An Integrative Review of Data and Theoretical Perspectives

Regarding Brain Function in the Vibrissal System

Since 1912, the vibrissae have been recognized as an important source of sensory input for the rat and other rodents (Vincent, 1912). The extreme mobility and sensitivity of the vibrissae in rats most likely developed as a partial compensation for poor vision and from the demands of their nocturnal environment. Rats can extend their whiskers as far as two inches (5cm) in front of them in order to sense objects. These constantly moving facial whiskers provide the rat with guiding sensations, a sense of support, aid in locomotion, and are intimately connected with equilibrium (Vincent, 1912). Carvell and Simons (1990) trained rats to discriminate between a smooth surface and a rough surface having shallow (30 mm) grooves spaced at 90 mm intervals, using only vibrissal cues. Their study demonstrated that by actively using their vibrissae, rats are able to distinguish between differently textured surfaces at a level that is comparable to that of primates using their fingertips.

The vibrissal system of the rat is an interesting system to explore. When the rat whisks an object, signals travel by way of the infraorbital branch of the fifth cranial nerve from the receptor cells in the whisker follicle to the trigeminal brainstem complex, and then to the thalamus and the primary somatosensory cortex. The spatial arrangement of the whiskers on the rat’s face is one of a matrix of large hairs (5 rows x 5-9 columns) which is represented in these brain areas by a topographically similar matrix of cell rings. This spatial arrangement on the face and in the brain is illustrated in Figures 1A and 1B, respectively. The aggregates of cell rings in layer IV of the cerebral cortex are referred to as barrels. This anatomically distinct area in the somatosensory cortex shown in Figure 1C, has been termed barrel cortex because the neurons are

Figure 1. The spatial arrangement of the whiskers on a rat’s face is illustrated

above (A), as well as the corresponding matrix of cell rings in the somatosensory

cortex (B). Actual barrels from layer IV are shown as well (C). (From Blakemore, 1977)

grouped in barrel- like arrangements, with a hollow center of lesser cell density surrounded by a circle of higher cell density, and because they appear as a stack of barrels when viewed from one end (Woolsey and Van der Loos, 1970). Based on anatomical and physiological evidence from histological and electrophysiological studies, Woolsey and Van der Loos (1970) concluded that a one- to- one relationship exists between each vibrissa and its corresponding barrel. Thus, each barrel was interpreted to primarily represent one contralateral mystacial vibrissa. In order to identify each whisker and its corresponding barrel, a naming system has been developed. The rat’s mystacial vibrissae are found on the upper lip arranged in rows on either side of the nasal fossae (Vincent, 1913). These whiskers are arranged in five horizontal rows (A, B, C, D, E) which are lettered dorsal to ventral and five to nine (depending on the row) columns which are numbered caudal to rostral beginning with the number 1.

Because of the functional and morphological correlation between the vibrissae and the barrels, the vibrissae-barrel neuraxis is an attractive model for studying structure, function, development and plasticity within the somatosensory system. However, the significance of this anatomy is uncertain because comparative studies have shown that cytoarchitectonically distinct barrels exist in only a few of the mammalian species that have prominent mystacial vibrissae (Woolsey, Welker, and Schwartz, 1975; Rice, Gomez, Barstow, Burnet, and Sands, 1985; Waite, Marlotte, and Mark, 1991). Barrels have been observed among marsupials, rodents and lagomorphs, but not among other orders such as carnivores and primates. Because the extent to which an animal relies on its whiskers is not known, it is not possible to predict the presence of histologically distinct cortical barrels. Two theories have attempted to explain why barrels exist in only certain species. The first proposes that barrels are a unique cortical specialization for vibrissae sensation (Woolsey and Van der Loos, 1970). According to this view, it is possible that the whiskers of rats, mice and similar rodents are more important in gathering information than they are for other mammals (with or without whiskers). Because they rely more on their whiskers, the corresponding cortex of these animals is more developed. The second theory suggests that barrels may represent an exaggeration of a fundamental somatosensory cortical organization that exists in all mammals (Woolsey and Van der Loos, 1970; Chapin, Waterhouse, and Woodward, 1987). According to this theory, barrels exist in all mammals, but appear to be more pronounced in rodents because they have a marginally low number of cells in layer IV. Because larger mammals have a much greater number of cells in layer IV, the basic pattern is obscured. Although these theories might eventually lead to a more complete understanding of why barrels exist, the significance of barrels has yet to be determined.

Anatomy of the Vibrissal System

Prior to investigating how sensory information is processed and transformed at each neuronal level between the periphery and the barrel cortex, it is important to outline the structural properties of the vibrissal system. The connections discussed in this section are illustrated in Figure 2. First, whisker- sensitive receptor cells synapse with the infraorbital branch of the trigeminal (V) nerve in order to relay signals to the trigeminal brainstem nuclear complex (TBNC). The infraorbital nerve bifurcates into ascending and descending limbs which innervate, respectively, the trigeminal nucleus principalis (PrV), and the three subnuclei of the spinal trigeminal nucleus: oralis (SpVo), interpolaris (SpVi) and caudalis (SpVc) (Diamond, 1995). The histological arrangements of cells in PrV are called barrelettes. Each barrelette contains neurons that respond to only one “principle” whisker (Ma, 1991). Histologically distinct barrelettes are not found in caudalis, interpolaris, or oralis.

Axons from the TBNC then project to portions of the contralateral thalamus, specifically the ventral posterior medial nucleus (VPm) and the medial portion of the posterior thalamic nucleus (POm). These trigeminothalamic projections to the VPm terminate in whisker- related cell clusters referred to as barreloids (Van der Loos, 1976). The VPm projects a vast majority of its axons to the barrels in layer IV of the somatosensory cortex, and only a minimal number of VPm axons extend to more than one barrel or barrel-column. Thus, the topographic barreloid-to-cortical barrel representation is preserved. The VPm is deemed the primary somatosensory pathway and can relay signals from the vibrissae to the barrel field within 8 msec. A corresponding specificity is not maintained, however, in the POm which relays signals from several vibrissae to the septal regions between the barrels. Cells in the POm, also known as the secondary somatosensory pathway, respond to stimulation with a latency of 15-20 msec after whisker stimulation and most likely serves as part of an inter-barrel communication pathway (Diamond, 1995).

Activity in barrel cortex, is projected to other areas of the brain; barrel cortex has several cortical and subcortical connections. Reciprocal ipsilateral and contralateral corticocortical connections include projections to the primary motor cortex (MsI), the secondary somatosensory area (SII) and the perirhinal cortex. The ipsilateral connections originate in layers III and IV and extend to layers III, IV and V of the adjacent barrel cortex. The contralateral connections extend from the barrel cortex on one side through the corpus callosum to the barrel field, SII and the perirhinal cortex on the other side (Diamond, 1995).

Several corticosubcortical connections have been identified as well. A nonreciprocal connection exists to contralateral and ipsilateral striatum (Diamond, 1995). Reciprocal projections have also been found between the vibrissal area of MsI and the nucleus ventralis lateralis (VL), the nucleus centralis lateralis (CL) and the zona incerta (ZI) of the ipsilateral thalamus. In addition to the connections with MsI, VL and CL also receive signals from the cerebellum and globus pallidus, which are subcortical motor-related structures. Nonreciprocal corticothalamic projections are evident in the thalamic

reticular nucleus (NRT) (Porter and White, 1983).

Follicle Brainstem Thalamus Cortex

Figure 2. A functional diagram, illustrating major projections in the vibrissal system.

Behavioral Functions of the Barrel Cortex

In order to have a more complete understanding of the vibrissal system, it is important to understand the functional properties of the system as well as the structural anatomy. Three experiments involving whisker removal and barrel ablation were conducted by Hutson and Masterton (1986) in order to investigate the role of barrel cortex. In the first experiment, the role of a barrel in the detection of movement of its corresponding vibrissa was assessed using a conditioned suppression technique. All but one (C1) of the whiskers from the left and right side of the rat’s face were removed and either contralateral or bilateral barrelfield ablations of the somatosensory cortex were done. Next, the rats were presented with an airstream paired with shock in order to induce conditioned suppression of drinking. If the rats detected the movement of air, then they would refrain from drinking. Results showed that rats' detection thresholds for sinusoidal oscillating airstreams with the contralateral vibrissa remained unchanged even after the contralateral ablations of the barrel cortex. These findings indicate that the barrel cortex is not necessary for rats to detect vibrissal movement.

In the second experiment, the role of cortical barrels in the discrimination of the frequency of vibrissal oscillation was evaluated using a similar conditioned suppression technique. The rats were either presented with a constant oscillation which was used as a safe signal, or a higher frequency oscillation which signaled shock. After contralateral and bilateral ablations, the rats were still able to discriminate between the two frequencies. The results indicate that the barrel cortex is not necessary for rats to detect differences in oscillation frequencies.

In the final experiment, the role of the barrelfield cortex in higher-order discriminations was assessed using a gap-crossing task. Rats were initially trained to either jump or not jump, depending on whether they could reach across a gap in an elevated runway for a food reward. In order to test the usefulness of the vibrissal system for this task, the ability of the rats to perform the task was measured after each of nine sequences of vibrissal removal or barrelfield ablations that gradually reduced the vibrissae- barrel compliments until none remained. Results showed that rats need at least one barrelfield and its contralateral whisker in order to successfully perform the discrimination task involving active exploration.

Guic-Robles, Valdivieso, and Guajardo (1989) extended the work of Hutson and Masterton (1986) by investigating the instrumental discriminatory characteristics of the vibrissal system. They demonstrated that rats can differentiate between two sandpaper surfaces with different degrees of roughness by only using their vibrissal system. The rats were trained to discriminate between a smooth surface (200 grains/cm²) that was associated with reinforcement and a rough surface (25 grains/cm²) that was not associated with reinforcement. After the rats reached an 85% criterion level of performance, the vibrissae were bilaterally trimmed. Performance then dropped to chance levels (around 50%). However, once the whiskers regrew, rats were again able to perform the discrimination task at the former criterion level. The results show that rats rely on their vibrissal system and not on extra-vibrissal cues in order to solve this behavioral problem.

A further investigation of the functional properties of the vibrissal system was conducted by Guic-Robles, Jenkins, and Bravo (1992). They showed that performance on a roughness discrimination task is dependent on the barrel cortex. Rats were initially trained to perform the discrimination task at a criterion level of 85%. However, after bilateral lesions to the posterior medial barrel subfield (PMBSF), the rats failed to exhibit any evidence of task retention. The rats were also unable to relearn the task after the lesion.

Together, these studies have helped demonstrate the functional properties of barrel cortex. According to Hutson and Masterton (1989) barrel cortex is not involved in the detection of vibrissae movement or in the detection of differences in oscillation frequencies. Instead, it has been shown that barrel cortex is involved in types of discrimination tasks that involve active movement and actual decision-making on the part of the animal.

Distributed Versus Labeled Line Processing

Through the 1960s and 1970s, most researchers accepted a “labeled line” view of sensory processing. According to this view, the functional properties of sensory representations established in early postnatal life are relatively static and enduring. For example, Lettvin, Maturana, McCulloch, and Pitts (1959) interpreted the frog’s brain as containing moving edge or dot detectors. Likewise, Woolsey and Van der Loos (1970) proposed a one-to-one correspondence between each whisker and its corresponding barrel. Neurons within the primary somatosensory cortex have been thought to preserve the spatial and temporal stimulation characteristics of their corresponding principal whisker (Simons,1978, 1985). Neurophysiological research has shown that many of these neurons have single-whisker receptive fields and distinct response properties. Receptive field sizes are smallest in layer IV, where the majority of cells are mainly activated by their principal whisker, and largest in layers V and VI where most cells respond to deflections of several adjacent whiskers. It is also the case that cortical vibrissa units in layer III respond differentially to a variety of stimulus parameters such as frequency, angular direction, velocity, and amplitude (Simons 1985). The labeled line approach to sensory processing views neural integration as a simple integration of individual features encoded by individual neurons. Such integration is accomplished through spatial and temporal summation of neural inputs within the sensory path.

For example, the various response properties of cells at different stages of the primary vibrissal pathway have been demonstrated through the use of controlled whisker stimulation with extracellular single-unit recordings. Reported differences exist between thalamocortical units in the thalamic barreloids of the VPm (TCUs), “regular-spike” neurons (RSUs) which are most prevalent throughout cortical layers II-VI, and “fast-spike” neurons (FSUs) which are mainly restricted to cortical layer IV. Five major distinctions between the properties of TCUs and RSUs have been made by Simons and Carvell (1989). First, TCUs displayed higher rates of spontaneous activity, and responded more vigorously to whisker stimulation. Second, cells in the thalamus tended to respond to a larger number of whiskers, thus having larger receptive fields than RSUs. Third, twice as many TCUs as RSUs respond selectively to angular direction of whisker displacement. Fourth, responses of slowly adapting TCUs were shown to be ~3.5 times greater than those of slowly adapting RSUs. Lastly, cross-whisker inhibition was observed less frequently in the thalamus. Similar research by Lichtenstein, Carvell, and Simons (1990) has supported these findings and concluded that TCUs respond with higher rates of firing to whisker stimuli, are more selective for the angular direction of whisker movement, and are more likely to respond in a slowly adapting fashion when compared to RSUs.

Distinctions have also been made between the response properties of RSUs and FSUs, both of which are found in the cortical barrels. Research has shown that FSUs tend to respond more similarly to TCUs. They display exceptionally high levels of spontaneous activity at rates of 15-50/s, while RSUs discharge spontaneously at rates of <1-15/s. FSUs also have larger receptive fields, respond more reliably to sinusoidal oscillations (possibly representing texture), and respond over a wider range of frequencies (3-40 HZ) than do RSUs. In addition, FSUs respond to whisker deflections over a broad range of angles as compared to RSUs, which have distinct spatial tuning characteristics and tend to respond to whisker deflections over a restricted range of < 90° (Simons, 1989). Collectively, these physiological data have been interpreted as providing support for an isomorphic structure-function view of sensory processing (i.e., a “labeled line”), in which each barrel is the morphological correlate in layer IV of a functional cortical column that extends throughout the thickness of the cortex, and encodes combinations of particular stimulus features. Further, each RSU in a barrel corresponds to the same principle whisker, but responds to a more selective range within a single stimulus dimension.

Now let us examine how the structure and/or function of the vibrissal system could be disrupted. Klein, Renehan, Jacquin, and Rhoades (1988) examined the effects of neonatal infraorbital nerve transection on the development of trigeminal patterns in the adult rat. Ipsilateral connectivity was preserved in order to provide a control; therefore, transections were unilateral. As expected, they found that unilateral infraorbital nerve transection in the neonate caused structural abnormalities in the trigeminal ganglion and vibrissa follicle nerves in the adult rat. They found that the myelinated follicle nerve axons were significantly fewer in number as well as smaller in diameter. They also found that while there was a significant reduction in the number of trigeminal ganglion cells innervating the vibrissa follicles, the peripheral branching among these ganglion cell axons was more pronounced. Finally, they observed that the somatotopic arrangement within the trigeminal ganglion was also altered.

While Klein et al.(1988) demonstrated how neonatal infraorbital nerve transection disrupts certain structural formations in the adult rat, Henderson, Woolsey, and Jacquin (1992) investigated the role of early neural activity in the development of the structure. They showed that simply blocking the activity of the infraorbital nerve does not affect central trigeminal pattern formation. In order to block signals transmitted via the infraorbital nerve, tetrodotoxin (a temporary sodium channel blocker) was injected in apposition to the left infraorbital nerve. They found that the average number of trigeminal ganglion cells was the same as controls. They also observed normal pattern formations in the trigeminal brainstem complex, thalamus, and barrel cortex. The results of this study demonstrate that actual activity in the infraorbital nerve is not necessary for the development of normal microscopic structure in the trigeminal complex.

While the previous two studies focused on structural development, Klein (1991) examined the functional properties of the trigeminal system. As a result of infraorbital nerve transections on adult rats, he found significant changes in size and functional reorganization of receptive fields for cells within SpVi. These results, with those of Klein et al. (1988), demonstrate that extensive functional reorganization can occur even without disruptions in microscopic structure. Thus, there is evidence that structure need not imply a particular mode of function, nor does one have to assume that altered function implies altered microscopic structure.

Other neurophysiological studies of the rat vibrissal system have also produced evidence alternative to the labeled line view implied by a traditional structure-function interpretation. In contrast to the findings of Simons (1978, 1985, 1989), Nicolelis, Lin, Woodward, and Chapin (1993) have demonstrated that neuronal receptive fields and topographic maps are dynamic and distributed. The plasticity of the receptive fields of the ventral posterior medial thalamus (VPm) was demonstrated during both awake and anesthetized experimental conditions. Researchers used a 64 microelectrode array in order to record the extracellular activity of populations of units after computer-controlled deflections of the rat’s whiskers. Responses to individual electrodes were measured using post-stimulus time histograms (PSTHs). PSTHs were used to represent the time of each recorded spike in relation to the onset times of computer controlled deflections. Three dimensional plots were constructed in order to demonstrate neuronal responses to whisker stimulation as a function of post-stimulus time. When the spatiotemporal receptive fields of approximately 23 neurons in the VPm were simultaneously monitored, they were found to be large and overlapping, covering up to 20 whiskers. The unit responses within these receptive fields were also found to shift dramatically over the first 35ms of post-stimulus time. Shifting was especially evident from the caudal-most to the rostral-most units.

These results contest the labeled line view of processing by proposing that the VPm contains a dynamic and distributed representation of the facial whiskers. If the distributive processing view holds true, it will have profound implications on how the somatosensory system actually works. It would mean that the brain would be able to reorganize sensory maps following changes in sensory experience or external injury (Nicolelis, 1997). It may also imply that the traditional conceptualization of the structure-function relationship, in which one neuron encodes one feature from one whisker, in the vibrissal system needs to be changed.

Pribram, Spinelli, and Kamback (1967) demonstrated how sensory experience could cause for changes in patterns of activity in primary sensory cortex. First, rhesus monkeys were trained to perform a discrimination task. They had to pull a lever in order to turn on a display, lasting 0.01msec, which was followed by either vertical stripes or a circle. If the stripes appeared, the monkeys were rewarded with a food pellet for depressing the left half of the display panel. If the circle appeared, the monkeys were rewarded for depressing the right half. Once the rhesus monkeys reached criterion, the researchers used a twelve electrode setup to record electrical potentials from the striate cortex of the monkeys during three stages of the visual discrimination trial. A different input pattern of brain activity was recorded for each lever that the monkey pressed: left or right. Another pattern of brain activity was correlated with each stimulus and responded to the stripes or the circle. Finally, a pattern of brain activity was correlated with whether the monkey was rewarded or not. The researchers found that prior to learning the discrimination task, there were not significant differences among patterns of brain activity recorded during the three stages of the trial. However, after learning occurred, the three types of brain activity were identifiable due to their unique patterns, which were distributed over the entire striate cortex. Therefore, sensory experience alone did not account for the different patterns of activity, but learning did.

Bridgeman (1982) performed a similar experiment which also showed how sensory experience could account for changes in patterns of brain activity. Macaque monkeys were trained to perform a two-part visual discrimination task. First each monkey was to look at a fixation light when it was illuminated. If they performed correctly, the first task was rewarded by presenting the second task, in which the monkey was to press a panel under the brighter of two illuminated discs. If the second task was performed correctly, the monkey then received a food reward. While the monkeys performed this visual discrimination task, electrode recordings were taken from groups of neurons in the parafoveal striate cortex. Bridgeman (1982) observed that prior to learning the discrimination task, cellular responses were similar in both the correct and incorrect trials. However, after learning occurred, groups of cells exhibited enhanced responses when the monkeys were about to make correct responses. These results demonstrate that responses among populations of neurons in the striate cortex are plastic and subject to change as a result of learning.

Stallings (1998) extended this line of research by recording the responses of multiple units in barrel cortex from awake animals during a discrimination task. The rats were trained to whisk two disks (one smooth and one rough) and then decide which one of the two corresponding goal boxes contained a food reward. Results were evaluated as frequency histograms of activity within a small (4-6 unit) neural ensemble, which showed that the pattern of cell activity within an ensemble was dependent upon combinations of stimulus as well as behavioral reinforcement parameters. Individual units did not just respond differentially to the features of the stimulus, they also responded to the "meaning" of the stimulus. When the food reward was changed from the smooth to the rough disks, the cortical representation of the stimulus also changed. These findings illustrate the plasticity of cell properties in support of a distributed view of processing in which populations of cells represent not only stimulus input, but behavioral correlates of such input.

Research conducted by Simons (1985; 1989) and Nicolelis (1993; 1997) has most likely produced different results in support of different perspectives due to different methodological approaches. Simons has proposed a structure-function view of sensory processing in which sensory representations are fixed. Thus, as features are encoded information becomes more specific and receptive field size becomes smaller, at least up to the input layer of the somatosensory cortex. In order to obtain these results, Simons used few electrodes in specific locations. By contrast, using a large array of electrodes, Nicolelis found results that imply a distributed view of processing in which sensory representations are widespread and plastic. Questions remain as to which, if either, view of sensory processing is the better.

Scales of Processing in Neural Systems

It is important to focus on why the interpretations of these researchers are at odds with one another. Perhaps Simons (1985,1989) and Nicolelis (1993,1997) hold opposing views because they are focusing on different scales of analysis, or on different types of processing within a given stage of analysis. Thus, it is important to make a distinction between two scales of processing. Secondly, it is also important to consider the nature of neural computation that takes place. One scale, a macroscale, focuses on the functional brain system in which there are anatomically distinct stages of processing. In the vibrissal system, for example, there are particular stages of processing that input must flow through. Whisker displacement results in stimulation of many receptor cells lining the whisker follicle. The processing of input from the receptor cells first begins in the trigeminal brainstem nuclear complex (TBNC). Signal processing then continues as input flows from various areas in the contralateral thalamus and proceeds to the primary somatosensory cortex (barrel cortex). When researchers study neural systems at this scale, they are concerned with what is being processed: i.e., the changes or transformations that take place between input and output at each stage. At each stage of processing, different types or sources of signals must be integrated or modified. For example, the TBNC consists of two parts: the trigeminal nucleus principalis and the spinal trigeminal nuclei. The output from these structures is integrated by the posterior thalamic nucleus (POm) and the ventral posterior medial thalamus (VPm) in the contralateral thalamus. Input from other systems, such as the primary motor cortex, is also integrated with information from the contralateral thalamus within the primary somatosensory cortex.

A second scale, a microscale, is concerned with the nature of processing within each stage of a functional system. Two different views that attempt to account for such processing are the classical receptive field (labeled line) view and a distributed view. According to the classical receptive field view, the focus should be on whether or not a given neuron fires in response to a particular stimulus. From the receptive field view, early in a functional system each neuron receives input from a collection of receptors, and processes that input in terms of an individual feature of the stimulus. As processing proceeds through the functional system, integration of features is accomplished through a convergence of neural inputs by way of temporal and spatial summation. Thus, neurons at a later stage of processing respond to more combinations of specific individual features.

There are at least three different distributed views. According to two of these distributed views of processing, population vectors and neural ensembles, the focus should not be on whether or not a particular neuron fires, but on the pattern of firing among a group of neurons. Population vectors describe the results of particular patterns of firing among a group of neurons from a computation in space-time. The changes in this pattern of firing over a spatially distributed group of neurons and time reflect the current integration of inputs at that particular stage. For example, Georgopoulos, Caminiti, Kalaska, and Massey (1983) measured the responses of individual units in the motor cortex of monkeys while they performed two-dimensional arm movements on a plane working surface. They found that although individual units in the motor cortex possessed directional preferences, they also responded in a lower firing rate to other directions of movement. The research indicated that a population of cells is tuned to several overlapping directions of movement. They used the vector hypothesis in order to represent the responses of single units. According to this hypothesis, the preferred orientation (direction) of a population of neurons could be determined by plotting vectors that represent the orientation of the movement and the rate of cell responsiveness.

The neural ensemble approach (Nicolelis et al., 1993; Nicolelis, 1997) relies on multiple electrode arrays to provide a pattern of activity among individual, neighboring units that changes as the input changes. For example, Nicolelis et al. (1993) used a 64 microelectrode array to record responses from populations of units in the ventral posterior medial thalamus in relation to whisker flicking. They found that the population responses changed as a function of space and time. The spatial locations of the receptive fields also shifted over time.

The holonomic approach (Pribram, 1991) focuses on how processing is done in a different manner. This approach accounts for processing in space and time, while adding a spectral component to local neural processing. The theory is based on both Fourier and Gabor relationships. The Fourier transform, which is explained later in greater detail, accounts for reversible transformations of spectrum to time and space. The Gabor function, however, is able to account for transformations within both spectrum and space-time. According to the holonomic approach, the dendrite network is the relevant structure-function unit. The distributed pattern of voltages within a synaptodendritic network is therefore the important consideration. Individual units sample from small sections of the dendritic field. Pairs of units are responsive to particular orientations (directions) and spatial frequencies within the portion of the dendritic web from which they sample. A given frequency pair responds as sine/cosine pairs if that spatial frequency is present in their portion of the receptive field. Thus, neural output from the integration of inputs to a synaptodendritic web reflects a type of spectral analysis which includes amplitude, frequency, and phase. Notice that according to the holonomic view, the receptive field on a receptive surface extends to the dendritic network to which it projects.

The purpose of this thesis is to attempt an integration of different viewpoints and methodologies in the study of cortical integration. In particular, I will focus on the utility of the spectral (holonomic) approach as it may apply to the visual and vibrissal systems. I will use the findings of research on the visual system as a model for interpreting results of experiments with the vibrissal system. Not only has more research been done on the visual system, but also, more developed theories have been formed as to how information is processed within the visual system. Although more work has been done in the visual system, the vibrissal system of the rat might also prove to be a good system to work with. This is because a unique topographic relationship exists between the whiskers on the face and the barrels in the primary somatosensory cortex. This architecture provides an ideal model for studying the processing of multiple inputs as they progress through the stages of processing.

Representations of Population Codes in Barrel Cortex

The response properties of neuronal populations have been represented in various ways. King, Xie, Zheng, and Pribram (1994) measured the responses of single units in the barrel cortex of 25 rats to combinations of spatial and temporal frequencies. Three cylinders with varying groove sizes were rotated against the rats’ vibrissae at eight different speeds. They used surface distributions to represent the distribution of dendritic field potentials. The surface distributions consisted of three dimensions: the spatial frequency of the stimulus, the temporal frequency of the stimulus, and the rate of response of the cell or group of cells. They found that the surface distributions were not similar with regard to spatial or temporal frequency. Thus, different cellular responses were obtained even when the flick rate would have been predicted to be the same. A cylinder with a low spatial frequency spinning fast did not produce the same pattern of cellular responses as a cylinder with a high spatial frequency rotating slowly. Thus, their results indicated that populations of neurons encode vibrissal activity in the spectral domain.

Santa Maria (1995) also measured the responses of cells in the barrel cortex to

combinations of spatial and temporal frequencies as buccal nerve stimulation caused sweeping of the mystacial whiskers. Five grooved disks, each with a different number of equally spaced grooves and teeth, were used to manipulate spatial frequency. Unit responses were recorded with disks placed at six different orientations from 180° and cell responses were measured at six different angular orientations. Hovis (1997) performed a similar experiment, but added a third variable- sweep rate. The whiskers were moved against the grooved disks at rates of 4 Hz, 8 Hz, and 12 Hz. Hovis’ (1997) study further indicated that the rat’s natural sweep rate of 8 Hz provided the greatest contrast between voltage patterns within a surface distribution.

Results from the two studies support the previous research by King, Xie, Zheng, and Pribram (1994). It was demonstrated that surface distributions provide an adequate method for representing the spike activity of small groups of neurons in the rat barrel cortex as recorded from a single electrode. However, surface distributions display a more detailed representation of the pattern of change for a cell’s response to changing texture, changing stimulation rate, and changing location in a three-dimensional view. In addition, the experiments illustrate that individual cell responses are not fixed, but vary according to combinations of different variables.

Fourier Representations of Complex Waveforms: Visual System

Late in the eighteenth century, Fourier developed the mathematics indicating that any complex pattern over space or time can be analyzed into different sine waves and their cosine counterparts varying in frequency, amplitude, and phase. In order to achieve this analysis, the waveform is decomposed into the linear sum of its sine and cosine components. These wave pairs are 90° (1/4 of a wavelength) out of phase with one another; therefore, the intersections of these waves at each frequency are represented mathematically as Fourier coefficients. From these coefficients one can derive each component frequency, its amplitude, and its phase.

Studying brain processes in terms of the spectral domain has proved to be useful in areas such as pattern recognition. The feature detector notion, proposed by Hubel and Wiesel (1962), that each cell responds only to one preferred edge or bar of a specific width has not been able to adequately explain how the visual system works. For example, this notion does not account for why it is possible to discriminate among different patterns of light and dark. The Fourier analysis, however, does account for such pattern recognition. Pollen, Lee, and Taylor (1971) used the Fourier analysis because it provides a computation whereby identification of an image can be accomplished. An image that is closer or brighter produces unit activity of greater magnitude over a wider range of the visual field, and tends to produce an increase in firing rate in more cells. Gabor ( see Pribram, 1991) also used the Fourier analysis to account for texture. He found that the Fourier provides better resolution of the stimulus that can explain texture recognition.

Over the last thirty years, many researchers have theorized that the brain performs something like a Fourier analysis in order to represent the outside world, and researched for evidence that the brain responds to the Fourier components of a stimulus. Campbell and Robson (1968) showed that simple cells within the visual cortex of cats respond selectively to a particular range of spatial frequencies. Thus, simple cells responded more if the animal was presented with a grating composed of preferred spacing of light and dark bars. According to their results, a relationship exists between a cell's firing rate and a particular spatial frequency of a given stationary stimulus. That is, cells in the visual cortex are "tuned" to their narrow range of spatial frequency. Therefore, Campbell and Robson (1968) speculate that selected cells in the visual cortex respond to the Fourier components of a stimulus.

Other researchers performed experiments similar to the ones performed by Campbell and Robson (1968) and found nearly the same results. Maffei and Fiorentini (1973) and Movshon, Thompson, and Tolhurst (1978) took single electrode recordings from simple cells in the striate cortex of cats. They drifted sinusoidal gratings of various spatial frequencies across the visual field, and found that different simple cells were tuned to different spatial frequencies. These researchers also found that at low frequencies the temporal phase of the cells' response corresponded with the spatial phase of the stimulus. Because they used moving gratings instead of stationary, flickered ones, the researchers were able to extend the generality of spatial frequency "tuning." Specifically, they concluded not only that simple cells, as Fourier analyzers, are sensitive to spatial frequency and contrast, as shown by Campbell and Robson (1968), but that these analyzers are tuned to particular spatial phases of light and dark. Therefore, they showed that simple cells not only analyze frequency and amplitude information, but also encode the spatial phase of a particular stimulus in terms of the temporal phase of the cells' response.

DeValois, DeValois, and Yund (1979) describe five experiments designed to test two hypotheses of how cells in the visual cortex process sensory input. One view was that proposed by Hubel and Wiesel (1961) in which cells act as feature detectors, responding maximally to bars and edges of specific orientations. The second was a view developed by Campbell and Robson (1968) who stated that cells act as spatial frequency selectors, responding to a particular range of spatial frequencies. These two views make different predictions as to how cells in the striate cortex should respond to gratings and checkerboards. Each experiment was conducted on either cats or macaque monkeys in which single electrode recordings were obtained from both simple and complex cells.

In the first experiment, gratings and checkerboard patterns were drifted across the animal's visual field at different orientations. If the feature detector notion was true and the cells were firing in response to the orientation of the edges of the checkerboard, then the cells should have maximal responses to the same orientation for both simple square-wave gratings and checkerboard patterns. However, if the cells were acting as spatial frequency selectors and responding to the fundamental Fourier component of the stimulus, then the cells should prefer checkerboard orientations with Fourier components corresponding to those of the checkerboard rather than the frequencies determined by the width of their checks. This is because the Fourier decompositions of the checkerboard pattern results in frequencies that are different from those defined by the widths of the bars that compose the checkerboard. Results supported the spatial frequency selector hypothesis. The researchers found that every simple and complex cell examined responded maximally to checkerboards that were oriented 45° off from that of their preferred square-wave grating. However, these results are inconclusive because it could be said that cells responded differently to checkerboards than gratings because checkerboards consist of orthogonal edges, in which case, edge-specific cells compute vector sums.

Thus, a second similar experiment was done that used plaids. If the feature detection notion was true, then cells would also compute vector sums for the plaids. If, however, the spatial frequency notion was true, then the cells should have maximal responses to the fundamental Fourier component of the plaid, which was oriented 45° off from their preferred checkerboard orientation. The researchers found that, indeed, simple and complex cells responded maximally to plaids that were oriented 45° off from that of their preferred checkerboard orientation.

In the third experiment, gratings and checkerboards were drifted across the cat and monkey's visual field at different spatial frequencies. If the cells behave as feature detectors and are tuned to specific bar widths, then they should respond maximally to gratings and checkerboards with the same bar widths. If the cells, however, are responding to the two-dimensional Fourier components of the stimuli, then they should prefer checkerboards with the same fundamental spatial frequency as that of the corresponding square-wave grating. This is because the fundamental of the checkerboard is 1.41 times the fundamental frequency of the bar width used to compose that checkerboard. Once again, the results confirmed Campbell and Robson's theory. DeValois et al. found that cells preferred checkerboards made up of bars that were 1.41 times the bar width of the preferred square-wave gratings. Thus, cells responded maximally to gratings and checkerboards having the same fundamental frequency, not the same bar width.

In the fourth experiment, the contrast sensitivities of cells to gratings and checkerboards were measured. If the feature detector notion was correct and cells were firing in response to the contrast of the stimuli, then cells should be more sensitive to checkerboards than gratings. If the spatial frequency selector view was right and cells were firing in response to the amplitude of the stimuli's Fourier components, then the cells should be more sensitive to gratings than checkerboards of the same contrast. Indeed, they found that cells responded less to checkerboards than to sine- and square-wave gratings of the same contrast.

In the final experiment, cell responses to higher harmonics of gratings and checkerboard patterns were examined. Checkerboards, in which the upper harmonics and not the fundamental harmonic were present, were drifted across the animal's visual field. If cells acted as feature detectors, then their orientation tuning could be predicted from the orientation of the edges of the gratings and checkerboards. However, if cells act as spatial frequency selectors, then their orientation tuning could be predicted from the orientation of the higher Fourier harmonic components. Thus, cells should be able to respond to higher harmonic components as well as fundamental components of a pattern. Indeed, the researchers found that cells responded to higher harmonics under the right conditions. Although, cells responded minimally to sine-wave gratings that were 1/3 the preferred spatial frequency, they did respond significantly to f/3 (where f is the preferred frequency of the cell) square-wave gratings. These responses to the f/3 square- wave gratings are assumed to be responses to the third harmonic.

These five experiments performed by DeValois et al. demonstrate that the cells in the visual cortex act more as spatial frequency analyzers than as feature detectors. Thus, the cells are not sensitive to different bar widths; instead, cells respond to spatial frequency components of a stimulus.

Albrecht, De Valois, and Thorell (1980) extended the work of DeValois et al. (1979) by looking more closely at cell responses to bar widths versus spatial frequencies. They took single electrode recordings from the striate cortex of cats and macaque monkeys. The researchers concluded that bar width was not important to the cells because they responded the same to narrow bars as they did to wide bars. Spatial frequency, however, was found to be very important to cortical cells. Each cell responded only to a limited range of spatial frequencies.

DeValois, Albrecht, and Thorell (1982) further extended this line of research by measuring the different band-pass characteristics of cells in the visual system of the macaque monkey. They wanted to investigate the tuning characteristics of simple cells in the striate cortex in order to find out how narrowly or broadly they are tuned. They drifted spatial sine-wave gratings across the receptive fields of the cells and then took single electrode recordings. They found that cells in the lateral geniculate nucleus (LGN) tend to be broadly tuned; whereas, cortical cells tend to be more narrowly tuned to either lower or higher frequencies. Cortical cells consist of simple and complex cells, both of which are tuned to higher frequencies within the center of the cortical region and lower frequencies toward the periphery. This provides further evidence that cells in the striate cortex, as well as the LGN, are sensitive to spatial frequency.

Pollen, Lee, and Taylor (1971) earlier demonstrated that cells acting as spatial frequency selectors perform strip Fourier transforms. They took single electrode recordings from visual cortical cells in primary visual cortex. By presenting bars of light varying in size, orientation, and luminance, they were able to determine certain properties, such as firing rate and response latency, of the cells. They found that different cells responded more to some bar widths and brightnesses than others. These researchers then deduced that the visual cortex is representing a given stimulus by decomposing the stimulus into its component spatial frequencies. Pollen et al. stated that this Fourier analysis occurs in strips, instead of all at once as would be predicted by the global Fourier hypothesis, over a number of parallel circuits at the simple cell stage. As this information goes from the simple cells and then to complex cells, the strips are integrated and the pieces of information are put together.

Pollen and Ronner (1981a) further investigated Fourier representations of gratings in the striate cortex by observing the response characteristics of simple and complex cells. They took single electrode recordings from cells in response to drifting sine-wave and square-wave gratings. They found that simple cells respond linearly to changes in the bar width, length, and luminance of the gratings. Thus, the bar widths of the grating can be predicted by adding and averaging the firing rate of simple cells in the visual field.

They also drifted sine-wave and square-wave gratings across the visual field at different speeds in order to determine each cell's preferred temporal frequency. As before, they found that certain cells respond more to some frequencies than others and that each cell has a preferred frequency to which it mostly responded. These researchers also found that neighboring cells responded to the same spatial frequency and orientation, and that they fired at or very closely to 90° out of phase with one another. This suggests that cells in the primary visual cortex represent a stimulus in terms of its sine and cosine Fourier components. They also expected to find an interaction between the cells’ fundamental frequency and the third harmonic of a square-wave grating. This is because a square wave is composed of a fundamental frequency plus all odd harmonics. Therefore, any square wave stimulus should induce firing in cells that represent the fundamental and third harmonic for each square wave stimulus. They found that when the actual cell frequency response profiles were compared to ones that would be predicted by a Fourier analysis, a few cells fit that predicted pattern.

Although, Pollen and Ronner’s (1981a) results seemed to fit the spatial frequency selectivity hypothesis, they did run into a problem of truncation. The level of truncation, how low the firing rate will go, for a given cell is dependent on how low its spontaneous firing rate is. Some of the cells they recorded from had low spontaneous firing rates and their troughs did not appear in the response profile. The researchers postulated that the problems of truncation could be resolved if they could find cells firing 180° out of phase with the sine/cosine pair.

In fact, Pollen and Ronner (1981b) published a follow up study in which they resolved the problem of truncation. They found that the information lost by truncation could be saved if there were two pairs, instead of one, of simple cells with the same preferred orientation, direction, and spatial frequency. The two pairs must have opposite polarities (odd and even) so that when the responses of one pair truncate, the second pair will begin responding.

Pollen and his colleagues have studied the brain in terms of Fourier analyzers. Pribram agrees, and has done so since the epilogue in Languages of the Brain (1971). According to Pribram’s holonomic theory, the Fourier transformation forms a distributed pattern of activity within the synaptodendritic network of a neuron activated by a stimulus in the receptive field. The voltage pattern across the synaptodendritic network is the result of integration of inputs to that network, and changes in this pattern over space and time are represented by a Fourier transform of the input. It is useful to study single units in order to gain an understanding of what they sample from the dendritic network, but the synaptodendritic network is where the processing takes place. This network is where information is integrated and where transformations are computed. Therefore, unit activity represents a sampling from a portion of the synaptodendritic network.

A Fourier analysis, however, depends not only on an analysis of frequency components of a stimulus, but also their phase relationships. Oppenheim and Lim (1981) have studied the importance of phase for pattern recognition. They found that phase information is useful for image, speech and crystallographic structure reconstruction. They also demonstrated that phase-only information is significantly more important than amplitude-only information for reconstructing an original waveform. In fact, if the frequency and phase of a signal are known, amplitude can be adequately estimated.

The usefulness of phase tuning may also be demonstrated in the visual system. Humphrey and Saul (1998) studied the relationship between directional tuning and spatiotemporal (S-T) structure of simple cells' receptive fields. Some simple cells are defined as S-T inseparable, meaning that the cell's response to a stationary, flickered sinusoidal bar of light is progressively phase lagged as one moves across the cell's receptive field. This property is illustrated in Figure 3A. This spatiotemporal inseparability confers directional selectivity for a particular simple cell. Other simple cells are defined as S-T separable, meaning that their receptive fields do not show spatiotemporal lags, and therefore are not responsive to movement in a particular direction. Thus, S-T separability and directional selectivity are highly correlated. The researchers hypothesized that strobe-rearing would affect the S-T structure of the receptive fields in the striate cortex. If the directional selectivity of the cells was also affected as a result of strobe-rearing, then there would be strong evidence to support the notion that the mechanism for S-T inseparability is at least part of the mechanism for directional tuning.

Humphrey and Saul (1988) used fourteen cats reared from birth to eight or nine months in a colony room illuminated by a 8 Hz strobe light. They also used five cats reared under normal lighting conditions to provide comparison data. After each cell's preferred orientation and spatial frequency tuning selectivity was determined by drifting sine-wave gratings across the receptive field, receptive field structure was assessed by placing stationary bars over different positions in the receptive field and flickering them in order to achieve temporal frequency. They then recorded response timing with respect to the phase of sinusoidal stimulation. If response timing changed as the bars were moved to different positions in the cell's receptive field, then the cells were classified as S-T inseparable. However, if response timing remained the same, the cells were classified as S-T separable.

The results showed that strobe-rearing reduced direction selectivity among simple cells in the striate cortex. This finding is illustrated in figure 3B. It was postulated that

this loss was due to the fact that the S-T inseparable receptive field structure was destroyed. While S-T inseparable receptive fields were abundant in layer 4 and sparse in layer 6 in normal cats, S-T inseparable receptive fields in strobe-reared cats were eliminated in both of these layers. That is, all simple cells in strobe-reared cats were S-T separable.

Figure 3. An S-T inseparable simple cell is illustrated above (A). The progressive phase

lags are quite apparent. After strobe-rearing, however, these phase lags are no longer

visible (B). (From Humphrey, 1998)

In a companion paper in the same issue, Humphrey, Saul, and Feidler (1998) examined specific changes in response timing within receptive fields of LGN and simple cells of strobe-reared cats, and related the S-T timing changes to the lack of directional selectivity. Unit response data were obtained from the LGN and simple cells in strobe-reared and normal cats. Once again, response timing in cortex was assessed by sinusoidally modulating stationary bars over time and over different positions in each cell's receptive field, and then recording the cell's response. Response timing in LGN was acquired in a similar manner, except that spots of light were used in place of bars. Response timing is categorized as either lagged or nonlagged. Results showed that the range of timings among populations of cells within cortex and LGN were not affected by strobe-rearing. Thus, a population of simple cells demonstrate a normal range of lagged and nonlagged inputs. However, timings within an individual cell's receptive field were affected in that they had only one or two lag responses instead of a progression of lags. Individual LGN cells' responses were not affected by strobe rearing. Thus, strobe rearing disrupts the S-T inseparability of simple cells by eliminating progressive phase lags across the spatial extent of the cell's receptive field. However, strobe rearing does not affect the lag profiles of LGN cells.

Humphrey, Saul, and Feidler (1998) developed a hypothesis of geniculocortical convergence to account for the development of S-T inseparability and its loss due to strobe rearing. This hypothesis is illustrated in figure 4. According to the researchers, the receptive fields of lagged and nonlagged cells in LGN exhibit spatial quadrature in that the cells respond to 1/4 the wavelength of a given spatial frequency. Both the lagged and nonlagged cells project to a simple cell. The receptive fields of lagged and nonlagged cells in the LGN also exhibit temporal structure because the cells fire 90° out of phase in response to

Figure 4. Model illustrating geniculocortical convergence hypothesis. A: receptive fields of

lagged and nonlagged LGN cells respond to ¼ the wavelength of a given spatial frequency.

Both cells project to a simple cell (not shown). B: LGN responses are 90° out of phase in

response to a sinusoidally modulated spot of light. C and D: luminance profiles for gratings

moved in the preferred and nonpreferred direction under normal lighting conditions. E and F:

temporal relationships of LGN responses under normal lighting conditions. G: luminance

profile for gratings moved in the preferred direction under strobe conditions. H: temporal

relationships of LGN responses under strobe conditions. These responses are shorter and differ

in latency compared to LGN responses under normal conditions. (From Humphrey, 1998)

sinusoidally modulated spots of light. When a sinusoidal grating is moved in the preferred direction, these lagged and nonlagged cells fire synchronously. Because the cells fire in such a synchronous manner, they act as sine and cosine Fourier components. These synchronous responses from many LGN cells help to strengthen synaptic connections with simple cells. Thus, a series of lags will be present in the simple cells' receptive field. When a sinusoidal grating is moved in the nonpreferred direction, the cells do not fire synchronously. Instead, they fire 180° out of phase with one another. During strobe-rearing, however, you don't get these firing patterns. Because the animals are exposed to light for such a short duration, cells respond only to a portion of the grating instead of responding to the entire grating. Due to this lack of synchrony, cells whose convergence could provide for a series of lags are lost.

While Humphrey, Saul, and Feidler (1998) focused on the geniculocortical mechanism responsible for S-T inseparability, Murthy and Humphrey (1999) focused on the two intracortical mechanisms. One such mechanism is linear inhibition, in which intracortical inhibition enhances S-T orientation, which then strengthens direction selectivity. The other is a nonlinear process in which inhibition either lowers membrane potentials relative to spike threshold, or raises spike threshold relative to the resting membrane potential. In order to determine which of these mechanisms contributes the most to direction selectivity, the researchers blocked intracortical inhibition using a GABA antagonist and then measured S-T orientation and direction selectivity. If the linear inhibition model is more prominent, then S-T orientation would be affected as well as direction selectivity. However, if the nonlinear mechanism accounts more for directional selectivity, then S-T orientation would not be altered. They found that after blocking intracortical inhibition, the S-T orientation of cells in layer four was significantly lowered. Thus, a linear mechanism takes place in this layer. However, the researchers also found that the S-T orientation of cells in layer six was not significantly affected by intracortical inhibition. This suggests that a nonlinear mechanism is also operating in the brain. Although the nonlinear mechanism is difficult to account for, the linear mechanism may be explained in terms of a Fourier analysis.

This research is important because it demonstrates that a Fourier decomposition can be used to account for classic receptive field properties, such as directional selectivity. While the classic notion of receptive fields can not account for phenomena such as texture, the Fourier model can. These findings allow one to look at classic receptive field research, and therefore our understanding of the structure-function relationship in brain processing, in a new light.

If the brain is performing a Fourier analysis of sensory input, then to reconstruct the input phase information is needed. Recent research by Jenson (1999) provides good evidence for the importance of phase information. He measured the firing patterns of cells in the hippocampus as rats proceeded through a maze with five locations. Neural activity was represented by phase codes which provided information as to the rat’s position in the maze. Communication among the neural networks allows the hippocampus to predict upcoming maze locations as well as have representations of the current location. Jenson claims that these principles might extend to areas of the brain other than the hippocampus. Thus, it is possible that other neural networks, such as those in the rat somatosensory cortex, might exchange information in a similar fashion.

Spectral Processing in the Vibrissal System

The main point of this thesis is that a spectral account of processing in the vibrissal system provides a more comprehensive, and therefore useful means of processing inputs than an account in purely spatiotemporal terms. This point is based on two branches of research, which, although presented here, occurred in parallel. The first compares the development of our understanding of processing in the visual and vibrissal systems. Although some differences exist, the two systems are similar enough to consider that the same basic laws apply to both. Because these similarities exist, theories used to account for visual phenomena may also be used to account for processing in the vibrissal system.

The second examines how research and theory in the vibrissal system has paralleled the progress of research and theory in the visual system. Over the last forty years, a great deal of research has been conducted in order to better understand visual processing. Through experimentation, researchers have developed theories attempting to account for such visual phenomena as pattern recognition, object constancy and directional selectivity within receptive fields. Research in the vibrissal system has attempted to account for similar phenomena, beginning by focusing on the relationship between structure and function in the whisker system. However, because there are inconsistencies that do not fit into the structure-function isomorphic view, there exists a need for a different model to account for such disparities. A discussion of these disparities compose the second part of this section.

The two branches of research lead to a common end result. They both lead to the conclusion to supplement traditional spatiotemporal views with a spectral view. According to the evolution of visual research, thinking in spatiotemporal terms will not allow for a complete understanding of how the visual system works. If, however, one were to look at the visual system in spectral terms, phenomena such as object constancy (Pribram, 1991), as well as classic receptive field properties such as directional selectivity (Humphrey and Saul, 1998, Humphrey et al., 1998; Murthy and Humphrey, 1999) can be accounted for.

Now let’s discuss the first branch of research mentioned earlier. Researchers have found that there are several ways in which the operation of the vibrissal system is like that of the visual system. One such similarity is that receptors in both systems are selective to certain aspects of a stimulus. The retina is composed of a “sheet” of receptors, groups of which are functionally connected with units in the primary visual pathway in a topographical arrangement. These functional groups of receptors enable units to be selective to stimulus components such as orientation, direction, luminance, and spatial frequency. More recent research has demonstrated that several space-time characteristics of visual units can be accounted for in spectral terms. In addition, a spectral analysis of visual processing provides for more complex visual phenomena, e.g., those dependent on texture. The receptor cells within whisker follicles may also be conceived as components of a “sheet” of receptors. Whiskers are generally stimulated in groups. Such groups on the mystacial pad thus provide a “sheet” of receptors stimulated by a stimulus. Like units in the visual pathway, units in the vibrissal system are also differentially responsive to various dimensions of stimulation, for example, direction, velocity, and texture. In addition, Simons (1983) used stimulators to deflect individual vibrissae either alone or in combination with other whiskers. He found that three factors in the experiment were important determinants as to how the vibrissa units would respond. The first was the direction of stimulation. While some units responded maximally to one particular angle of whisker deflection, others responded little or not at all. The second was the sequence in which a pair of whiskers was deflected. For example, when whisker D2 was stimulated before D3, units responded only to the deflection of D2. However, When D3 was stimulated first, units responded to deflections of both whiskers. The third was the particular combination of whiskers that were stimulated. For example, when whiskers B1 and b (a neighboring whisker) were simultaneously deflected, units exhibited little or no response. However, when whiskers B1 and B2 were simultaneously deflected, units fired a great deal. These findings demonstrate that unique spatial and temporal components of vibrissal stimuli are integrated within the vibrissal system, resulting in differential responses by individual units.

Another similarity between the visual system and the vibrissal system, which accounts for the previously discussed neuronal properties, is that receptive field formation in both systems is dependent on inhibitory processes. Phelps (1973) investigated the importance of lateral inhibitory interactions between cortical neurons. Single electrode recordings were taken from 250 neurons in the striate cortex of 29 cats. After mapping the receptive fields of these cortical neurons, either one or two spots (or in other experiments, bars) of light were passed in front of the cat’s receptive field along an axis parallel to the preferred direction. The researcher then constructed an interaction map, which depicted each cells response over time. By comparing these responses, it was found that interactions occurred between neighboring neurons, each of which received projections from about three to six fibers along an input/output pathway. Such interactions were found to moderate the responses along the input/output pathway. Thus, Phelps (1973) concluded that lateral inhibitory interaction between the cortical neurons was responsible for direction selectivity.

Batuev, Alexandrov, Scheynikov, Kcharazia, and An (1989) also investigated the role of inhibitory processes in the formation of receptive fields. They took single electrode recordings from the vibrissal system of adult cats. After determining if neurons were directionally sensitive or not, inhibition was either blocked or enhanced in order to determine its effects on directional “tuning.” Picrotoxin and bicuculline, two GABA antagonists, were used to block inhibition and distant glutamate application was used to activate inhibition. After inhibition processes were blocked, previously directionally selective cells lost their selective properties. After inhibitory mechanisms were activated, cells that were not previously directionally selective exhibited properties of directionally selective cells. The cells that were directionally selective prior to the activation of inhibitory mechanisms, were either more selective, less selective, or exhibited changes in their directional preferences after the activation. These results suggest that intracortical inhibitory mechanisms are responsible for the formation of, or changes in, receptive field properties such as directional selectivity.

The results of these two experiments are consistent with those of Murthy and Humphrey (1999), who measured direction selectivity in the visual system of the cat. They also blocked intracortical inhibition using a GABA antagonist and found that cells lost their directionally selective properties after the blocking.

Despite their similarities, at least three differences exist between processes in the visual system and the vibrissal system. The first, is that certain cells in the two systems exhibit opposite properties. The receptive fields of ganglion and thalamic cells in the visual system are small and specific to spatiotemporal dimensions, yet as one moves through cortex, the receptive fields become broad and less specific to these dimensions and more specific to spectral dimensions. However, the receptive fields of receptor and thalamic cells in the vibrissal system are broad and unspecific, while those of barrel cells are small and specific to spatiotemporal dimensions. The second, is that the mechanisms of directional selectivity in the visual system have been determined to be geniculocortical, while those in the vibrissal system are so far considered to be intracortical. The third, is that the processes responsible for directional selectivity in the visual system are described in linear terms (Humphrey and Saul, 1998, Humphrey et al., 1998; Murthy and Humphrey, 1999), while such operations for directional selectivity in the vibrissal system are described in nonlinear terms (Simons, 1983). These differences could be due to the fact that in the vibrissa experiments, single receptors are stimulated while in the visual experiments, the entire retinal surface is always involved in the stimulation.

Having compared properties of the vibrissal and visual systems, we may now turn our attention to the theoretical trends that have been set by research on the visual system and followed by that on the vibrissal system. Just as visual system researchers began by studying basic receptive field properties, so did researchers in the vibrissal system. Vibrissal system research initially focused on structure-function relationships which relied on the neuron as the functional unit of neural integration. Recordings were taken from individual units in response to systematic variations in a single stimulus parameter. However, there are inconsistencies that this structure-function view can not account for. For example, barrels exist in other parts of the rodent brain that are unrelated to the whiskers. Also, cytoarchitectonically distinct barrels exist in only a few of the mammalian species that have prominent mystacial vibrissae (Woolsey et al., 1975; Rice et al., 1985; Waite et al., 1991). Another inconsistency is that many barrels respond to adjacent whiskers as well as their principle whisker. Further, barrel neurons respond to multiple stimulus dimensions. Although the structure-function view based on arrangements of neurons is appealing, it does not hold for all of the properties of the vibrissal system.

Thus, researchers began to consider computation across “ensembles” of units. In fact, that is what Nicolelis et al., (1993) and Nicolelis, (1997) did. They used a large array of electrodes in order to record from populations of units. They looked at the cell responses in barrel cortex as three-dimensional plots, changing as a function of space and time. This shift from looking at the vibrissal system in terms of single units to viewing it in terms of groups of units paralleled a similar shift in visual research.

Although, space-time representations of unit activity are useful for gaining a better understanding of processing, they are not the end-all. Processing in the spectral domain might prove to be a complementary brain mechanism and more comprehensive theoretical strategy. Holonomy, for example, takes the more basic notion of neural ensembles and adds to it a complementary view of structure and function. As mentioned earlier, Pribram’s (1991) holonomic view considers the synaptodendritic network as another basic functional unit in the nervous system. Voltage patterns within the network are the result of the timing of various inputs into the network. Therefore, the changes in voltage patterns over time comprises neural integration. Sampling from the network in spectral terms provides for an account of the perceptual world, that adds to an account provided by neural ensemble views. If trends in vibrissal system research continue to follow those in the visual system, then there should soon be another shift from looking at unit activity solely in spatiotemporal terms among groups of units to looking at unit activity in spatiotemporal and spectral terms as a reflection of neural integration within the synaptodendritic network.

It is important to note, however, that such a shift does not completely discount the importance of the classical relationship between structure and function. Structure-function relationships are necessary to biological understanding, but the key lies in identifying the relevant structure to which particular functions should be attached. While the classic structure-function view considers the neuron and its axons to be the fundamental functional unit, the holonomic view places such importance on the synaptodendritic network. The function associated with the synaptodendritic network is spectral; whereas, the function associated with the neuron is one of sampling from the spectrum and transmission of the sample. Thus, the spectral view is able to better account for information processing than a purely spatiotemporal view. It provides a computational breadth (Pribram, 1991) that accounts for how the brain makes correlations. An additional benefit of the spectral view is that it can account for inconsistencies left behind by the classical receptive field view. For example, when the structure of focus is the synaptodendritic web instead of the neuron, the fact that a barrel responds to adjacent whiskers as well as the principle whisker or that barrel units respond to multiple stimulus dimensions, no longer presents a problem. Since each neuron samples from a particular region of the synaptodendritic network, and each stimulus affects the voltage pattern within the network, then each neuron responds differentially to each stimulus. In addition, the spectral approach can also account for the enhanced selectivity of units that results from inhibition interacting within a barrel (Batuev et al., 1989). For example, Hovis (1997) demonstrated that the sharpening of responses in the dendritic network actually provides for enhanced differences between different patterns of input.

Summary and Conclusions

The somatosensory system of the rat, provides a model system to better understand the manner in which sensory processing proceeds in the brain. The vibrissal system serves as a model somatosensory system due to four basic characteristics that it possesses. First, it is easy to manipulate stimulus parameters. For example, it is a fairly simple task to control the number of whiskers stimulated or the order in which they are stimulated. Second, a distinct anatomical pathway exists. Projections in the vibrissal system may easily be followed from the receptor cells in the whisker follicle, to the trigeminal brainstem nuclear complex, to the thalamus, and then to the primary somatosensory cortex. Third, a unique topographical structure exists in which a one-to-one relationship exists between each vibrissa and its corresponding barrel. Fourth, classical receptive field properties (i.e., functional correlations) are apparent between the vibrissae and barrels. This architecture as well as the functional correlation between the vibrissae and the barrels, suggests that the vibrissa-barrel neuraxis is an attractive model for studying the structure, function, and processing within the somatosensory system.

Despite the fact that large segments of retinal receptor surfaces are stimulated, the visual system was explored as well because its functional properties seem to parallel those of the vibrissal system, and because its computational properties have been so thoroughly explored. Also, our understanding of processing in the visual system has evolved in a manner which may serve as a model for the development of our understanding of vibrissal processing. Researchers of both the visual and the vibrissal system began by studying basic receptive field properties. However, the classic receptive field view was not able to account for such properties as texture in the visual system. In the vibrissal system, this approach explores why many barrels respond to more than their principle whisker, why destruction of the whisker-barrel relationship does not disturb its functional properties, and why a multidimensional stimulus does not produce a single peak in a surface distribution. Thus, a shift is occurring in which researchers are beginning to consider computation among ensembles of units instead of looking at the functional properties of single units.

The further evolution of research and theory in the visual system has demonstrated that neural processing should be looked at in spectral as well as spatiotemporal terms. Likewise, the evolution of research and theory in the vibrissal system is also heading toward understanding processing in spectral terms. By taking a holonomic approach, a more unique view of the structure-function relationship may be added to the more basic notion of neural ensemble processing in strictly spatial and temporal terms. Also, instead of focusing on the neuron as the primary structure of study, it is advantageous to look at processing that takes place in the synaptodendritic web. Because the function associated with the synaptodendritic network is computational and distributed, it is better able to take advantage of the spectral dimension. The spectral approach is able to account for inconsistencies in the results of research in the vibrissal system as well as phenomena such as texture; whereas, a classical receptive field view cannot. In