Rats explore their surroundings through rhythmic movement of their mystacial vibrissae. At any given moment, multiple whiskers are simultaneously moved and may contact the surface of an object. The aim of this work is to understand how simultaneous multiple-whisker deflections are processed in the somatosensory cortex. Arrays of 25 electrodes were inserted into the vibrissal representation of barrel cortex of adult rats. Multi-unit responses were recorded during (i) stimulation of single whiskers, and (ii) simultaneous stimulation of two, three or four whiskers of a whisker arc or whisker row. The whole-array response elicited by the simultaneous stimulation of multiple-whiskers (observed response) was compared to a multiple-whisker response predictor, defined as the sum of the whole-array responses to the separate stimulation of the corresponding single whiskers. The observed response to stimulation of four whiskers was nearly always less than the predicted response, indicating a sublinear summation of multiple coincident inputs. Examining the poststimulus time course of sublinearity, we found that the earliest cortical response to whisker deflection – reflecting the thalamocortical volley – was linear, whereas the successive cortical response was highly sublinear. This suggests a cortical origin of the phenomenon.
Rats use their facial whiskers to explore the environment (Vincent, 1912; Welker, 1964; Wineski, 1983; Carvell and Simons, 1990). Tactile information is acquired during ‘whisking’, repetitive cycles of forward and backward whisker movements across objects (Welker, 1964; Carvell and Simons, 1990). Multiple whiskers can be positioned to contact objects, either in sequence or simultaneously. Therefore, understanding the basic functioning of the vibrissal barrel cortex requires an examination of the integration of information from spatially and temporally patterned deflections of mystacial vibrissae.
The whisker representation of rat somatosensory cortex (SI) is characterized by the presence of anatomically identifiable groups of neurons in layer IV called ‘barrels’ (Woolsey and Van der Loos, 1970). The barrel forms the basis of a cortical column extending from layers II–VI, and each column is associated with a single whisker on the controlateral mystacial pad. Single-unit studies have shown that neurons in a given cortical column respond maximally to the deflection of the principal whisker (PW), while they respond more weakly to the deflection of several surrounding whiskers (SWs) (Welker, 1971; Simons, 1978). With regard to the interaction of multi-whisker inputs in barrel cortex, previous studies gave contrasting results. Single-unit and optical imaging studies have shown that deflection of SWs diminishes the excitatory responses elicited by a subsequent stimulation of the PW (Carvell and Simons, 1988; Simons and Carvell, 1989; Goldreich et al., 1998). Furthermore, the amount of inhibition exerted upon the excitatory cells of a barrel increases with the number of stimulated adjacent whiskers (Brumberg et al., 1996). On the other hand, in other reports it was argued that simultaneous (Ghazanfar and Nicolelis, 1997) or nearly simultaneous (Shimegi et al., 1999) stimulation of multiple whiskers produces mainly supralinear summation.
These contradictory views have important implications in understanding information processing in rat barrel cortex. The purpose of this study is to re-examine this problem by comparing the linearity of responses of large populations of cortical neurons recorded in parallel during simultaneous stimulation of multiple whiskers. The results of the present study suggest that coincident inputs from four whiskers are summated sublinearly: cortical neurons respond with lower magnitude than they would if they integrated the inputs associated with each whisker independently. Besides confirming the basic phenomenon of sublinear summation of tactile inputs in somatosensory cortex, one goal of the present work is to generate a global picture of cortical population responses during single and multi-whisker stimulation.
Materials and Methods
Subjects, Animal Preparation and Histology
All procedures conformed to NIH and international standards concerning the use of experimental animals. Adult male Wistar rats (n = 18) weighing ~350 g were used. Of these, only the rats (n = 9) in which the electrode array sampled three rows of barrels are included in the dataset. Extended experimental methodology is given elsewhere (Rousche et al., 1999). Animals were anesthetized with urethane (1.5 g/kg body wt, i.p.) and placed in a Narashige (Tokyo) stereotaxic apparatus. Left somatosensory cortex was exposed by a 5 mm diameter craniotomy. During the recording session body temperature was maintained at 37.5°C and anesthesia was held at a consistent depth signaled by a 2–4 Hz cortical burst rate (Armstrong-James and George, 1988; Erchova et al., 1998). Supplemental doses of urethane were administered as necessary. At the end of the experiment, subjects were perfused with saline and 4% paraformaldehyde. After postfixation in 20% sucrose, the cortex was separated from the underlying white matter and either frozen or else embedded in paraffin. The block of tissue was cut in 60 μm sections (if frozen) or in 20 μm sections (if paraffin-embedded) in the plane of section. The sections were processed with cresyl violet or thionin.
The vibrissal region of left somatosensory cortex was identified according to vascular and cranial landmarks and stereotaxic coordinates (Hall and Lindholm, 1974). Physiological recordings were made using arrays of 25 silicon microelectrodes (impedance of 80–300 kΩ at 1 kHz) geometrically arranged in a 5 × 5 grid-like pattern, with 400 μm inter-electrode spacing (Bionic Technologies, Inc., Salt Lake City, UT). Microelectrodes were implanted orthogonal to the pial surface by a pneumatic impulse inserter (Bionic Technologies, Inc.) with the dura left intact, reducing cortical trauma (Rousche et al., 1999). The insertion force was such that the microelectrode tips reached the middle layers of cortex. Figure 1 illustrates the electrode array position in a representative case (also see subsequent sections of Materials and Methods).
Neural signals were amplified at 5000×, filtered at band-pass 250–7500 Hz and sent to a digital signal processor (Bionic Technologies). Digitized signals (30 000 samples/s) were collected by the ISA board installed in the PentiumTM PC and were viewed and stored by data acquisition software (Bionic Technologies). Utilizing this software, experimental status was assessed on-line on the computer monitor, while the audio speaker played above-threshold neural events.
To make certain that the same neurons were not recorded within the neuronal clusters at adjacent electrodes, we carried out a cross-correlation analysis on spontaneous activity. The reasoning was that if the neuronal clusters at two electrodes ‘shared’ some neurons in common, then the cross-correlation histogram (CCH) would show a very high peak at the central bin, since the same action potentials would be recorded at both electrodes. In every experiment, correlations between all neighboring electrode pairs were examined using 3 min of spontaneous activity. The resulting CCHs, with 2 ms bins, usually showed peaks spreading across ~10–20 ms before and after 0 ms (where 0 ms indicates simultaneous events at the two channels). These peaks reflect the ‘bursty’ firing pattern of cortical neurons under urethane anesthesia, which is known to be highly correlated across barrel cortex (Fox and Armstrong-James, 1986; Armstrong-James and George, 1988; Erchova et al., 1998) The essential observation is that the central peak, at 0 ms, never contained a greater number of coincidences than could be expected based on the ‘burst’ of correlated activity occuring on both sides of 0 ms. We conclude that the activity of the neuronal cluster recorded at one electrode was never recorded at nearby electrodes.
Before collecting data, all whiskers on the right side of the snout were cut to 3–4 mm. Whiskers were deflected, either individually or in sets, by hooks fixed to a piezoelectric ceramic bimorph wafer (Morgan Matroc, Bedford, OH) positioned just below the whisker shaft, 2 mm from the skin. Figure 2 illustrates the stimulation protocols used in these experiments: (A) single whisker deflection; (B) arc deflection, defined as the simultaneous stimulation of two, three or four whiskers of the same arc; (C) row deflection, defined as the simultaneous stimulation of two, three or four whiskers of the same row. The stimulus was an up-down deflection of 80 μm amplitude and 500 ms duration delivered 50 times for each site at 1/s.
Multi-unit action potentials were discriminated off-line and action potentials were time-stamped with 0.1 ms resolution. To avoid the sampling bias that might be introduced by selecting one single-unit at each electrode, all action potentials recorded at a single electrode were summated to create a neural cluster recording.
The evoked sensory response at each electrode was calculated as the number of spikes occurring during the first 100 ms after stimulus onset minus the number of spikes in the 100 ms preceding stimulus onset, averaged across 50 whisker deflections. Electrodes were identified as responding to a given stimulus by comparing across trials the number of spikes occurring after stimulus onset (0–100 ms; stimulus onset = 0) with the number of spikes before stimulus onset (–100–0 ms) using the Wilcoxon signed-ranks test (P < 0.01 accepted as a significant response). To calculate the cortical territory activated by a given stimulus we counted the number of responsive electrodes (Wilcoxon signed-ranks test, P 0.01).
Laminar and Columnar Location of Recording Sites
To support the estimate of electrode depth gained from in situ photography (Fig. 1), at each electrode the modal latency for principal whisker stimulation was calculated (Armstrong-James and Fox, 1987). By measuring the time of the first spike after stimulus onset on every trial, a first spike time histogram was constructed with 1 ms bins in the interval 0–100 ms post-stimulus. The modal latency was the mode of this histogram. Neurons in layers receiving input from the ventral posterior medial nucleus (VPM) are known to respond to stimulation of the principal whisker with an average modal latency of 8.5 ms. In contrast, neurons in other layers respond with average latencies of at least 10.5 ms post-stimulus (Armstrong-James and Fox, 1987; Armstrong-James et al., 1992). In our sample, the mean value of the total distribution of principal whisker modal latency was 8.1 ms (SD = 2.3, IQR = 6–9 ms, n = 105). This finding, together with the photomicrograph of the implanted array, suggests that in all cases we were recording from neurons located in layers IIIb and IV, the laminae receiving strongest VPM input (Lu and Lin, 1993).
Examination of the barrel cortex after the physiological experiment was carried out on each experimental subject. The hope was to be able match the principal whisker at each electrode, defined by sensory response magnitude, to the position of that electrode in a barrel map derived from stained tangential sections. The analysis met with methodological difficulties: it was difficult to extract the array from the brain without causing some damage to sectors of the sampled region (Rousche et al., 1999). As a result, even in the best cases, columnar positions could be revealed only for a portion of the array; a complete barrel map containing all recording sites could not be produced. Therefore we relied on functional responses to define electrode position. We adopted the criterion of response magnitude since the strongest response of a cortical barrel-column is always evoked by deflection of the topographically associated whisker (Simons, 1978; Armstrong-James and Fox, 1987). An electrode was designated to be in the barrel-column of a given whisker if the activity evoked by stimulation of that whisker was statistically significant and was at least 1.5 times bigger than the activity evoked by the next strongest input.
Index of Response Linearity
To estimate the linearity of the cortical integration of multiple whisker inputs, a predicted response p was defined at each electrode as the sum of the separate responses at that electrode to stimulation of the two, three or four single whiskers of interest. A whole-array linearity index was then calculated as:
where oi is the observed activity at the ith electrode evoked by the multi-whisker stimulus in a selected interval and pi is the predicted activity at the same electrode in the same interval. Linearity index equal to 0 indicates linear summation; index values >0 and <0 indicate supralinear and sublinear summation respectively.
Cortical Response Linearity
The first question was whether cortical summation of multiple inputs across the post-stimulus interval of 0–100 ms depended on the number of stimulated whiskers. Figure 3 shows the average of the whole-array response linearity in four rats during simultaneous stimulation of two, three or four whiskers. A trend toward sublinearity is evident as the number of whiskers stimulated increased: when two or three whiskers were stimulated simultaneously the response was not significantly non-linear (two whiskers: P = 0.33, Wilcoxon signed rank on difference, n = 20; three whiskers: P = 0.43, n = 13). However, when four whiskers were simultaneously stimulated the response became significantly sublinear (P < 0.001, n = 10). On average, the observed four-whisker evoked response was 33.5% less than the linearly predicted response.
Figure 4 illustrates the typical sublinear effect of stimulation of four whiskers. The microelectrode array was placed in the caudal region of the barrel cortex, sampling from many different barrels (Fig. 4A). When the four whiskers of arc 2 were coincidentally displaced the observed cortical response was less than that of the linear predictor (Fig. 4C). The difference between the observed and the predicted response maps (Fig. 4D) underlines the sublinearity in the evoked response. To find out whether the whole-array sublinearity consisted of a reduction in the magnitude of the evoked activity at a subset of electrodes or, alternatively, if it was caused by a uniform response reduction, the observed and predicted responses were plotted for every electrode (Fig. 4E). All electrodes, both those strongly activated and those weakly activated, showed sublinearity. This phenomenon was characteristic of all response maps.
To directly test whether sublinear summation involved a reduced spatial extent of activated cortical territory, the observed and the predicted areas were computed as follows. From the spatial sampling density of the array, each electrode was taken to be representative of a 0.16 mm2 block of cortical territory. For the observed response map, an electrode was considered to be active according to spike counts in the interval 0–100 ms after stimulus onset (see Materials and Methods). For the predicted response map, an electrode was considered to be active if it was responsive, by the same criterion, to any of the four single whiskers. For both cases, the total active territory simply equaled the number of active electrodes multiplied by 0.16 mm2. This analysis showed that the amount of barrel cortex engaged by four-whisker stimulation usually was less than the territory predicted if summation were linear (Fig. 5). Average values were 2.7 mm2 (SD = 0.48 mm2, n = 31) for the observed response map and 3.1 mm2 (SD = 0.48 mm2, n = 31) for the predicted response map (P < 0.001, paired t-test). Sublinear integration therefore had the effect of restricting the activated territory to the more strongly activated electrodes, as suggested by Figure 4E.
To see if there was any difference between horizontally and vertically oriented stimuli, as reported previously (Ghazanfar and Nicolelis, 1997), index values for four-whisker arc and row stimuli of all rats were plotted separately (Fig. 6). The mean value of the index was less than zero for both arcs (mean index value = –0.25, SD = 0.17, n = 14), and rows (mean index value = –0.23, SD = 0.13, n = 17), and stimulus orientation did not affect linearity (P = 0.53, Mann–Whitney rank sum test). For one observation the index value was positive (0.07), indicating that supralinear summation can take place.
Time Course of Sublinearity
The poststimulus latency at which sublinearity emerges can give insights concerning the neural mechanisms of the phenomenon. The activity evoked in the barrel cortex within 7 ms of stimulus onset reflects a direct thalamocortical relay, whereas the activity which takes place >10 ms after stimulus onset derives largely from intracortical circuits; activity occurring 7–10 ms after stimulus onset has a mixed origin (Armstrong-James et al., 1993). Does the linearity of cortical responses differ in these intervals? Figure 7A shows the whole-array activity in 7 ms time bins evoked by the simultaneous stimulation of four whiskers, together with the corresponding predicted values, averaged across all rats. In the first 7 ms poststimulus, when barrel neurons are activated exclusively by the direct thalamic inputs, the observed activity had risen above the pre-stimulus level of activity (n = 31; P = 0.03, Wilcoxon signed rank) but was not significantly different from the predicted activity (n = 31; P = 0.125, Wilcoxon signed rank). In the next time interval, when intracortical circuitry presumably began to be engaged, the observed activity was significantly less than the predicted activity and remained so across successive intervals. The linearity index was computed from 0 to 91 ms after stimulus onset, in bins of 7 ms, and its time course is illustrated in Figure 7B. In the period 0–7 ms the index value was slightly negative, though not significantly different from 0 (n = 31, P = 0.159, Wilcoxon signed rank), suggesting that the cortical activity evoked by the first thalamic volley was nearly equivalent to the sum of the responses produced by the stimulation of each corresponding whisker alone. If sublinearity were generated at the thalamic level, then the initial cortical response would be expected to be sublinear. Index values were significantly sublinear for all successive time bins until 70 ms. These observations are consistent with predominantly cortical mechanisms of sublinear summation.
The purpose of this study was to address how multi-whisker stimuli are represented in barrel cortex. The issue is relevant since this pattern resembles more closely the stimulation that the sensory system is likely to receive during active exploration. Although rats are able to detect objects using a single whisker (Hutson and Masterton, 1986; Harris et al., 1999), this is not the more common situation. Our data show that simultaneous inputs from four but not from two or three whiskers are summated sublinearly by neurons in the middle layers of barrel cortex. Responses were sublinear both at the most strongly activated sites, as well as the more weakly activated sites. Thus, the more weakly activated sites ‘dropped out,’ leading to a reduction in the spatial extent of the engaged territory.
These results agree with previous reports (Goldreich et al., 1998; Simons and Carvell, 1989) that analyzed the response of cortical single-units or small neuronal clusters during the deflection of two whiskers with various time differences. Both studies showed that the response to the principal whisker (PW) was inhibited when preceded by 10, 20, 30 and 40 ms by surround whisker (SW) deflection. When the delay was 0 ms (simultaneous deflection), response to the PW was not inhibited. Brumberg and co-workers showed that cortical neurons' response to the PW was inhibited if one neighboring whisker received a ‘conditioning’ stimulus, a low amplitude vibration initiated 1.3 s before the deflection of the PW (Brumberg et al., 1996). In our experiments and in those of Brumberg et al., sublinearity increased with the number of whiskers stimulated. In the latter case, stimulation of three whiskers produced a significant sublinear effect; in our data a tendency toward sublinearity was evident with simultaneous stimulation of three whiskers (on average, observed cortical response was 11.1% lower than the linear predictor), but the effect was not statistically significant. The discrepancy could be explained by differences in the stimulation paradigms. First, in our case whiskers were deflected at the same instant, whereas Brumberg and co-workers applied conditioning stimuli prior to deflection of the principal whisker, perhaps causing the ‘build up’ of a higher level of inhibition (Brumberg et al., 1996). Second, we used stimuli of a smaller amplitude (80 μm as compared to 1 mm), which is likely to evoke less excitation as well as less inhibition. With stimulation of four whiskers, similar degrees of sublinearity were found. In the earlier report (Brumberg et al., 1996), response to deflection of the principal whisker was reduced by 46% when accompanied by conditioning stimulation of three neighboring whiskers; in our case, the average cortical response to simultaneous four-whisker stimulation was 33% lower than the linear sum of the responses to the four whiskers separately.
Laminar differences in Cortical Integration of Multi-whisker Inputs
Recently Shimegi and co-workers found that 37% of sampled neurons (42 out of 114 units) in the rat barrel cortex showed supralinear summation, or ‘facilitation,’ to the combined deflection of two whiskers with brief interstimulus intervals (ISI) (Shimegi et al., 1999). A conspicuous laminar difference was observed. Supralinear summation occurred for 69% of the neurons in the supragranular layers, but for only 15 and 24% of cells in layer IV and infragranular layers, respectively.
The ISI that elicited maximal facilitation differed according to the layer, but most neurons showed the greatest supralinear summation for a very short ISI (1–4 ms). When the time difference between the deflection SW and PW exceeded a few milliseconds [thus in the time ranges examined by Simons and Carvell (Simons and Carvell, 1989)], most cortical neurons showed sublinear summation. For supragranular neurons, sublinearity was maximal for ISIs of 30–100 ms.
The work of Shimegi and co-workers (Shimegi et al., 1999) is also consistent with results of Ghazanfar and Nicolelis (Ghazanfar and Nicolelis, 1997), who argued that simultaneous stimulation of three whiskers usually produced a supralinear response among infragranular neurons of rat barrel cortex. The latter recorded the response of each neuron to a fixed number of horizontal and vertical three-whisker stimuli, finding that supralinear summation occurred in about 35% of the cases [see Fig. 1B, p. 507 (Ghazanfar and Nicolelis, 1997)]. However the frequency of sublinear summation, ~65%, is similar to that observed in our experiments: in the present study, 16 out of 25 three-whisker stimuli (64%) produced sublinear summation in the whole-array response (individual datapoints are not illustrated, however the mean values are given in Fig. 3).
Origin of Sublinear Integration of Multi-whisker Inputs
In principle, the sublinearity in cortical responsiveness to multiple-whisker stimulation could exist if the trigeminal or thalamic neurons were themselves performing a sublinear summation, and relaying this information to cortex. However, in our data the linearity index values begin to be significantly negative only 7–14 ms after stimulus onset. In other words, the number of evoked short-latency (0–7 ms post-stimulus) cortical spikes, reflecting the most ‘direct’ thalamic inputs, were equivalent for observed responses to multi-whisker stimuli and for the linear predictor, consistent with the idea that sublinearity originates as a product of cortical processing.
Further support for this view comes from Shosaku (Shosaku, 1986), who investigated the functional organization of the recurrent inhibitory action of neurons in the somatosensory part of the thalamic reticular nucleus (Rt) on VPM neurons in rats. The spontaneous activities of vibrissa-responding neurons in VPM and Rt were simultaneously recorded and analyzed with cross-correlation techniques. The analysis revealed that both inhibitory and excitatory interaction between Rt and VPM are restricted to neurons with receptive fields on the same vibrissa. Rt neurons associated with one whisker do not appear to inhibit the VPM neurons with a different principal whisker. Therefore, intra-thalamic inhibitory circuitry is not likely to account for the sublinear multi-whisker interactions reported here.
Even though we argue — based upon the linearity of the early (0–7 ms after stimulus onset) cortical response — that the initial thalamocortical volley is generated by linear summation, we cannot rule out the possibility that later thalamic activity might reflect sublinear summation at subcortical levels. Projections from the barrel cortex to the thalamic reticular nucleus are somatotopically arranged: axons originating in infragranular layers of one barrel-column terminate precisely in VPM along the entire whisker arc representation (Hoogland et al., 1987; Bourassa et al., 1995). In contrast collaterals of the same axons terminate in regions of S-RT representing the entire whisker row (Hoogland et al., 1987). These projections could have the function of suppressing a given VPM neuron's response to non-principal whiskers.
Nevertheless, the most important mechanism leading to sublinearity appears to be intracortical inhibition. Inhibitory neurons in layer IV receive convergent input from multiple whiskers, and suppress the nearby excitatory neurons in proportion to the number of whiskers stimulated [reviewed by Brumberg et al. (Brumberg et al., 1996)]. Support for this idea comes from a study on awake behaving rats (Kelly et al., 1999) (also M.K. Kelly, personal communication). Neural activity was recorded while rats were palpating a wire mesh screen with their whiskers. When the whiskers surrounding the principal whisker of the sampled neuron were trimmed off and the animal was again exposed to the experimental apparatus, neurons in the column of the intact whisker showed a marked increase in stimulus-evoked activity, presumably because of the loss of inhibitory influence provoked by removal of adjacent whiskers. A key finding in support of the role of intracortical inhibition is that sublinearity is dramatically reduced by bicuculline methiodide iontophoresis in layer IV (Kyriazi et al., 1996).
Functional Significance of Sublinearity
Single whisker and multi-whisker deflections might have different behavioral relevance for rats. While exploring, contact of any whisker with an object is an important event, and detection of the presence of an object [e.g. in the ‘gap-crossing task’ (Hutson and Masterton, 1986; Harris et al., 1999)] might be enhanced by the absence of inhibition of the afferent signal related to deflection of one or a few whiskers. On the other hand, once an object is detected, a rat is likely to orient its snout, and then to sweep multiple whiskers across the surface of the object (Welker, 1964). During the second phase of tactile exploration, when the sensory system analyzes the contacted object, sublinear summation might play a fundamental role by reducing the total magnitude of the response.
Yet another possible role of sublinear summation could be to reduce the total amplitude of the cortical response evoked by the simultaneous movement of all vibrissae during the whisking cycle. The above possibilities are not mutually exclusive. In both cases the reduction of response magnitude would allow the full dynamic range of response to be available for representation of the features of contacted objects.
To summarize, the available data suggest that with the increase in the number of simultaneously deflected whiskers the overall cortical response tends to be smaller than the sum of the responses to single whiskers. How general is the phenomenon of sublinear summation to multiple simultaneous inputs? Gardner and Costanzo compared the response of single somatosensory cortical neurons in alert rhesus monkeys to brief air-puff stimuli applied either at a single point or simultaneously at several points along of the forearm (Gardner and Costanzo 1980a,b). Simultaneous inputs from three different sites were found to be summated sublinearly.
We are grateful to S. Giannotta and to M. Cesarato for technical contributions. Supported by NIH grant NS32647, Telethon Foundation grant 984, and MURST.
Address correspondence to M.E. Diamond, Cognitive Neuroscience Sector, International School for Advanced Studies, Via Beirut 9, 34013 Trieste, Italy. Email: firstname.lastname@example.org.