how these mechanisms emerge and influence one another during childhood.

Following that last point, a second key period for developmental clarification is the adolescent transition, when relevant neural networks are consolidating and maturing. The ability to rapidly suppress unintended or prepotent responses continues to develop through adolescence (Bedard, Nichols, Barbosa, Schachar, Logan, & Tannock, 2002), presumably aided by ongoing myelination and pruning of frontal cortical neural networks (Benes, 2001) in conjunction with socialization and learning experiences. Symptomatically, in cases of ADHD, hyperactivity tends to improve, whereas impulsive and disorganized behaviors tend to become salient problems for children with ADHD in this developmental period. In addition, gender ratios appear to shift, with girls gaining in relative frequency of diagnosis. Hormonal effects may be important here as well.

Progress on these fronts will ensue in part from studies that examine developmental progressions in neural development (e.g., brain imaging over time; see Giedd, Vaituzis, Hamburger, Lange, Rajapakse, Kaysen, Vauss, & Rapoport, 1996) in relation to changes in behavioral expression (e.g., decrease versus continuity in impulsivity in adolescence; worsening of disorganization relative to peers in early childhood and elementary school). A key question here is the extent to which behaviors that are developmentally normal at one age (e.g., impulsive behavior at age 3 years) fail to consolidate in the context of cognitive control, instead persisting until they are developmentally inappropriate (e.g., similar age 3-year level impulsivity continuing at age 6 years). ADHD may in some cases reflect residual activation patterns that fail to diminish with maturation. However, at this point knowledge about even the normal progression of and individual differences in the neural systems at issue remains limited.

A second overarching issue that we highlight here is heterogeneity, both in the pathways to ADHD and in the phenotype of ADHD. We have been careful to posit that the mechanisms we are discussing likely apply to many children with ADHD and may cut across to include children with other disorders such as some cases of conduct disorder (most of which also have ADHD). However, multiple genetic and other etiologies likely are involved in ADHD. Individual differences in, for example, neural development, have not begun to be seriously explored in samples of children or adults with ADHD.

A number of other issues in ADHD development that could be related to neuroscience have been bypassed here in the interest of brevity. For example, literatures are beginning to emerge in terms of differences between ADHD DSM-IV subtypes, gender differences in ADHD symptoms and in concomitant neural activation patterns, hormone effects on ADHD, and the role of motor circuitry (see footnote 2). We defer discussion of those topics.