Address correspondence to Dr. Jeffrey Loeb, 421 East Canfield St., Elliman Research Building Rm # 3122, Detroit, MI 48201, U.S.A. E-mail: firstname.lastname@example.org
One of the major challenges in developing novel therapeutics for human epileptic disorders comes from the wide range of brain abnormalities capable of producing epilepsy. In children and adults that undergo epilepsy surgery for treatment of refractory seizures, these abnormalities range from developmental defects to injuries, infections, tumors, and ischemia. Given the many molecular mechanisms likely involved in each of these, finding common therapeutic targets seems a futile task. However, patients undergoing surgery for neocortical seizures have surprisingly similar electrophysiologic abnormalities, which consist of the synchronous firing of large neuronal populations. Surgical removal of these regions is the only means at present time to permanently reduce or eliminate seizures. The precise locations of these hypersynchronous firing regions that produce seizures can be revealed using long-term subdural electrical high-density recordings. This therapeutic strategy not only can dramatically reduce seizures, but also offers the potential to generate molecular and cellular information that can be used to ask why certain regions of the cortex become and remain epileptic. We have taken advantage of these detailed clinical and electrophysiologic human studies by taking a “systems biology” approach to identify novel biomarkers and drug targets in neocortical human epilepsy. In this article, we describe our multidisciplinary systems approach that utilizes a relational database to interrelate clinical, quantitative electrophysiologic, pathologic, and gene expression profiling data together as a means to identify and validate new biomarkers and potential drug targets for human epilepsy.
Epilepsy is a disabling neurologic disorder of recurrent seizures affecting up to 1% of the population worldwide (Annegers, 1993). Although single gene defects in ion channels or neurotransmitter receptors are associated with some inherited forms of epilepsy (Berkovic & Steinlein, 1999; Noebels, 2003; Steinlein, 2004; Reid et al., 2009), these mutations cannot account for the majority of patients with epilepsy. In most patients with partial epilepsy, seizures start in focal brain regions in response to a wide variety of brain insults, often with no clear histopathologic abnormalities (Babb & Pretorius, 1993). Patients who fail to respond to antiepileptic medications can greatly benefit from surgery to remove brain regions where seizures originate. This suggests that these focal epileptic brain regions are both necessary and sufficient to produce clinical seizures. Although in young children epileptic foci are mostly in the neocortex, in adults they frequently involve the hippocampus (Annegers, 1993). Little is known about how these often normal-appearing brain regions become and remain epileptic. Regardless of the original brain insult, neocortical epileptic foci show a remarkably similar electrophysiologic pattern of localized, abnormal electrical discharges that can become rhythmic and spread to widespread brain regions to produce clinical seizures. Between seizures, and far more frequent than seizures, these focal brain regions generate localized “interictal” discharges that can be used to help identify regions of seizure onset (Asano et al., 2003, 2009).
One thing that has become clear for a variety of both cortical malformations as well as other secondary lesions is that the structural abnormalities may not themselves be the main epileptic generator (Chugani et al., 1998; Asano et al., 2000). This is not surprising when one considers that lesions composed of tumor cells, infarcts, or glia scars are not likely to be intrinsically, electrically active. Given that most neocortical epileptic foci do not have any pathologically demonstrated malformations, it would not be unreasonable to hypothesize that epilepsy can develop in relatively normal neocortex adjacent to any brain lesion. It also raises the possibility that the final common pathway for epileptogenesis can be entirely independent of the nature of the inciting lesion.
One of the hypotheses that our laboratory has been testing for a number of years is that focal regions of neocortex that display spontaneous ictal and interictal activities are maintained in an epileptic state through the local expression of specific genes and proteins, and that by comparing these abnormal electrical regions in human cortex to nearby less abnormal regions we can identify molecular pathways that maintain the epileptic state. Our work using this approach in human neocortex has thus far has identified a number of robust biomarkers of human epileptic activity that appear to be independent of the “cause” of epilepsy and should make it possible to determine the functional as well as spatial relationship of these activity-dependent genes within “normal” appearing six-layered neocortex and in adjacent cortical lesions (Rakhade et al., 2005).
A Systems Biology Approach to Generate Unbiased Drug Targets
Systems biology can be variously defined, but one appealing definition is the ability to obtain, integrate, and analyze complex data from multiple experimental sources using interdisciplinary tools (Parker et al., 2009). The goal of this seemingly diffuse approach is to find ways to condense and focus diverse and highly variant types of data into meaningful insights. This can be achieved by developing focused hypotheses and testing each with carefully designed studies that have sufficient biologic and technical replicates to generate statistically meaningful results. The most highly utilized high-throughput method today involves gene expression profiling of known coding genes through microarray studies. Microarrays are high-density arrays of small sequences of DNA that are “spotted” onto glass slides or other surfaces so that a single microscope slide with thousands of these spots can probe a majority of genes spanning the entire human genome (Loeb & Beaumont, 2009). These slides can be costly. As a result many microarray experiments are severely underpowered statistically and produce both false-positive and false-negative results. Even more problematic are study designs that examine too many variables simultaneously, making it nearly impossible to identify those gene expression changes that are truly due to the variable under study.
For our epilepsy work, we have been comparing relative gene expression between two or more nearby regions of human neocortex from the same patient. Our goal has been to reduce the number of variables we are measuring and ask the simple question: what is different between neocortex that is more and less epileptic? All samples we take, including our nearby “control” tissues, have been subjected to long-term surface recordings. An alternative experimental design that has been used more extensively compares epileptic tissue to “control” tissue taken from a nonepileptic patient with another disorder such as acute trauma or a brain tumor (Colantuoni et al., 2001; Becker et al., 2003; Arion et al., 2006; Crino, 2007). Although this study design may also pick up important transcriptional differences due to epilepsy, it will also pick up many other differences due to other variables including: (1) Genetic background differences between the two individuals; (2) Effects of different medications (such as anticonvulsants); (3) Differences in tissue processing times and techniques; and, perhaps most importantly, (4) Not knowing whether the “control” tissue is truly electrically quiet, since the tissue is not electrically mapped with subdural recordings. Having too many gene expression changes creates a “needle in the haystack” problem that makes it nearly impossible to differentiate “epilepsy genes” from hundreds or thousands of other genes changed by these other variables.
Keeping Track of Different Types of Data
One of the greatest challenges for high-throughput experimentation is keeping track of many different types of data that range from gene expression changes to what medications the patient was on at the time of surgery (see Fig. 1). This is no small task for human epilepsy work where the electrophysiologic data alone consists of often 100 different electrode recording sites producing up to a week of continuous data each. As part of our systems biology approach, we have developed a database that incorporates many data types including clinical, electrophysiologic, tissue, and gene data into a relational database. Some of the data we collect are patient-specific such as age, sex, and seizure medications. Other types of data are electrode-specific and include spike frequency, location, and, if the tissue was removed, histology, concentrations of RNA and protein extracts, and gene expression levels. Because the neocortical anatomy of each patient is different, a critical component of this database is placing each of the electrodes into a three-dimensional brain rendering for each patient as shown in Fig. 2 (Hua et al., 2004).
The development of a user-friendly and clinically helpful web-based interface to the database is also important. When an interface is helpful only for research, but not clinically useful, it makes it likely that busy clinicians will not take the extra time needed to enter data. Creating such an infrastructure for this systems biology approach requires coordination of clinical as well as computational experts, but can create a powerful new way to identify both biomarkers and drug targets for epilepsy (Fig. 1).
Role of Ictal Versus Interictal Activity
Both the spatial and temporal aspects of epileptic activities are likely to have important roles in both epileptogenesis and the generation of clinical seizures. Quite simply, an epileptic discharge is an abnormal, hypersynchronous firing of very large populations of neurons in the brain. Epileptic discharges can occur acutely after brain injuries, as well as chronically in patients with recurrent epileptic seizures (Staley et al., 2005). Although clinical symptoms of seizures are most readily apparent, long-term brain recordings show that seizures are relatively infrequent compared to smaller, isolated or grouped epileptic discharges often referred to as interictal spikes (Tao et al., 2005). Despite the relatively higher frequency of interictal spiking compared to seizures, the roles that interictal spiking plays in epileptogenesis are not clear and their clinical significance is controversial. Although the identification and removal of regions of seizure onset have beneficial effects, the importance of removing additional regions with these frequent “interictal” spikes is less clear. For example, although interictal spikes have been qualitatively associated with increased seizures (Overweg et al., 1987), other studies suggest that because increased interictal spiking can be seen immediately after seizures that they may actually prevent seizures (Gotman, 1991). Nonetheless, studies have shown that resection of regions with frequent interictal spike activity in the temporal lobe is associated with a good surgical outcome (McBride et al., 1991; Kanazawa et al., 1996; Bautista et al., 1999). Given these clinical findings, it is not surprising that there is a growing interest in the molecular roles of interictal spiking in epilepsy (Staley et al., 2005). In fact, interictal spiking and seizure activity have been shown to arise from the same neuronal networks in hippocampal–entorhinal cortex slices (Dzhala & Staley, 2003).
Many of the genes we identified at zones of seizure onset are induced in all patients we have examined, regardless of the underlying “cause” of the epilepsy, making them excellent epileptic biomarkers. We recently asked whether the expression levels of these genes correlate with ictal versus interictal activity (Rakhade et al., 2005, 2007). Surprisingly, we found no clear correlation between a patient’s seizures frequency and the expression levels of four genes (EGR1, EGR2, CFOS, and DUSP6). Specifically, patients with more than 10 seizures per day did not necessarily have higher gene expression levels than those with one seizure or less per week. In contrast, we found that these same genes correlate precisely to the degree of interictal spiking at each electrode location, suggesting a close association or perhaps even a causal relationship between interictal spiking and genes that may keep these focal regions epileptic. These findings also highlight a lack of understanding of the relationship between interictal and ictal activities. This relationship will need to be further elucidated in both humans and animal models to determine whether interictal spiking is a prerequisite for seizures or simply a separate but related form of abnormal hypersynchronous activity, and thus whether a biomarker or drug target that affects interictal spiking is also a useful biomarker for seizures.
Quantitative In Vivo Electrophysiology
Because of the tremendous amount of data generated from subdural electrical recordings, improved, automated methods for identification and analysis of both ictal and interictal activities are needed. This is critical, since one of our central goals is to relate these activities to the molecular, cellular, and synaptic bases of these abnormal electrical potentials. In addition to identifying regions of seizure onset and seizure spread, we have developed methods to identify and quantify interictal spiking and then determine their frequency, amplitude, and duration (Rakhade et al., 2007) (Barkmeier DT, unpublished data). These data are stored in our database and used to generate detailed three-dimensional electrophysiologic heat maps applied to each patient’s anatomic brain surface rendering (Fig. 2, top).
Tissue Collection and Processing
It is important to emphasize that in no way does the research program influence the clinical decisions about what tissues should be removed; and that each patient has signed a consent form indicating that any tissues that are removed for their treatment could be used for research. As there are a number of steps and individuals involved in the identification and removal of epileptic tissues during epilepsy surgery, it is critical that we are certain that a particular tissue we record from is the same tissue that we use in subsequent experiments after the tissue is removed. To be certain of this, we take digital photographs in the operating room with the grids in place and after grid removal. These images are then placed alongside our three-dimensional renderings. At each electrode location a block of tissue is removed and divided in half (Fig. 2, bottom). One piece is fixed in 4% paraformaldehyde, sectioned, and stained for histology and other markers. The orientation of the opposing piece is maintained and stored frozen at −80°C. After we identify regions from the fixed tissues that we are interested in, we go back to the frozen piece to prepare various extracts for genomics (RNA), proteomics (protein fractions), and metabolomics (frozen tissues) (Rakhade et al., 2005; Rakhade & Loeb, 2008). This method allows us to obtain both histologic and molecular information from each electrode-recording location. Although we cannot use the same piece of tissue for all studies, we pool alternating strips within the frozen block of tissue to average out local differences. Our database is also helpful here, as it enables us keep track of which tissues have been sampled, where they are stored, and whether specific molecular components have been extracted.
Identification and Validation of Biomarkers and Drug Targets for Human Epilepsy
The most exciting next step is to identify biomarkers of epileptic activity as well as whether any of these biomarkers are potential drug targets (Fig. 3). This is where all of the work in generating a relational database of many different systems can pay off. While each patient is linked to a variety of variables including age, sex, medications, and family history, each recording electrode location is linked to a variety of electrophysiologic parameters, molecular profiles, including gene and protein expression patterns, and tissue histology. We can then generate a large number of hypotheses that can be “tested” for statistical significance through a process of data mining. For example, one hypothesis we have tested is whether or not common gene expression changes exist between regions where seizures start and nearby “control” areas without spontaneous seizures (Rakhade et al., 2005). The simple premise is that, while any high-throughput microarray experiment will show gene expression differences between two samples, the probably of finding the same changes across many patients by random chance is quite low, but can be realized if the right number of biologic and technical replicates are performed (Wolfinger et al., 2001; Yao et al., 2004).
Although the expression of statistically significant individual genes may serve as epileptic biomarkers, identifying groups of differentially expressed genes with common functions can be an even more powerful way to identify molecular pathways. Analysis of pathways based on groups of genes is often referred to as gene ontology and is a commonly used approach to make sense out of seemingly diverse groups of genes (Khatri et al., 2007). Furthermore, molecular pathway biomarkers, such as a signaling pathway, can be more readily converted from a biomarker to a drug target.
Before determining whether or not a given biomarker is a good drug target, it needs to be validated. At a minimum, validation requires reproducing the same gene expression change seen in the microarray experiment using another method such as quantitative reverse-transcriptase polymerase chain reaction. Equally important, validation of predicted pathways can be achieved by demonstrating that the implicated pathway is indeed altered (increased/decreased) in tissues displaying variable epileptic activity. Based on the tissue sampling method shown in Fig. 2, we use portions of subdivided human tissues from each electrode location for validation studies. For example, if a particular signaling pathway is implicated, such as the activation of the transcription factor CREB (cyclic AMP response element binding) (Rakhade et al., 2005), it can be validated by showing changes in CREB phosphorylation using western blots of protein fractions and by tissue staining experiments that have the added advantage of implicating specific cell populations that might be involved in the epileptic process.
Once a specific biomarker is validated in human tissues, the next step is to determine whether it is a good drug target (see Fig. 3). Although human tissue studies can be invaluable for identifying genes and pathways, they cannot prove or disprove whether a particular gene or pathway is causative and hence a target for drug development. Animal studies are required for the next stage of validation to prove or disprove the value of a given biomarker as a drug target for epilepsy. The first question is how to choose an animal model from a large number of animal epileptic models, many of which use combinations of genetic, chemical, electrical, or injurious processes to elicit seizures (Schwartzkroin, 1993). For our purposes, we have found that a model using tetanus toxin in the neocortex replicates many of the same electrical and molecular changes we have seen in human neocortex (Barkmeier & Loeb, 2009). Even after a single model is chosen and found to be useful, validation with additional animal models will be needed to support a particular drug target. Finally, a means to test the drug target with a small molecule inhibitor or through genetic manipulations of the animal is truly the critical step required to validate a particular drug target. The ultimate goal is to demonstrate that disrupting a particular gene or pathway is sufficient to reduce or eliminate seizures in animals so that therapeutics can be designed for clinical trials in humans.
This work was funded by NIH/NINDS R01NS045207 and R01-NS058802 and a Wayne State University President’s Research Enhancement Program in computational biology.
The author has no conflicts of interest to disclose.