1. We have studied how neurons of primary somatic sensory (Sl) cortex encode the direction and speed of moving tactile stimuli delivered to the glabrous skin of the contralateral hand. 2. From a total of 178 neurons recorded in SI cortex of 3 awake Macaca mulatta monkeys. 103 were selected for quantitative analysis. Forty-six neurons had slowly adapting (SA) responses. 43 quickly adapting (QA), and 14 mixed SA-QA properties. All possessed cutaneous receptive fields on the distal segments of digits 2, 3, or 4. Receptive fields were scanned with a metal probe (2 mm diam, hemispheric tip) in four different directions (0, 90, 180, and 270 degrees), over a fixed traverse distance of 6 mm, at a variety of speeds (4-100 mm/s). and with a static normal force of 20 g by means of a computer-controlled tactile stimulator. 3. Most neurons gave statistically significant differences in mean impulse rate during the moving stimuli (Wilcoxon, P < 0.01), in at least one of the four directions, compared with the control (nonstimulus) period. The Kruskal-Wallis test(P < 0.01) and the direction index (DI > 35%) determined that about one- of SI cortical neurons showed significant differences in mean impulse rates associated with the direction of the stimuli at the speeds of 23, 50, and 100 mm/s. and about one-third at 4 mm/s. 4. We determined how the temporal covariance of the neural activity was associated with the parameters of the moving stimuli by calculating the coefficients of the Karhunen-Loeve (KL) transform for each set of stimulus responses. Decomposition of the neural activity into principal components indicated that similar to 85% of the impulse train variance during the stimulus responses was contained in the 1st 10 coefficients of the KL transform for the speeds of 23, 50, and 100 mm/s, and similar to 75% at the speed of 4 mm/s. The line spectra calculated from the coefficients of the KL transform showed that the variance contained in the impulse trains in about one-half of the neurons is related to the stimuli. 5. We investigated how the temporal covariance of the neuronal activity was correlated with the direction of the stimulus, by fitting the first coefficient of the KL transform to a weighting function model. This analysis showed that the first coefficient of the KL transform varied as an orderly function of the direction of the moving stimuli. The percentage of neurons that fitted this model (R(2) > 0.7, P < 0.001), as a function of the speed are the following: 18% (11 of 61) for 4 mm/s, 30.5% (29 of 95) for 23 mm/s, 33.7% (27 of 80) for 50 mm/s. and 25% (14 of 56) for 100 mm/s. None of these neurons showed a correlation between the first coefficient if the KL transform and the mean impulse rates, indicating that the variance contained in this coefficient is exclusively related with tile temporal covariance induced by the stimuli. Very few neurons, by contrast, presented this orderly variation, when the mean impulse frequencies were fitted to the weighting function model. 5 (8%) for 4 mm/s, 9 (9.5%) for 23 mm/s, 5 (6.3%) for 50 mm/s, and 7 (12.5%) for 100 mm/s. 6. More neurons of area 3b than area 1 fitted the first coefficient of the KL transform to the weighting function model (62.1% for area 3b and 37.9% for area 1, when the stimulus moved at 23 mm/s), with a higher percentage for SAs in both areas 3b (55.5% for SAs, 27.8% for QAs, and 16.7% for SAs-QAs) and 1 (63.6% for SAs and 36.4% for QAs). 7. A vector-averaging approach was applied to the set of neurons that fitted the first coefficient of the KL transform to the weighting function model (Georgopoulos et al. 1988, 1993). The hypothesis tested was that the resulting population vector is congruent with the direction of the moving stimuli. Indeed, a correlation analysis showed that the population vector is highly correlated with the direction of the moving stimulus (with circular correlation coefficients of 0.769, 0.955, 0.928, and 0.931 for 4, 23, 50, and 100 mm/s, respectively). The analysis of variability of the resulting population vector showed the following confidence intervals, averaged for eight directions as a function of the speed: 45.54 degrees for 4 mm/s, 16.04 degrees for 23 mm/s, 17.61 degrees for 50 mm/s. and 25.75 degrees for 100 mm/s. It was also shown that the magnitude of the population vector is modulated by the stimulus speed. Very poor correlation coefficients were obtained for the population vectors calculated from the small number of neurons for which mean frequency fitted the weighting function model. 8. A Monte Carlo simulation estimated the number of neurons needed in the vector averaging model (with about an order of magnitude of 10(-3)), to obtain a population vector that is congruent with the direction of the stimulus. This simulation showed that the angle between the resulting population vector and the stimulus direction drops asymptotically to zero as the size of the population increases. Also, the same asymptotic behavior was obtained for the confidence intervals. 9. The results obtained in this study suggest that the direction of a moving tactile stimuli can be represented in the neuronal activity of SI cortex. This representation is in the form of a neuronal population vector, whose magnitude is modulated by the speed. We propose that this dynamic internal representation in SI cortex is necessary for the higher order processing of the tactile stimuli.
