We propose that increased intra-individual behavioural variability in continual responses to rapid, externally-paced stimuli are related to other factors than those causing increased variability in free-operant, non-externally paced tasks 8920. The former is timed by objective clock time, the latter by behavioural events not necessarily highly correlated with objective time. We define intra-individual variability as short-term fluctuations in the performance of an individual over a time-scale of seconds. It is differentiated from systematic and longer-lasting time-related changes in task performance due to practice, learning, developmental growth, or to remission or progression of a clinical condition or disease. This intra-individual variability will be the cause of inter-task variability. While not central in ADHD research, intra-individual response variability has been used as a primary dependent variable in research in normal ageing and in neurological populations 21222324252627. Evidence that individuals with mild dementia exhibit greater intra-individual variability than both healthy adults and those with arthritis suggests that it reflects neurological compromise rather than deterioration of general health 2228. Moreover, findings from studies incorporating structural or functional human brain mapping methods indicate that intra-individual variability is not simply a sequela of general brain dysfunction, but is likely related to the functioning of neural circuits that engage the prefrontal cortex, particularly the dorsolateral areas 252930. Behavioural findings indicate that intra-individual response variability is a strong predictor of success 30, suggesting that poor performance on control tasks like Go-No-Go, inhibition versions of continuous performance tests (CPT), and stop-signal task may reflect problems in response variability rather than poor inhibitory control per se.

Prior to summarizing the empirical research on intra-individual variability in ADHD, a comment is warranted on the various approaches to its measurement. The most common method of characterizing variability in response or reaction times (both denoted here by RT) consists of a single-point estimate of the standard deviation (SD) around the mean reaction time (MRT) for each individual (ISD). This method has the advantage of simplicity, but because it aggregates response indices across time intervals, it conflates effects due to practice, learning, overall speed of responding, and randomness, with possible systematic periodic fluctuations in response times. Most importantly, response variability is usually strongly correlated with MRT 31; computation of an intra-individual coefficient of variation (ICV = ISD/MRT) controls to some extent for the individual's overall speed of response. Two other measures capitalize to some extent on information that may be derived from serial analysis of response, but do not permit examination of potential periodicities within the moment-to-moment fluctuations in performance. Consecutive variability (CONV = √(∑(RTi - RT_i-1)²/(n - 1)), where I = trial number, n = number of trials, √ = square root) provides a measure of local unpredictability (trial-to-trial variability) as well as controlling overall MRT. Autocorrelations (correlations between consecutive or subsequent observations) measure the extent of linear association between one response and subsequent responding, thereby providing insights into the consistency and predictability of behaviour 825. Another approach has been to fit trial-by-trial RT data to the ex-Gaussian distribution, which provides parameters that may be related to more theoretically based distributions, such as the Wald 32. This method permits determination of whether the increase in intra-individual RT variability in ADHD is a general phenomenon occurring across the whole RT distribution, or reflects a specific variability-producing process, such as attentional lapses, restricted to the tail at the slow end of the distribution 27. The ex-Gaussian analysis delivers two measures of variability: one from the fast, Gaussian portion of the distribution, and one from the slow exponential tail 31. In normal development and ageing, age-related change in intra-individual variability throughout childhood affects the distribution as a whole, whereas in adulthood, differences in variability appears to be due to factors influencing primarily the slow end of the RT distribution, such as attentional lapses 2233.

The fast Fourier transform, which measures the power of periodic changes at different temporal frequencies, permits an assessment of periodicity over time 34. Any periodically recurring patterns of responding within the RT series are manifest as peaks of power at particular frequencies.

Significant and reliable differences in the speed and variability of responses have been documented between ADHD and comparison groups across a wide variety of neuropsychological tasks. Increased variability is seen in tasks requiring continual responses to rapid, externally-paced stimuli as well as in "free-operant" tasks without such a requirement. Recent studies of South African children from 7 ethnic groups and Norwegian children, using a computerized game-like free-operant task 89, showed significantly lower predictability of responding in ADHD than in non-ADHD groups. Predictability of response location and timing were measured in terms of variance explained by autocorrelations. Interestingly, in the free-operant task without externally-paced stimuli, response location – but not response timing – was a sensitive behavioural measure. In tasks requiring continual responses to rapid, externally-paced stimuli, however, children with ADHD display greater intra-individual variability in response times and are often found to respond more slowly and less accurately than their typically developing peers 6. These performance differences have been found on the stop-signal task 35363738, a variety of CPT 394041, time perception, time reproduction and motor timing tests 42434445, and visuomotor preparation tests 46. Notably, a recent psychometric analysis of several measures of intra-individual variability (ISD, ICV, CONV) as well as measures of central tendency and shape of the reaction time distribution derived from children's performance of four different neuropsychological tasks (stop signal task, CPT, Go-No-Go, n-back working memory task) yield two important findings 47. First, measures of intra-individual variability best discriminated between the ADHD and control groups; second, intra-individual variability appeared to be a unitary construct in ADHD (individuals with high variability on one task tended to have high variability on other tasks).

Poorer performance on neuropsychological tasks in ADHD is typically interpreted as evidence of executive function deficits, an interpretation that is based on slower and less accurate responses averaged across time. Accompanying group differences in intra-individual response variability are typically ignored – or, worse, become a nuisance parameter that keeps the inferential statistics from achieving significance. For example, the interpretation "poor inhibitory control" is commonly based on slower RTs to the stop signal or more commission errors in the CPT. Yet, it is often intra-individual variability in RT that correlates with the behavioural symptoms of ADHD, not the mean of the criterion variable 38404849. Response variability correlates more strongly and reliably with ratings of ADHD symptoms compared to commission errors or other outcome measures on the CPT in a large epidemiological sample 40 and compared with stop signal reaction time (i.e. index of inhibitory control) in a large twin sample 38 or average response speed in a community sample of typically developing children 48.

Application of the ex-Gaussian model to RT data indicates that the greater response time variability in individuals with ADHD is most evident in the slow tail 505152: There appears to be some variability-producing process, such as momentary attentional lapses, that affects the slow RTs 50. Recent evidence from Fourier analysis found that children with ADHD showed increased variability at all parts of the RT spectrum 53, consistent with Williams et al54 who showed that intra-individual response variability in the fast tail of the distribution is distinct from that found in the slow tail. However, there is significantly (50%) more intra-individual variability in reaction time of individuals with ADHD at a modal frequency of around 0.05 Hz, indicative of lapses of attention 2 to 4 times per minute 53.

Variability is increased at all parts of the distribution 53 which is consistent with Williams et al. 54 who showed that that intra-individual response variability in the fast tail of the distribution is distinct from that found in the slow tail.

Intra-individual variability in response time has been proposed as a possible endophenotype with the potential to index genetic vulnerability to ADHD 1055. Further support for this proposition comes from two molecular genetic studies that report associations between the 10-repeat risk allele within the dopamine transporter gene (DAT1) and response variability in ADHD 5657. The dopamine transporter is the main site of action of methylphenidate, which reduces intra-individual response variability in ADHD 355358.

Transient fluctuations have also been revealed in recent studies using innovative methods to capture shifts in attentional processing within relatively short time-scales, ranging from milliseconds to minutes. For example, rapid serial visual presentation (RSVP) paradigms such as the attentional blink design permit evaluation of temporal characteristics of information processing within a time-scale of milliseconds 6364. In this paradigm a stream of stimuli is presented briefly (100 ms) in the same location and in rapid succession (e.g., 10/s) and subjects are required to detect two targets in the stream. When the targets are presented within about 500 ms of one another, detection of the second target is impaired – a phenomenon termed the "attentional blink". Children and adults with ADHD and highly impulsive adolescents have been found to show a larger and more protracted attentional blink than their non-ADHD peers, suggesting less efficient attentional processing and/or more rapid depletion of processing resources 646566.

Transient fluctuations in attentional state over a somewhat longer time-scale (30 s periods) have been demonstrated by means of error analysis on computerized CPT coupled with an infra-red motion analysis system. Analysis of attentional states over a period of 30-s revealed that children with ADHD exhibit many more shifts in attention states and spend much less time in an on-task attention state than their peers 67. Fidgetiness, as detected by the motion-analysis system, was found to be related to the percent of time spent in the 'distracted' attentional state, in which the children's awareness of the relevant stimuli was diminished. These measures of attentional state provided more robust indicators of the performance differences between children with ADHD and comparison children than did the traditional time-averaged CPT summary measures of error rates, latency and variability 67. Fidgetiness, that most public characteristic of ADHD, might be a by-product of fatigue of relevant neuronal circuits, and a search for new ones to engage.

Reaction time variability is reduced with appropriate monetary reward 45, but such rewards are not differentially effective for children with ADHD. The reason for this may be that reinforcers may be less effective for tasks requiring continual responses to rapid, externally-paced stimuli than for subject-paced (free-operant) responding 68. The reason for this is probably that reinforcers work on free-operant behaviour, but not on instructed behaviour typical in tasks requiring continual responses to rapid, externally-paced stimuli (cf., 68). Methylphenidate medication improves accuracy, speeds reaction time, and reduces reaction time variability in individuals with ADHD 3558. Moreover, methylphenidate has been shown to improve time on-task, with concomitant decreases in the number of shifts in attention as the task progressed 67. Such treatment effects suggest that the increased availability of extra-neuronal catecholamines arising from the blockade of their uptake by methylphenidate could lead to increased activity of noradrenaline at β-adrenoceptors and possibly also dopamine acting at D1 receptors found on astrocytes 1569. A key fact is that stimulation of α₁- and β-adrenoceptors on astrocytes enhances glycolysis and lactate production 69. Dopamine acting on D1 receptors may have similar effects, but these are not well documented.

At present, various tasks used to measure different aspects of executive function usually have moderately heavy cognitive demands and share requirements for continuous speeded responding to computer- or experimenter-produced stimuli, rather than being self-paced. Experimenter-paced measures of working memory are substantially better predictors of reading letter span, continuous operation span, literacy and mathematics scores than are more traditional, subject-paced measures 70. We believe they provide better diagnostics of ADHD because, like stress tests for cardiac function, they most directly challenge the subjects' short-term reserves of energy, and that is a major resource compromised in ADHD.

Slowing of MRT and increasing RT variability over time has been postulated to reflect mental fatigue, 'resource depletion', motivational deficits, fluctuations in executive control, or underlying neurobiological disturbances (e.g., motor timing deficits), which are caused by the need for continuous speeded responding to high-demand tasks 6207172737475. Most of these models cannot account for the moment-to-moment fluctuations in response accuracy characteristic of ADHD.

The dynamic developmental theory of ADHD 8920 explains intra-individual behavioural variability as deficient acquisition of stimulus control of long chains of behaviour. This deficiency is rooted in reduced efficacy of reinforcers ("rewards") combined with poorer extinction ("unlearning") of inefficient behaviour. The dynamic developmental theory addresses free-operant behaviour without time constraints. However, neither the dynamic developmental theory of ADHD, nor other theories and explanations account for the intra-individual performance variability of ADHD on tasks that require continual responses to rapid, externally-paced stimuli. The present paper addresses this aspect by postulating that reaction time variability and other manifestations of transient fluctuations in performance, which arise in the context of the need for continuous rapid neuronal firing, reflect the effects of transient depletion of neuronal energy on the efficiency of information processing.

The ionic composition of the cell cytoplasm is very different from extracellular ion concentrations. In the neuron the ionic gradients across the membrane constitute a store of potential energy that can drive the influx of Na⁺ and efflux of K⁺ ions to generate an action potential, and the influx of Ca²⁺ ions to trigger neurotransmitter release and to generate Ca²⁺ waves in and between both neurons and glial cells. To maintain repeated firing over an extended time, neurons require energy to restore the trans-membrane gradients of Na⁺, K⁺ and Ca²⁺ ions. Neurons generate the necessary energy in the form of ATP; this drives the membrane-associated Na⁺/K⁺ATPase to pump Na⁺ out of the cell and K⁺ back into the cell. ATP is also required by Ca²⁺ATPase to pump Ca²⁺ out of the cell or into intracellular stores (Figure 2).

Neuronal activity is tightly coupled to glucose utilization 91. Neural activity triggers oxidative metabolism (to produce ATP) within neurons followed by breakdown of energy stores (glycogen in astrocytes) and uptake of glucose from blood capillaries into astrocytes to produce lactate 91. Rapid neuronal firing, however, is sustained by the astrocyte-neuron lactate shuttle 92. Lactate is the essential energy source for rapidly firing neurons. It is a more efficient fuel than glucose because it is metabolized to form ATP more rapidly and, unlike glucose, does not require ATP for its metabolism 76. It is imperative for neurons to make use of the most efficient energy supplies when rapid neural processing is required and demands for energy are high, and the brain has evolved ways to do so 93. The neurotransmitter glutamate (released by the majority of excitatory neurons in the CNS) stimulates glycolysis (glucose utilization and lactate production) in astrocytes and release of lactate into the extracellular fluid 9194. An inadequate supply of lactate during periods of rapid neuronal firing, when local energy demand is high, briefly impairs neuronal function, particularly during the latter part of a train of neural impulses and shortly thereafter, causing an extended refractory period. This increased latency to new information is measured in the attentional blink paradigm (section 2.1.3). Compromised energy supply, as hypothesized for ADHD, should impair response inhibition or alternation at these times.

Periods of rapid neuronal firing will be followed by slow unsynchronized firing exerting less demand on energy resources, allowing replenishment of energy reserves and restoration of function. Energy reserves will depend on the prior history of neural and astrocyte activity. Brief periods of energy insufficiency followed by periods of normal supply are proposed to account for the variability of behavioural response seen in ADHD when performing complex tasks that require speed and accuracy.

This section first considers the limited direct neuroimaging evidence of altered energy utilization in the CNS of people with ADHD that could potentially derive from deficient astrocyte function 116. We then discuss electroencephalographic (EEG) recordings indicating inefficient communication within and between neuron clusters that could be attributed to impaired energy supplies.

Energy utilization in the brain associated with the cognitive challenge of CPT performance was measured by positron emission tomography (PET) and deoxyglucose in 25 adults with childhood onset of ADHD 123. The authors reported decreases in utilization in 30 of 60 regions analysed, and a global reduction of >8%. The largest decreases were registered in adjacent regions of the superior frontal, premotor, and somatosensory cortices. Methylphenidate in healthy volunteers has been shown to increase glucose utilization in these and other brain regions (e.g. cerebellum), perhaps by elevating extracellular noradrenaline concentrations. Decreased glucose utilization in ADHD may result from inadequate noradrenergic stimulation of astrocytes. But decreased glucose utilization in striatum and differences following acute vs. chronic methylphenidate treatment are results that warn against making simple generalizations about the effects of stimulants on energy utilization 124.

Some EEG measures indicative of ADHD may reflect impaired immediate energy supply from astrocytes (H-1) and others impaired myelination (H-2, section 3.4.3/4) deriving from a longer-term lactate deficiency. In either case the presence or absence of effort to overcome the inefficiency should be visible in terms of the amplitude of event-related potential (ERP) components in the EEG representing the stages of information processing, and deficient levels of 'activation' in the ERP latencies.

The contingent negative variation (CNV) is an excitatory, slow, negative-going waveform that occurs in the second before a stimulus. The CNV is usually considered to index anticipatory and preparatory processes to a stimulus that may require a response. It has been reported to be reduced in amplitude in people with ADHD 125126127, although this was not replicated in two further reports 128129. The extent to which the reduction of CNV reflects reduced effort in or motivation for preparation is secondary to the point at issue here, namely that subjects with ADHD are poorly prepared for the stimulus. This may be due to impaired learning, or poor association of actions with delayed outcomes in previous experiences, as suggested by Sagvolden et al. 20, and this is non-adaptive to the demands of the situation. The P3 component elicited by a rare or meaningful stimulus represents updating of working memory-like templates of its associations. The amplitude is reduced in most studies of ADHD. It has been argued that reduced effort makes a contribution to this effect 130131. These are two of several ERP markers reflecting inefficient information processing in ADHD. We propose that insufficient energy reserves underlie this deficiency.

Variable ERP latencies are widely reported, especially for the P3. This variability is evident in 8–9 year-old children born very prematurely and given the diagnosis of ADHD, compared to those who had a similarly low birth weight without the ADHD diagnosis 132. Methylphenidate treatment decreased variability both of the P3 component 133134 and the response times recorded 135136. The number of responses that were too early (premature, impulsive) or too late were reduced with psychostimulant treatment 135137. Effective therapeutic drugs blocked dopamine and noradrenaline transporters and increased extracellular levels of catecholamines permitting activation of β-adrenoceptors (and/or D1 receptors) on astrocytes to stimulate lactate production.

In the human brain the myelination of axons – a principle function of the oligodendrocytes – starts in utero and continues through the first 3–4 decades of life 138. The increase of white matter during development puts high demands on metabolism. Myelin starts to form before birth with much being laid down in the first two years of life. Myelination is prominent in the prefrontal cortex, and of course in the major fibre tracts, especially the corpus callosum – with levels of "adult functionality" achieved normally during adolescence 139140. Impaired myelin synthesis as a result of intermittently insufficient lactate production by astrocytes would have a detrimental effect on the coordinated myelination of axons during development, and lead to impaired communication between brain regions and poor integration of information in these target brain structures.

Oligodendrocytes normally oxidise glucose and lactate at far higher rates than either neurons or astrocytes 14141. Glucose is metabolised in the pentose phosphate pathway to produce NADPH and in the glycolytic pathway to produce lactate, both of which are used to synthesize lipids: lactate is the preferred substrate for fatty acid synthesis and myelin formation. As oligodendrocytes are the highest lactate and glucose consumers in the developing brain, any deficiency in the supply of lactate would impair lipid synthesis and retard myelination of axons. Impaired myelination reduces the speed of conduction of action potentials in individual neurons and could account for slow reaction times. Axons that are not properly insulated by the myelin sheath would be less efficient at conducting action potentials and require more energy resources than normal. Partial myelination of fibre pathways would also contribute to impaired integration of information in target regions and provide a further source of response variability.

Myelination is a sensitive indicator of functional brain maturation. Across at least the first two decades of development myelination consists of a broad increase in overall white matter density as well as a more region-specific progression, proceeding from posterior to more anterior regions, 142143144145. In magnetic resonance spectroscopy (MRS) and diffusion tensor imaging (DTI) studies, maturation of relatively restricted regions of white matter has been found to correlate with the development of cognitive functions such as working memory capacity (left frontal regions), reading ability (left temporal lobe) and even with IQ (bilateral association areas 144146. In a study of 100 children with evidence of delayed development but otherwise normal MR-scans, Pujol and colleagues 143 were able to show a reduction of myelination (in part asymmetric) equivalent to normally developing children who were some 3 years younger. Also, at the other end of the lifespan, white matter damage due to axonal loss that occurs in normal ageing has been found to correlate with working memory performance, even after controlling for age 19. This suggests that working memory performance may be particularly dependent on complex networks, which in turn depend upon white matter connections. These studies conclude that there is a positive relationship between the density and organization of myelinated fibres and the efficiency (maturity) of cognitive function 144.

PKU results from high concentrations of phenylalanine (Phe), that arise from an inability to convert it into tyrosine, and which inhibit transport across the blood brain barrier of neutral amino acids such as tyrosine and tryptophan necessary for the synthesis of the three principle monoamine transmitters. If children are placed on Phe-free diet early enough, microcephaly, mental retardation and motor problems can be avoided, although sub-clinical symptoms may remain, particularly in the cognitive domain 183184. Children with PKU (off-diet) are rated by parents and teachers as more distractible, hyperactive, and impulsive than healthy controls 185, with many symptoms similar to ADHD (e.g. restlessness, fidgeting, concentration difficulty, short attention span, low frustration tolerance 186). Realmuto et al. 187 noted that 9/13 of their subjects either manifested or had a history of comorbidity with ADHD. Prenatal exposure was associated with a higher likelihood of expressing hyperactive/impulsive symptoms and postnatal exposure was associated with a higher likelihood of expressing inattentive symptoms 188. Indeed, a considerable proportion of people with treated PKU take psychostimulant medication for their attentional problems (e.g. 26% of a sample of 38 school-aged children with PKU 189).

Cognitive impairments of individuals with PKU like those with ADHD include variable reaction times, sustained attention, working memory, executive function, planning, learning, tests of colour-naming, arithmetic, verbal abilities and academic performance 187190191192193194195196197198199. Inter-hemispheric interactions, as measured by slowed transfer times 200, and a lack of the across-hemisphere advantage in performing a name-identity task 183, are affected in comparison to neurologically intact children. Neurophysiological studies report slow latencies for several early EPs 201, the later P3 component and slower more variable motor reaction times 194195202. In the EEG slow (theta) rhythms are characteristic of individuals with PKU. Indeed their high theta/alpha ratio (recalling ADHD) was sensitive to treatment reducing the levels of Phe 203204.

Functionally, these features reflect delayed myelination, low monoamine levels, and impaired energy availability. Impaired myelination is the primary neuropathological feature of treated or untreated people with PKU: A delay is typical of the younger ages and diffuse demyelination is reported at older ages 205206207208. The finding of low HVA levels (dopamine metabolite) in the CSF 209 confirms the second feature, low dopamine levels. Importantly, changes of dopamine in the rodent model go hand in hand with the depletion and restoration of Phe levels 210. Noradrenaline and serotonin would also be expected to be affected. The third feature was addressed in 11 young adults with PKU using phosphate MRS. Pietz et al. 204 describe changes of cerebral energy metabolism that could underlie reduced transmission speed, myelination and catecholamine availability 205. Among 11 measures taken at baseline, only ADP was significantly elevated, and inorganic phosphate decreased. Phe loading then decreased phosphocreatine and ATP levels while further increasing ADP. This is consistent with Phe inhibition of pyruvate kinase and the concurrent conversion of ADP to ATP 211. Impaired pyruvate synthesis would reduce lactate production and the ability of astrocytes to meet the energy requirements of sustained rapid neural firing and a shortage of lactate would also impair the ability of oligodendrocytes to synthesize myelin.

Narcolepsy is a neurological disorder characterized by excessive daytime sleepiness 212. Some of the neuropsychological characteristics of narcolepsy are strikingly similar to those seen in ADHD and children with ADHD have an increased tendency to daytime sleepiness 213. Individuals with narcolepsy have slower reaction times and more within-task variability of performance than control subjects on a variety of attentional tasks ranging from those sensitive to arousal and sustained attention, to the executive control of attention 214. Recent studies report narcolepsy-related deficits in attentional and executive function which place high demands on inhibition and task management, but not on simple tasks of memory and attention 214215. The pattern of findings was thought to be indicative of a depletion of available cognitive processing resources because of the need for continuous allocation of resources to monitoring.

Like ADHD, hypoarousal (sleepiness) is induced in situations of low stimulation, such as reading or boring repetitive tasks 112216217218. Similarly, children with ADHD or narcolepsy appear to use motor overactivity and fidgetiness to counteract their drowsiness 218219220. It is interesting to speculate how a deficiency in energy supply might contribute to the other symptoms of ADHD, such as hyperactivity. Depression of sensory modules could certainly lead to a suboptimal level of stimulation, inducing sensation-seeking behaviour, with motor, vocal, and other activities displacing the contextually "appropriate" behaviours, which through local energy deficiency can no longer provide the necessary arousal. Derangements of calcium-dependent protein phosphatase and kinase activity impair working memory 221. These and other implications of the energetics hypothesis require further consideration, lying beyond the scope of this paper. Lastly, as in ADHD analyses, EEG rhythms show the ratio of theta to alpha or beta power in narcolepsy to be higher than normal 172. Narcolepsy has been attributed to depletion of the transmitter hypocretin, a hypothalamic neuropeptide that regulates energy metabolism and dopamine activity in certain brain regions 222223224. Modafinil, the drug normally used to treat narcolepsy, is effective in treating ADHD and can inhibit dopamine re-uptake 220225226227228. It is also an α₁-adrenoceptor agonist and can therefore facilitate β-adrenoceptor-stimulated lactate production in astrocytes 229230. Thus, the mechanism of treatment could be the same in both disorders. This highlights the similarities between the disorders that could reflect common underlying disturbances.

There is a need for more precise measures of the dynamics within and between the major compartments of energy production, storage and utilization through studies using labelling, stimulation and inhibition of the various constituents in animal models of ADHD and tissue culture (e.g. delineation of the quantitative relationship between neuronal firing rate and lactate utilization, intra/extracellular glutamate flow, mitochondrial function (ATP/ADP ratio and oxygen consumption), effects of lactate restriction on ionic gradients, function of regulatory factors (e.g. neurotransmitter/neuropeptide receptors, particularly noradrenergic α₁-, α₂- and β-adrenoceptors), effect of impaired lactate production on myelin synthesis (and accumulation of NAA), influence of NAA availability on acetyl-group incorporation into myelin). Regional differences in measures of effects of transient energy deficiency are important to note, since the hypothesis does not predict that all parts of the brain will be affected equally. Greater effects (e.g. reduced size) should be observed in those brain areas that contain or form part of rapidly firing neural circuits that transiently deplete local energy reserves and hinder synapse formation.

Measures of some of these effects are becoming technically feasible for human studies, in vivo, with MRS. For example, changes in glutamate production (tricarboxylic acid cycle), lactate synthesis (glycolysis) and glutamine synthesis have been demonstrated in neurological patients using labelled carbon (¹³C) MRS 231232. Further data on the energy metabolites ATP, ADP, inorganic phosphate, phosphocreatine and creatine obtained from more conventional proton and phosphate MRS studies would also be useful (cf. section 3.5.1). A second example is based on the changes of intracellular calcium oscillations and extracellular calcium waves generated by neurotransmitter stimulated glia. These calcium signals stimulate glutamate release, modulate neuronal excitability and carbohydrate metabolism 8689. Changes in calcium concentration can be monitored by new contrast substances being developed for MRI investigations. These techniques, and recent applications of 2-photon optical calcium imaging in animals, could also address the role of dopamine D1 and D2 receptors 233 as well as metabotropic glutamate receptors and their modulation by α₁-adrenoceptors in the stimulation of calcium responses in astrocytes 85234235. A third example is the need for cross-sectional if not longitudinal MRS data on myo-inositol. This is suggested to be a marker of the integrity of glia and glial transport mechanisms, but the signal has proved difficult to separate from that for the large amounts of glycine present that also resonate at 3.6 ppm 236. Whereas increased levels of myo-inositol are claimed to reflect gliosis that would not be expected in ADHD, decreased levels, as reported for major depressive disorder, may reflect glial loss or altered glial metabolism 237. Indeed a functional decrease may apply to ADHD. A preliminary report on 15 subjects with ADHD found an increased glutamate/myo-inositol ratio 238, that would be consistent with either increased extracellular glutamate concentrations or an inadequate supply of myo-inositol.

The principle claim of our hypothesis, that lactate production and availability is impaired, may be difficult to measure directly in human subjects, but may be best tested with a pharmacological challenge in animal models of ADHD such as the spontaneously hypertensive rat (SHR) 239240 and poor (and impulsive) performers on the 5-choice serial reaction time task that show low 2-deoxyglucose uptake (index of brain glucose uptake) in the cingulate and ventrolateral orbital cortices during performance of the visuospatial task 241. Lactate production should be recorded at baseline, during performance of the task, and after methylphenidate, atomoxetine or venlafaxin treatment (transport inhibitors of the three monoamines). The effect of pretreatment with monoamine receptor antagonists could also be investigated. The link between performance, energy availability and at least some of these treatments should be established. Other cognitive tests that make use of dynamic strategies (change in contingency) can be used to evaluate possible correlation between changes in lactate production/utilization and performance speed/accuracy.

Glutamate availability for uptake into astrocytes and stimulation of lactate production requires special attention. Impaired release would reduce astrocytic lactate synthesis. The impairment could lie with SNAP-25, a protein important for the release of the transmitter. There is some evidence for an association between polymorphisms of the SNAP-25 gene and ADHD 242243. A potential relationship to energy production should be investigated. Lactate synthesis is dependent on glutamate uptake into astrocytes. Problems with glutamate transport could give rise to deficient lactate synthesis. Metabotropic glutamate receptor function and the glutamate/aspartate transporter (GLAST) have been suggested to play a role the development of fatigue 78244. To test this hypothesis, glutamate metabolism could be measured in astrocytes of animal models of ADHD using an experimental setup that simulates the role of neurons (glutamate producers and glutamine consumers) by the addition of glutaminase to the culture medium 245. A steady supply of glutamate can be imposed at the expense of glutamine, and the stress intensity manipulated by changing the glutaminase concentration 246. The extracellular concentration of glutamate will provide information on the efficiency of flux through the glutamate transporter and glutamine synthetase system in restoring the extracellular concentration of glutamate to a low level 247. This will provide a measure of the glutamate drive of lactate formation.

More quantification of developmental changes in myelination in ADHD, using sensitive neuroimaging techniques such as DTI, MRS, EEG or MEG would be informative. More simply, for indications of impaired white matter generation blood samples should be taken from healthy children and those with ADHD to investigate lipid availability (serum fatty acid levels and the nature of red blood corpuscle membranes). This approach has been useful in showing changes in patients with schizophrenia and depression (e.g. reduced polyunsaturated fatty acids like arachidonic and docosahexaenoic acid 246247, decreased linoleic acid 248, less high-density lipoprotein levels 249). An early report on hyperactive children suggested there were slightly decreased levels of poly-unsaturated, but increased levels of saturated fatty acids 250. Chen et al. 251 found decreased nervonic acid, linoleic acid, arachidonic acid, and docosahexaenoic acid in red blood cell membranes of children with ADHD while plasma gamma-linolenic acid and red blood cell oleic acid were increased. Irmisch noted a potential relationship with stress experienced by their subjects 250. As glucocorticoids modulate membrane stability and the adrenal hormone dehydroepiandrosterone influences lipid synthesis, both steroids should be monitored in the recommended study 252 (section 3.4.1.3). The availability of these fatty acids (particularly of the omega series), as well as marked prostaglandin synthesis (e.g. cyclooxygenase) around puberty are crucial determinants of synapse modification, pruning and maturation processes 253 that may be delayed in ADHD. These studies would usefully supplement further delineation of the fibre bundles that are 254 or in ADHD are not showing normal developmental increases of anisotropy studied with DTI.

Of the numerous potential genetic sources of the modification of energy availability, we would select those that regulate lactate metabolism (lactate, glutamate and glucose transporters, glycogen synthase, glycogen phosphorylase, glycolytic enzymes), factors that regulate glutamate release from synaptic terminals and hence glutamate-stimulated lactate production (e.g. SNAP25, vesicle transporter, metabotropic glutamate receptors, adenosine A1 receptors, neurexins, factors that regulate intracellular Ca²⁺), and astrocyte function (e.g. noradrenergic α₁, α₂ and β-adrenoceptors, dopamine D1 receptors). Mitochondrial function should also be considered, as impairment of gene transcription can occur through the accumulation of mitochondrial DNA mutations 255. One potential candidate is Ant1, a mitochondrial ATP/ADP exchanger that facilitates efflux of ATP out of the mitochondria and glutamate uptake into astrocytes 256. Metabotropic glutamate receptors are of special interest as stimulation can produce marked Ca²⁺ oscillations. Its expression represents a glial sensor of the extracellular glutamate concentration that can be used to acutely regulate excitatory transmission 257.

Likewise, there are many genetic influences on lipid synthesis and myelin formation of potential relevance to our hypothesis on oligodendrocyte function in ADHD. We select but one for attention. Lingo-1 is expressed in oligodendrocytes and is critical for CNS myelin formation. Treatment of neuron and oligodendrocyte cell culture with soluble Lingo-1 (a transmembrane protein binding to the Nogo-66 receptor/p75 signaling complex) led to highly developed and differentiated axons 258259. In animal studies it appears that its up-regulation may be a characteristic of activity-induced neural plasticity responses 260. Thus, its impaired expression could be a feature of several developmental disorders. We would also advocate the initiation of microarray studies of gene expression in samples taken from subjects with ADHD. This large scale profiling technique applied to patients with schizophrenia has already shown a surprising number of genes related to energy metabolism, oligodendrocyte function and myelin formation that are associated with psychopathology 261.