Computational Neuroscience: A Powerful Tool for Today's Clinician

Some patients with epilepsy have more treatment options today because of constantly expanding computer power, the development of model neurons and neuronal networks, and the ability of neurologists and neurosurgeons to translate medical and scientific research into clinical practice.Computational neuroscience is no longer a tool of the future. It's a tool of today and for the future, according to speakers at the American Epilepsy Society (AES) annual meeting that was held in December.In a symposium titled "Epilepsy and Computational Neuroscience: At the Threshold of a Whole New Era," 3 speakers presented the full historical spectrum of this topic, from the basic science of developing models of neurons, through the intermediate stage of designing and refining models of neuronal networks, to using these models and other computational power and technology to help treat patients. "Although the computational approaches for the understanding of seizures and epilepsy have been studied for decades, the last few years have brought major conceptual advances," said Daniel H. Lowenstein, MD, a professor of neurology at the University of California, San Francisco, and last year's AES president, in his opening remarks."It is . . . quite clear that we are just at the beginning of having a framework with which to understand the behavior of complex biological systems. Epilepsy, in particular, is clearly a manifestation of multiple pathophysiology processes impinging upon intricate neuronal networks, with the added complexity of a network dysfunction being intermittent and difficult to predict," said Lowenstein, who last year was a coauthor of a paper describing a Web-based prototype of a multimedia data warehouse on refractory epilepsy and other neurologic disorders.1 The prototype provides clinicians who treat patients who have epilepsy with a way to access shared data on surgical planning and research applications.Brian Litt, MD, associate professor of neurology and bioengineering at the University of Pennsylvania, who spoke about the clinical aspects of computational neuroscience, broke his topic into 4 basic areas: seizure localization, seizure generation, seizure prediction, "and then the exciting translation of these ideas in computational neuroscience into devices, which are going into humans right now." These devices are being designed for severely affected patients who don't respond to antiepileptic medications, he said.BRIDGING THE GAP"Epilepsy surgery has really created revolutionary results and helped patients enormously, particularly in those patients who have focal lesions, for those who have concordant medial temporal lobe epilepsy," Litt said. "But then we go into the nonlesional temporal and extratemporal patients, where our success at making them seizure-free is decreased and the morbidity and the costs of doing the surgery are high." So the tasks at hand are to create models of neuronal networks, plus perform rapid computations on neural signals and their properties, according to Litt.Some developments-not so new but being observed in a different light-will help accomplish those tasks. Electroencephalographic (EEG) technology has been in the clinic for a long time now, but use of high-frequency recordings is broadening epilepsy research and treatment. "We're starting to appreciate the importance of high-frequency [EEG] signals. No longer can we sit outside the concert hall and try to listen to the symphony of what goes on in the brain, but we need to put electrodes in and look at the bandwidth of the actual cellular structures," Litt said.In a paper published last year, Litt, Gregory A. Worrell, MD, PhD, of the Mayo Clinic in Rochester, Minn, and colleagues described new findings after using high-frequency electroencephalography to study a group of patients with neocortical epilepsy.2 They found that high-frequency (60 to 100 Hz) epileptiform oscillations were highly localized in the seizure-onset zone in 6 of 23 patients with neocortical epilepsy who had intracranial electrodes implanted for presurgical evaluation. The findings showed that current clinical EEG analysis that does not use high-frequency signals "largely ignores fundamental oscillations that are signatures of an epileptogenic brain. It may prove that many currently available clinical EEG systems and EEG analysis methods utilize a dynamic range that discards clinically important information," according to the paper.Litt explained at the December symposium that the study "demonstrated that bursts of energy in the hippocampus-the epilep-tic hippocampus-behave like a critical system, sort of like snow building up before an avalanche." He said that the filter settings on intracranial electroencephalography used commonly in clinical practice are based on machines from the 1950s. Now manufacturers are making higher-frequency machines.When Worrell looked over the high-frequency bandwidth of the EEG in the above study, he noticed that about 40 minutes before onset of the seizure, "there was a pretty impressive increase in high-frequency activity," Litt said. Although he stopped short of calling this seizure prediction, "It's identification that there are changes associated that herald the onset of the seizures."In a study published in 2001, Litt and colleagues3 analyzed intracranial EEG recordings from 5 patients with medial temporal lobe epilepsy. They identified complex activity that began 7 hours before seizures, localized activity 2 hours before seizures, and accumulated energy increases 50 minutes before seizure onset. Their conclusion was that "epileptic seizures may begin as a cascade of electrophysiological events that evolve over hours and that quantitative measures of preseizure electrical activity could possibly be used to predict seizures far in advance of clinical onset." Now, groups of physicians in the United States and Germany are making recordings in humans using microelectrode probes and are in the preliminary stages of identifying brain areas where seizures begin, Litt said.He later told Applied Neurology, "There are many tools widely available to [make use of computational neuroscience]. One generally needs fast computers, often in a cluster-we have clusters of 20 to 100 very fast computers all hooked together in our lab, and others have larger ones-and modeling tools." Among the most popular modeling tools are software programs that have been developed through government grants and are available free, including Neuron (www.neuron.yale.edu).Potential applications of these programs in the clinic would be as "algorithms to detect seizures when they occur or to predict them and direct therapy to stop these events. We are using these types of algorithms to locate where seizures come from in the brain, to run implantable brain devices, and to guide surgery," he said. "In research, we are using these methods to understand how circuits are wired in the brain to understand normal function and disease. These research tools have helped us develop new drugs and new treatment, again through devices, for Parkinson's disease, tremor, other movement disorders, pain, epilepsy, incontinence, and paralysis, to name a few applications. We are also using these tools to understand how memory works and how information is stored and retrieved in the brain."In an open-label pilot study published last year, Litt and colleagues used bilateral programmable implant devices to stimulate the anterior nucleus of the thalamus (ANT) in 5 patients with intractable partial epilepsy, 4 of whom also had secondary generalized seizures.4 Four of the 5 patients had less severe seizures, compared with preimplantation, and 1 of the patients had a reduction in total seizure frequency.Although such treatments are mostly for patients whose condition is resistant to drugs, he added, "We do analyze data from all patients, including those cured with surgery. Frankly, in the cured patients, this tells us that if our findings correlated with the removal of a brain region that makes the patient seizure-free, then we're on to something."CURRENT CLINICAL TRIALSTwo ongoing clinical trials currently recruiting patients are based on the recent advances in brain stimulation research. The trials are the Stimulation of the Anterior Nucleus of the Thalamus for Epilepsy (SANTE) trial, sponsored by Medtronic (www.clinicaltrials.gov/ct/show/NCT00101933? order = 10), and the Study of a Responsive Neurostimulator System to Treat Epilepsy trial, sponsored by NeuroPace (www.clinicaltrials.gov/ct/show/NCT00079781?der=15).The Medtronic trial, in phase 3, is testing that company's Intercept Epilepsy Control System, a brain-stimulation treatment for patients with refractory epilepsy. It is to include about 124 patients at 15 sites in the United States and Canada. Patients will receive an implanted device and will be monitored for 13 months following implant. The purpose is to test whether seizure frequency can be reduced through bilateral stimulation of the ANT. Patient candidates are adults with partial-onset epilepsy whose treatments have included at least 3 antiepileptic drugs that have been ineffective.The NeuroPace trial, in phase 2, is testing that company's Responsive Neurostimulator System (RNS). Expected total patient enrollment is 80, and 2 different protocols are involved: assessment and treatment. Physicians using the assessment protocol monitor seizure type, frequency, and severity and assess the patients' epilepsy-related physical and emotional health before and after implantation of the RNS. Physicians using the treatment protocol manage the RNS system by monitoring and adjusting parameters, when necessary, before and after implant. Patients aged 18 to 65 years are eligible, and multiple selection criteria are listed on the above Web site.PRIME FOR ADVANCESEpilepsy is a prime candidate for benefit from computational neuroscience, according to Ivan Soltesz, PhD, the second speaker at the symposium. Researchers have large databases that can be used to build realistic control and epileptic network models, and 2 areas where epilepsy and computational neuroscience have merged productively are ion channels and epileptic networks, said Soltesz, professor of anatomy and neurobiology at the University of California, Irvine.Traditional experimental research is still necessary, Soltesz said, and "it's ongoing, is very expensive, and it's time-consuming. . . . [However,] computational modeling is much faster, it is less expensive, and it has great predictive power."Speaking of models already developed by researchers, Soltesz said the next step is to make them fully Web-based and user-friendly. "If you're interested in finding out how a particular ion channel changes in a particular neuron, you can just fire up your computer, go to the Web site, and tweak a knob, if you want, on the model and see how a channel up-regulation in one particular cell type changes network behavior. We believe these models will be generally applicable," he said. "As computer prices go down every year and the power goes up, we'll be able to model more and more larger and larger networks with cheaper and cheaper computers.""The models are no longer just simulating what's happening; they're predicting behavior, they're giving us new findings, they're going to be a key toward new therapies," Litt said, and the establishment of shared data archives and processing tools that Soltesz cited sets an important precedent. "Perhaps we need a 'human seizure project,' where we have central collations of data from animal and human investigations that we can use to share and work on problems together. And perhaps we can share and analyze at least clinical downloads from our devices."Eve Marder, PhD, professor of biology at Brandeis University, gave the symposium presentation on the basic science aspects of computational neuroscience. She and her colleagues published a paper in 2003 that described a database they generated of about 1.7 million model neurons.5 That database can be searched for various combinations of neuron properties and can be screened for models that reproduce a specific neuron's biologic behavior. The database is available on-line at www.bio.brandeis.edu/marderlab/database.html and is available on a set of 2 DVD discs if requested. The neuron models in the database have no direct clinical applicability, but the database shows a foundation in basic science for neuronal modeling.References1. Cao X, Wong ST, Hoo KS Jr, et al. A web-based federated neuroinformatics model for surgical planning and clinical research applications in epilepsy. Neuroinformatics. 2004;2:101-118.2. Worrell GA, Parish L, Cranstoun SD, et al. High-frequency oscillations and seizure generation in neocortical epilepsy. Brain. 2004;127(pt 7):1496-1506.3. Litt B, Esteller R, Echauz J, et al. Epileptic seizures may begin hours in advance of clinical onset: a report of five patients. Neuron. 2001;30:1-3.4. Kerrigan JF, Litt B, Fisher RS, et al. Electrical stimulation of the anterior nucleus of the thalamus for the treatment of epilepsy. Epilepsia. 2004;45:346-354.5. Prinz AA, Billimoria CP, Marder E. Alternative to hand-tuning conductance-based models: construction and analysis of databases of model neurons. J Neurophysiol. 2003;90:3998-4015.