EDITORIALS Microarrays, Markers of Disease, and the Myth of “Nonhypothesis-Driven” Research The article by Tang and colleagues1 in this issue of the Annals reports that differences in gene expression patterns in blood mononuclear leukocytes, measured with DNA microarrays, can distinguish a variety of experimental cerebral disorders. This is an important contribution because of the potential for applying genomic and proteomic technology to identifying peripheral markers of neurological disease. While a handful of articles have characterized gene expression patterns in the brain in neurological disorders, this is the first to demonstrate the feasibility of using microarray data for neurological differential diagnosis. Moreover, because distinct patterns of altered gene expression were detectable in blood, this approach could lead to minimally invasive diagnosis of neurological disorders. DNA microarrays, or gene chips, consist of hundreds or thousands of DNA fragments spotted onto a glass slide or other support, and provide a means for comparing the profiles of messenger RNA (mRNA) expression in two tissue samples. This is accomplished by labeling mRNA from the two samples with different fluorescent dyes and measuring their relative binding to each spot on the array. Like any other investigative method, DNA microarrays have strengths and weaknesses. A major strength is the ability to monitor the expression of thousands of genes simultaneously, and thereby to profile changes in their expression associated with a biological or pathological process. However, because gene-chip microarrays measure mRNA abundance, they identify only alterations in gene transcription and cannot detect whether changes in the level or function of proteins occur as well. These and other issues surrounding the use of DNA microarrays in neurobiological research are discussed in detail in an excellent recent review.2 Two of the principal uses of DNA microarrays are for fingerprinting biological or pathological processes and for discovering new genes involved in these processes. An example of the former is the finding by Tang and colleagues that disorders such as cerebral ischemia, intracerebral hemorrhage, seizures, hypoglycemia, and hypoxia produce distinct gene-expression profiles in blood leukocytes. This suggests that genechip technology could be useful as a minimally invasive approach for differential diagnosis. In addition, there is the prospect for improved understanding of pathogen- esis and perhaps, as a result, new treatments. Thus, recent studies published in the Annals of Neurology have used microarrays to identify changes in the expression of genes not previously implicated in multiple sclerosis3 and Alzheimer’s disease.4 When microarrays are used for fingerprinting, reproducibility is critical; for discovering clinically relevant genes, however, it is most important that observed changes in gene transcription are accompanied by changes in protein expression and associated biological effects. In short, assessing the validity and value of microarray results requires that the experimental context be considered. This brings up a criticism of microarray studies that has achieved some currency, namely, that such studies are by their nature “nonhypothesis-driven.” The implication seems to be that if one tests 10,000 genes seriatim for disease-related changes in expression, this represents 10,000 hypothesis-driven experiments, but if all the genes are tested simultaneously on a single chip, the hypotheses are somehow annulled. The clear hypothesis of Tang and colleagues—that “the gene expression profile in white blood cells could be used as a fingerprint of different disease states”—was tested and affirmed. Microarray technology is starting to be used for additional applications, such as protein arrays to detect protein–protein interactions5 and even arrays of living cells.6 We do not yet know how clinically useful any of these so-called massively parallel approaches will prove to be, but the study by Tang and colleagues augurs well. As microarray technology develops, it will be important that disdain for “nonhypothesis-driven” research does not prevent its potential benefits from being realized. David A. Greenberg, MD, PhD Buck Institute for Age Research Novato, CA References 1. Tang Y, Lu A, Aronow BJ, Sharp FR. Blood genomic responses differ after stroke, seizures, hypoglycemia and hypoxia: blood genomic fingerprints of disease. Ann Neurol 2001;50:699 –707. 2. Greenberg SA. DNA microarray gene expression analysis technology and its application to neurological disorders. Neurology 2001;57:755–761. 3. Whitney LW, Becker KG, Tresser NJ, et al. Analysis of gene expression in multiple sclerosis lesions using cDNA microarrays. Ann Neurol 1999;46:425– 428. 4. Ginsberg SD, Hemby SE, Lee VM, et al. Expression profile of transcripts in Alzheimer’s disease tangle-bearing CA1 neurons. Ann Neurol 2000;48:77– 87. 5. MacBeath G, Schreiber SL. Printing proteins as microarrays for high-throughput function determination. Science 2000;289: 1760 –1763. 6. Ziauddin J, Sabatini DM. Microarrays of cells expressing defined cDNAs. Nature 2001;411:107–110. © 2001 Wiley-Liss, Inc. 695 Epilepsy After Head Injury: The Impact of Impact A bone of the skull of a 12-year old youth had been broken and depressed by a fall and by negligence had not been restored . . . . Consequently in his 18th year the youth suffered from epilepsy because of oppression of the brain. —Duretus, (1527–1586)1 But the secret I was seeking eluded me as it has eluded others before me and since that time: What is it that corrupts the nerve cells of the remaining gray matter from their true function while scars form and wounds heal, changing the orderly formation and conduction of nervous impulse into disorderly mass discharges? —Wilder Penfield2 Although head injury accounts for a relatively small proportion of the causes of epilepsy overall, the risk of epilepsy in patients who have suffered from severe head trauma is remarkably high. For example, soldiers with penetrating missile injuries of the brain have a 30 to 50% incidence of epilepsy,3 and civilian patients with trauma-induced loss of consciousness, prolonged amnesia, and subdural hematomas or brain contusion have at least a 12-fold increased risk of developing epilepsy within the next 10 years compared with the general population.4 At the other end of the scale, epidemiological studies suggest that there is minimal risk of epilepsy after concussive head injury in which the loss of consciousness or amnesia is less than 30 minutes.4 Nonetheless, many epileptologists can offer compelling anecdotes about previously healthy patients who have focal complex seizures immediately after relatively minor head injury, and who subsequently develop classic features of temporal lobe epilepsy. The clear link between head injury and epilepsy presents an opportunity for achieving a thus-far unattained goal: the prevention of epilepsy in an at-risk population. Conceptually, antiepileptogenesis strategies are the ultimate goal of epilepsy therapeutics, and the high incidence of head injury in an adult population and the predictability of subsequent epilepsy have led to a number of large-scale clinical trials looking at the potential benefits of anticonvulsant therapy in preventing post-traumatic epilepsy. Unfortunately, the results of these trials have been disappointing—a recent metaanalysis concluded that although anticonvulsants prevent seizures in the first week after injury, they do not prevent the development of epilepsy.5 The need to prevent epilepsy in a high-risk population and the current lack of any effective treatment is magnified by our limited understanding of the basic, underlying mecha- 696 © 2001 Wiley-Liss, Inc. nisms of trauma-induced epileptogenesis. This is not for want of experimental models or intensive research on head injury itself–the literature is replete with studies of the structural, biochemical, and overall functional consequences of traumatic brain injury.6 However, relatively little effort has been focused on the effects of trauma on network excitability. In 1992, my colleagues and I made the fortuitous discovery that experimental fluid-percussion injury (FPI) in the rat causes a selective loss of hilar interneurons in the dentate gyrus.7 Our observation grew from the work of Robert Sloviter, who had shown that a similar pattern of cell loss could be induced by pure electrical activation of the main afferent input into the dentate gyrus, and predicted that the selective loss of cells led to a pathological disinhibition of the dentate granule cells.8 In both models, the pattern of injury is strikingly similar to that seen in many cases of human temporal lobe epilepsy.9 Prompted by Sloviter’s work, we found that the immediate cell loss induced by the injury was associated with abnormal hyperexcitability of the dentate granule cells 1 week later, and we proposed this as a potential mechanism of trauma-induced epileptogenesis. The model turns out to be far from perfect, since the animals fail to develop clinically evident, spontaneous seizures. Nonetheless, the fact that a single, isolated, 25msec impact to the dura could lead to a prolonged defect in network excitability at some distance from the injury provided at least a new system for studying the electrophysiological consequences of head trauma. Over the next few years, additional studies both verified the pattern of selective cell loss and provided a more sophisticated analysis of the short-term abnormalities of excitability.10 –12 However, until now, studies of the long-term electrophysiological effects of FPI have been lacking. In this issue of Annals, Soltesz and colleagues provide us with some initial information about the persistent changes in network properties of the hippocampal formation following FPI in the rodent.13 Using hippocampal-entorhinal cortical slices prepared acutely from animals sacrificed at various times after FPI, they found that the evoked population spike amplitude of dentate granule cells in response to perforant path stimulation, which was abnormally elevated 1 week after injury, had returned to control levels 1 and 3 months later. However, studies of tetanic stimulation of the Schaeffer collaterals (the main input to CA1 pyramidal cells) 3 months after FPI revealed a significant decrease in the threshold for the generation of self-sustaining epileptiform activity compared with control slices (ie, abnormal hyperexcitability). Somewhat surprisingly, they also observed a long-lasting increase in the frequency of spontaneous inhibitory postsynaptic currents in dentate granule cells after FPI, an increase that appeared to be a consequence of enhanced excitatory drive onto inhibitory interneurons within the circuit. The finding of en- hanced inhibition and hyperexcitability in an epileptogenic process may seem paradoxical but is hardly unprecedented, and, as the authors mention in their discussion, there are a number of theories and supportive evidence to suggest a causal link between the two. Where do these new findings leave us? Soltesz and colleagues have provided unambiguous evidence for a chronic network defect in the hippocampal formation many months after a single episode of head trauma. Along with the work of Prince and coworkers using a different model of mechanical trauma (the cortical undercut model),14 we now have at least two systems for studying epileptogenesis associated with the types of injuries seen in patients. Given the need for the development of preventative therapies, investigators may be tempted to use these models to screen potential antiepileptogenic compounds. This would be an exceptionally risky venture, at least at the present time. The models do not lend themselves to high through-put drug screening, and the relationship between the abnormal network properties and the clinical phenotype of epilepsy, particularly in humans, is uncertain at best. Furthermore, we need to be especially selective about the choice of interventions that are proposed for prospective clinical trials looking at head injury and the subsequent onset of epilepsy. By their nature, these trials require large numbers of patients, prolonged periods of observation to verify the presence or absence of a beneficial effect, and they are too expensive to justify an approach lacking a well-founded, rational basis. This leaves us with only one viable option—determined and sustained efforts to identify the precise molecular, cellular, and systems events that are induced by trauma and transform a normal neural circuit into a seizure focus. By documenting for the first time the long-term electrophysiological changes seen in the FPI model, Soltesz and colleagues have made a significant contribution toward this goal. Daniel Lowenstein, MD Harvard Medical School Boston, Massachusetts References 1. Temkin O. The falling sickness: a history of epilepsy from the Greeks to the beginnings of modern neurology. Baltimore: The Johns Hopkins Press, 1945. 2. Penfield W. Introduction (to the Symposium on PostTraumatic Epilepsy at the American Epilepsy Society Meeting). Epilepsia 1961;2:109 –110. 3. Jennett B. Epilepsy after non-missile head injuries. 2nd ed. London: William Heinemann, 1975. 4. Annegers J, Hauser W, Coan S, et al. A population based study of seizures after traumatic brain injuries. N Engl J Med 1998; 338:20 –24. 5. Temkin N. Antiepileptogenesis and seizure prevention trials with antiepileptic drugs: meta-analysis of controlled trials. Epilepsia 2001;42:515–524. 6. Laurer H, McIntosh T. Experimental models of brain trauma. Curr Opin Neurol 1999;12:715–721. 7. Lowenstein D, Thomas M, Smith D, et al. Selective vulnerability of dentate hilar neurons following traumatic brain injury: a potential mechanistic link between head trauma and disorders of the hippocampus. J Neurosci 1992;12: 4846 – 4853. 8. Sloviter R. “Epileptic” brain damage in rats induced by sustained electrical stimulation of the perforant path. I. Acute electrophysiological and light microscopic studies. Brain Res Bull 1983;10:675– 697. 9. Margerison J, Corsellis J. Epilepsy and the temporal lobes. Brain 1966;89:499 –530. 10. Toth Z, Hollrigel G, Gorcs T, et al. Instantaneous pertubation of interneuronal networks by a pressure wave-transient delivered to the neocortex. J Neurosci 1997;17:8106 – 8117. 11. Santhakumar V, Bender R, Frotscher M, et al. Granule cell hyperexcitability in the early post-traumatic rat dentate gyrus: the ‘irritable mossy cell’ hypothesis. J Physiol (Lond) 2000;524: 117–134. 12. Coulter D, Rafiq A, Shumate M, et al. Brain injury-induced enhanced limbic epileptogenesis: anatomical and physiological parallels to an animal model of temporal lobe epilepsy. Epilepsy Res 1996;26:81–91. 13. Santhakumar V, Ratzliff A, Jeng J, et al. Long-term hyperexcitability in the hippocampus after experimental head trauma. Ann Neurol 2001;708 –717. 14. Prince D, Jacobs K, Salin P, et al. Chronic focal neocortical epileptogenesis: does disinhibition play a role? Can J Physiol Pharmacol 1997;75:500 –507. Imaging Stroke Recovery: Lessons from Prozac The past decade has seen selected but highly significant advances in the prevention and treatment of stroke, although stroke remains the most common cause of neurological dysfunction. The primary interest of stroke victims, along with secondary prevention, is the opportunity for recovery of function. Fortunately, most stroke symptoms do improve, although to varying degrees, and with variable rates. Basic research on stroke recovery has shown that the recovery process can be modulated through pharmacological and environmental factors. The extent to which these principles can be successfully applied to human stroke victims could have a major impact on the overall morbidity of stroke. The past decade has also seen a rapid expansion of opportunities to examine human brain function using noninvasive imaging, primarily with magnetic resonance. Functional magnetic resonance imaging (fMRI) has not only reduced the cost and invasiveness of functional neuroimaging, but also has increased its sensitivity such that task-specific signal changes can typically © 2001 Wiley-Liss, Inc. 697 be reliably observed in individual subjects. One of the major potential clinical applications of fMRI is to gain a better understanding of the mechanisms of functional recovery in response to brain injury, with the hopes of using this information to develop strategies to enhance this process. In the present study, Pariente and colleagues1 demonstrate that a single dose of fluoxetine increased motor performance and also produced increased fMRI activation in the ipsilesional motor cortex. While these observations are consistent with a causal role of fluoxetine on motor cortex resulting in improved performance, it also remains possible that motor activation increased in response to improved performance through some other mechanism, such as attention. Attentional effects have been previously shown to alter fMRI activation in motor areas.2 The absence of fluoxetine-correlated changes in brain regions thought to subserve attention does not provide much evidence against this alternative, as the statistical approaches currently used for fMRI analysis do not well address the issue of false-negative activation. It is also unfortunate that motor performance and motor cortex activation could not be assessed using identical tasks. This would have allowed task performance to be included as a covariate in the image analysis. The present study illustrates the challenges in determining causality based on associations between performance effects and fMRI activation. Existing data indicate that activation may increase with performance in certain regions, at least over certain ranges, while in other regions activation may decrease with increasing performance. The latter response has been observed in dorsolateral prefrontal cortex during pharmacological modulation of working memory performance.3,4 Another factor related to performance is task difficulty, which is a particular concern when activation in patients with deficits is compared with controls, even if subjects in both groups perform equally. Perhaps more convincing neuroimaging evidence of therapeutic causality might be the recruitment of brain regions not previously activated by the task, for example, the recruitment of homolateral brain regions or a spatial shift in the activation centroid. Such effects have also been observed during stroke recovery. A fundamental question concerning the specific application of fMRI to stroke recovery is the extent to which the normal coupling between neural function and cerebral blood flow (CBF) is preserved, as well as the extent to which changes in CBF with functional activation reflect neural versus vascular reorganization. In the present study, the potential influence of this uncer- 698 Annals of Neurology Vol 50 No 6 December 2001 tainty was minimized because fMRI was used to characterize changes in cortex in response to small subcortical infarcts, which presumably do not significantly affect cortical CBF. Furthermore, the authors also studied a passive control task that did not show a fluoxetine effect. At the very least, fMRI studies such as these may be viewed as hypothesis generating. In the present case, the hypothesis that fluoxetine effects on motor cortex are mediated by serotonin can be tested through further manipulations of this system. Neuroimaging methods that measure resting CBF or metabolism might also contribute to the characterization of pharmacological interventions by demonstrating their effects on regional activity independent of task performance. For example, an increase in resting metabolism in motor cortex in response to fluoxetine would further support a causal role for this therapy in motor performance. In the case of novel activations in response to intervention or recovery, their role in task performance can be tested using methods that inhibit the activity, such as pharmacological manipulation or focal transcranial magnetic stimulation. At the very best, studies such as these illustrate how functional neuroimaging performed with careful attention to experimental design and data analysis can elucidate the mechanisms and modulators of functional recovery in the human brain. With improved understanding of the relationships between task difficulty, task performance, and brain activation, such data should provide insights that will ultimately result in opportunities to improve patients’ functional outcome from stroke and other types of brain injury. John A. Detre, MD Department of Neurology University of Pennsylvania Philadelphia, Pennsylvania References 1. Pariente J, Loubinoux I, Carel C, et al. Fluoxetine modulates motor performance and cerebral activation of patients recovering from stroke. Ann Neurol 2001;718 –729. 2. Baker JT, Donoghue JP, Sanes JN. Gaze direction modulates finger movement activation patterns in human cerebral cortex. J Neurosci 1999;19:10044 –10052. 3. Mehta MA, Owen AM, Sahakian BJ, et al. Methylphenidate enhances working memory by modulating discrete frontal and parietal lobe regions in the human brain. J Neurosci 2000;20: RC65. 4. Furey ML, Pietrini P, Haxby JV. Cholinergic enhancement and increased selectivity of perceptual processing during working memory. Science 2000;290:2315–2319.