Recording From Over 1,000 Cells: A New Toy in Place for Epilepsy Research?

Commentary

Long-Term Dynamics of CA Hippocampal Place Codes.

Ziv Y, Burns LD, Cocker ED, Hamel EO, Ghosh KK, Kitch LJ, Gamal AE, Schnitzer MJ. Nat Neurosci 2013;16(3):264–268

Using Ca²⁺ imaging in freely behaving mice that repeatedly explored a familiar environment, we tracked thousands of CA1 pyramidal cells' place fields over weeks. Place coding was dynamic, as each day the ensemble representation of this environment involved a unique subset of cells. However, cells in the ~15–25% overlap between any two of these subsets retained the same place fields, which sufficed to preserve an accurate spatial representation across weeks.

Determining how seizures originate at the individual cell and network levels is beyond our current technology. Our present definition of a seizure is based upon the dynamics visible with 60-year-old technology. We view 10 to 15 seconds of data on a “page” of computer screen, and the fastest events occur at 30 Hz. The data come from large electrodes that record the combination of local field potentials from millions of cells—an effect we still do not fully understand (1). This low-resolution view of brain activity has been useful clinically but does little to help us understand the mechanisms of seizures. Some recent technologic advances have focused on improving this resolution, and researchers have found intriguing dynamic phenomena such as spiral waves with high-density electrodes (2) or unique firing properties in epileptiform discharges using faster calcium imaging (3). Several other recent advances are challenging the standard teaching about seizures, such as whether seizures really represent synchronous neuronal activity and whether pyramidal cells are the instigators of seizures (4). We still have a long way to go before understanding the intricate network dynamics that produce seizures.

A primary reason we understand so little about seizure generation is that our technology has not been able to see what happens on the network level. The main technologic challenge in measuring seizure dynamics is dealing with the difficult balance between spatial and temporal resolution. For years, this balance has led to a kind of neurophysiological Heisenberg uncertainty: one can achieve either temporal or spatial resolution, but never both. Single electrodes have very good temporal resolution but are typically used to record from either a large volume of brain (EEG) or from a tiny sample of cells (patch clamp, tetrodes). The recent Utah microarray improves resolution to 100 electrodes within a 4-mm square (5), the scale of a small number individual cortical columns. Imaging can provide a wide range of excellent spatial resolution at specific instants in time, from MRI of the entire brain to high-resolution microscopy of individual cells. The recent developments in fMRI, voltage-sensitive dyes, and calcium imaging now provide a temporal dimension to imaging, but they have traditionally been too slow to resolve individual action potentials. However, in recent years, investigators have been pushing the envelope from both sides, creating large arrays of small electrodes (2) and developing methods to identify individual cells firing in calcium imaging (3). New technology is now poised to monitor thousands of individual cells to finally determine how neuronal networks work.

Ziv and colleagues have developed a remarkable tool that allows them to monitor the firing of over 1,000 individual cells over the course of weeks in awake, behaving animals. They used a viral vector to induce expression of GCaMP3 within hippocampal pyramidal cells, then implanted a microscope capable of imaging the calcium activity in those cells. Their research did not involve epilepsy; rather, they examined the dynamics of place-cell firing. As mice explore an area, certain pyramidal cells tend to fire whenever they reach a similar “place” in the track. Place cells have been known for many years to encode spatial information but, as in seizures, their temporal dynamics have been difficult to determine (6). The primary biological outcome of this study was that, while a small percentage of cells tended to “remember their place,” most of the cells changed their spatial coding over time. This outcome in normal mice is intriguing and will likely impact future work on epilepsy and memory, as past work has shown that seizures disturb normal place cell firing (7). Additionally, development of the method itself is a dramatic technological step forward. The authors were able to label, identify, and follow an enormous number of pyramidal cells and determine the firing order of each one during normal behavior. Their microscope was small enough to allow the mouse to ambulate without difficulty, and they demonstrated that the imaging continued to work after ten sessions over the course of weeks. It is not hard to imagine how similar techniques could be a useful tool in epilepsy research.

One of the most important aspects of this work is the method in which the data are handled. Processing data from over 1,000 cells is no trivial task, and one of the caveats of acquiring such high-resolution data is to know what to do with it. Ziv and colleagues used an algorithm that they published previously (8) to coregister spatial locations with fluorescence images, using a combination of principal components analysis, independent components analysis, and image segmentation to identify individual cells. This method is especially powerful because it labels and tracks somata as well as dendrite processes. Because the microscopes were removed for several days between experiments, the authors also developed a method to identify and coregister cells between sessions, adjusting the microscope's trajectory to match previous days. This painstaking approach allowed them to determine which specific cells fired at the same place as before.

While this technology has obvious implications for epilepsy research, there are several issues that must be addressed. First, investigators must be prepared to handle and correlate the massive amounts of data that come with such high resolution. In this work, as in the electrode and imaging articles mentioned previously (2, 3), advanced mathematical and statistical tools were necessary to make sense of the data. Second, this technique computationally extracts individual cells by identifying short spike bursts in networks with rather sparse firing. It will be difficult to translate those conditions to epileptiform discharges, in which neither of those conditions may be present. Third, GCaMP3 activates quickly when a cell starts to fire, but has slow “off” kinetics, taking several seconds to return to baseline; this would be very limiting when recording seizures. Fourth, while this study demonstrated that the imaging integrity was maintained after weeks with intermittent imaging, it did not test what would happen if the microscope was left on for prolonged periods, as would be necessary to monitor for seizures. Despite these caveats, this technology may prove applicable to epilepsy research. The authors found that 75% of cells change their firing patterns every 5 days. Is this type of plasticity only present in normal physiology? If so, could the stereotypy of electroclinical seizures be a manifestation of an aberrant deterministic pathway similar to that seen in brain slices (3)? Or, will there also be variability in epileptic tissue, potentially confounding attempts to identify a “focus”? Either way, epilepsy now has a new toy to play with, and it will be fascinating to see what it uncovers.