Multielectrode Recordings in the Somatosensory System

A fundamental goal in systems neuroscience is to explain animal behavior in terms of the dynamics of neural ensembles. Multielectrode techniques greatly facilitate the approach toward this goal. Aside from the fact that each experiment provides a higher yield of data as compared to single-site recordings, some questions simply cannot be addressed using only one electrode at a time. For example, only multisite recordings can determine whether different neurons respond independently to stimuli, or covary from trial to trial. The purpose of this chapter is to review methods used in multielectrode studies of the rat somatosensory system, with an emphasis on the whisker system. We present a basic toolbox of methods we have used to probe the functions of populations of somatosensory neurons in a behavioral context. The basic toolbox includes techniques for applying controlled whisker stimuli, behavioral training in tactile discrimination tasks, multielectrode recordings, reversibly inactivating specific brain areas, and analysis of the ensemble neural data. These methods have already revealed fundamental properties of the somatosensory system that would have been difficult or impossible to uncover using single-electrode recordings. For example, cortical (Zhu and Connors, 1999; Ghazanfar and Nicolelis, 2001; Diamond et al., 1992; Ghazanfar et al., 2000; Schubert et al., 2001) and thalamic (Armstrong-James and Fox, 1987; Nicolelis and Chapin, 1994) neurons have large multiwhisker receptive fields that are dynamic over poststimulus time (Nicolelis and Chapin, 1994; Ghazanfar and Nicolelis, 1999; Ghazanfar et al., 2000). These data, together with observations of supralinear summation of multiwhisker inputs (Ghazanfar and Nicolelis, 1997; Shimegi, 2000), suggest that tactile receptive field dynamics function to integrate time-varying multiwhisker inputs (Ghazanfar and Nicolelis, 2001). For example, analysis of multineuron response data revealed additional stimulus-coding properties of somatosensory ensembles. S1 ensembles code stimulus location in single-neuron temporal patterns and the relative response latencies of their neurons, but not in single-trial covariations among the neurons (Nicolelis et al., 1998; Ghazanfar et al., 2000). In S2 of the primate, on the other hand, single-trial covariations among multiple neurons did contribute significantly to coding the location of a punctate stimulus (Nicolelis et al., 1998; Ghazanfar et al., 2000). Even in S1, the contribution of coordinated firing may increase with greater stimulus complexity, because multiple whisker stimuli lead to a higher prevalence of synchronous responses between neurons in the infragranular layers of S1 than in other layers (Zhang and Alloway, 2005). Combining methods for inactivating specific neural inputs with ensemble recordings led to the further conclusion that spatiotemporal RF properties of somatosensory neurons arise not only from intrinsic local properties of neurons and their neighbor connections, but rather from interactions among multiple levels of the somatosensory system. For example, recording thalamic tactile responses in the presence and absence of cortical feedback revealed that corticofugal projections contributed to both the short- and long-latency components of ventral posterior medial nucleus (VPM) responses (Krupa et al., 1999; Ghazanfar et al., 2001). These interlevel interactions were reflected in simultaneous recordings in trigeminal areas in brain stem, thalamus, and cortex, which revealed widespread oscillatory synchronization of neural firing (Nicolelis et al., 1995). The correlated activity remains even after transection of the facial nerve, which suggests that such synchronous activity is generated centrally. Although the high coherence among large populations of neurons associated with this oscillatory 7–12 Hz brain state suggested absence seizures to a number of authors (Marescaux et al., 1992; Shaw et al., 2006; Shaw, 2007), a direct test showed that rats respond robustly to mild tactile stimulation during bouts of 7–12 Hz oscillations in S1, contradicting the absence interpretation (Wiest and Nicolelis, 2003). Thus, widespread synchronized neural firing need not preclude perception; in fact, it can enhance aspects of sensory representation (Fontanini and Katz, 2006) as well as long-term plasticity (Erchova and Diamond, 2004). These demonstrations of fast interactions among neurons distributed across the somatosensory maps at multiple processing stages were paralleled by demonstrations of a tight coupling between the two hemispheres of S1 (Shuler et al., 2001; Wiest et al., 2005). This cross-talk challenges the classical conception of the S1 barrel cortex as an encoder for exclusively contralateral whisker activity and suggests the potential importance of bilateral interactions in S1 for whisker-guided discriminations (Krupa, 2001b; Shuler et al., 2002). Multielectrode recordings in different layers of S1 while rats performed a bilateral whisker-guided discrimination revealed that a feed-forward model of tactile signal processing cannot explain S1 response properties (Krupa et al., 2004). For example, firing rate modulations that began before tactile stimulation clearly could not be explained in terms of bottom-up propagation of a stimulus signal. Rather, other inputs to S1 must contribute to shaping the task-related responses. Similarly, tactile responses were found to vary significantly in different spontaneously occurring behavioral states (Nelson, 1996; Fanselow and Nicolelis, 1999; Moore et al., 1999; Nicolelis and Fanselow, 2002; Castro-Alamancos, 2004; Moore, 2004). These data collectively suggest that widely distributed neurons coordinate their activities on millisecond time scales, and that the functional connectivity among them can be quickly adjusted in different behavioral contexts. The preceding examples are meant to indicate the range of results that have already been achieved using multielectrode arrays (MEAs). In the following sections we present specific methods developed in the past 15 years. The examples have been selected to represent methods from each major phase of a typical study, from electrode design and surgical implantation, through ensemble recording involving somatosensory stimulation, behavioral monitoring, and reversible inactivation of specific brain areas, to analysis of the recorded many-neuron data.