The neural mechanisms of roughness perception have been studied in a series of combined psychophysical and neurophysiological experiments (Connor and others 1990
; Connor and Johnson 1992
; Blake and others 1997a
; Yoshioka and others 2001
) that followed the method of multiple working hypotheses and sequential elimination of hypotheses by falsification (Popper 1959
; Platt 1964
). In each study, the surfaces varied over two dimensions. The aim in the first study was to produce psychophysical results across a wide range of surfaces—results that would severely challenge any neural coding hypothesis. The result was that among a dozen broad neural coding categories, eight were eliminated by the consistency test described earlier. The experimental designs in the following three studies were aimed at testing the four remaining possibilities.
The combined result of the four studies is that all hypotheses but one have been rejected in at least one study; most of the more likely hypotheses have been rejected in two or more studies. The single hypothesis that survived was a measure of the spatial variation in the impulse rates of SA1 (Merkel) afferent fibers innervating the skin, which is computed by a simple, well-studied neural mechanism. To be specific, the measure, which will be called “SA1 spatial variation,” is the mean absolute difference in firing rates between SA1 afferent fibers with receptive field centers separated by about 2 mm. The point that is most relevant for this review is that perceived roughness was a linear function of SA1 spatial variation in every study. Because these studies have been reviewed recently (Johnson and others 2000
), they will be described here only briefly.
The skin of the fingerpad is innervated by four types of mechanoreceptive afferent fibers that have distinctly different response properties and serve distinctly different perceptual functions (Johnson and others 2000
; Johnson 2001
). Roughness perception could depend on any of these four types or a combination of them. The four types comprise slowly adapting type 1 (SA1) afferents, rapidly adapting cutaneous (RA) afferents, Pacinian (PC) afferents, and slowly adapting type 2 (SA2) afferents. The SA1 afferents, which innervate the epidermis densely (100/cm2
), terminate within Merkel cells at the base of the deepest epidermal ridges. They resolve the spatial details of tactile stimuli acutely; all the available evidence suggests that they are responsible for both form and texture perception (Johnson and others 2000
). The RA afferents, which also innervate the skin densely (150/cm2
), terminate in Meissner corpuscles at the margin between the dermis and epidermis. The RA afferents have relatively poor spatial resolution, but they are very sensitive to skin movement and they gather information about skin movement from a large skin area (5–25 mm2
). They are responsible for the detection of minute motion on the surface of the skin. Each PC afferent terminates within a single Pacinian corpuscle. PC afferents, which are less numerous than SA1 or RA afferents, number about 2000 in the human hand. PC afferents are sensitive to high-frequency vibration with amplitudes in the nanometer range, and they are therefore responsible for the detection of high-frequency vibrations. The SA2 afferents, which are also much less numerous, are said to terminate in Ruffini corpuscles within the connective tissue matrix of the dermis. The available evidence suggests that they are responsible for the sense of skin stretch and may, therefore, play a critical role in the perception of hand shape.
The response properties of these mechanoreceptors and psychophysical studies provide some clues to the neural mechanisms of roughness perception. Except for the SA2 afferents, all the afferents are very sensitive to skin deformation and respond robustly to textured surfaces scanned across the skin (Phillips and others 1990
). The SA2 afferents are not very sensitive to skin deformation apart from stretch, and they respond very poorly to scanned, textured surfaces (Phillips and others 1990
). No measure of the SA2 afferent responses is consistent with roughness perception (Johnson and others 2000
The constancy of roughness judgments over a very wide range of scanning velocities argues against a temporal mechanism as the basis for roughness perception—that is, some temporal measure of the neural response. A change in scanning velocity from 20 to 70 mm/s, for example, causes the temporal cadence of the stimulus features driving each afferent fiber to change almost fourfold, and it therefore has a large effect on any temporal measure of the neural responses, but it has no effect on roughness judgments (Lederman 1983
). Nonetheless, central neural mechanisms might use information about scanning velocity to compensate for this effect. Vibratory adaptation that reduces magnitude estimates at 20 and 250 Hz (which indicates reduction of the sensitivity of RA and PC afferents) has little or no effect on roughness judgments (Lederman and others 1982
). This argues against a role for RA and PC afferents (however, see Hollins and others 2000b
). None of this circumstantial evidence was used to select or reject neural coding hypotheses in the studies described here.
The first combined psychophysical and neurophysiological study of roughness perception was by Sathian and others (1989)
. They were unable to conclusively eliminate any hypotheses, because their stimulus range was too small. The first study to employ the consistency test was by Connor and others (1990)
, which used dot patterns with varying center-to-center spacings between dots (0.8–6.5 mm) and varying dot diameters (0.5–1.2 mm). Roughness perception was an inverted U-shaped function of dot spacing, which peaked at 3.2 mm; surfaces with dot spacings smaller and greater than 3.2 were perceived as less rough. At each dot spacing, roughness declined with increasing dot diameter (because the dots felt less sharp). The result was three inverted U-shaped functions of dot spacing (). By drawing a horizontal line across the three functions, it can be seen that six surfaces produce the same subjective roughness although they evoke very different neural responses (). The neural measure upon which roughness perception depends must be constant for these six surfaces even though the neural responses differ. This provides a severe test of any neural coding hypothesis.
Fig. 2 Responses of single, typical slowly adapting type 1 (SA1) afferents, rapidly adapting cutaneous (RA) afferents, and Pacinian (PC) afferents to six raised-dot surfaces with 0.5 mm dot diameters. The dot patterns were scanned repeatedly from right to left (more ...)
The consistency test is shown in for nine neural coding measures derived from the study by Connor and others (1990)
. As in all four studies described here, identical surfaces were used in the psychophysical and neurophysiological studies. The surfaces, which ranged from feeling almost glassy smooth to very rough, were scanned repeatedly across the receptive fields of monkey SA1, RA, and PC afferents to obtain a statistically accurate description of the population responses to these surfaces. Identical surfaces were also scanned across the receptive fields of human SA1, SA2, RA, and PC afferents (Phillips and others 1992
). The result of the human studies was that, except for SA2 afferent responses, there were no significant differences between human and monkey neural responses (Johnson and others 2000
Fig. 3 Consistency plots for nine possible neural coding measures. The ordinate of each graph represents the mean subjective roughness judgment for each of the 18 surfaces in the study by Connor and others (1990). The abscissa represents one of the neural response (more ...)
It is evident in that there is no consistent relationship between roughness perception and any measure of mean impulse rate. Spatial and temporal neural coding mechanisms based on PC responses failed because PCs are very sensitive and respond vigorously to smooth and rough surfaces alike (); there was little or no gradation in their spatial or temporal responses and therefore no basis for roughness perception. This rejection of all neural codes based on PC responses is consistent with the study by Lederman and others (1982)
mentioned above, which showed that strong vibratory adaptation (that depressed the responses of PC afferents and reduced the subjective magnitudes of high-frequency stimuli) had no effect on perceived roughness. The most consistent relationships in the study were between spatial and temporal variations in SA1 impulse rates and the roughness judgments (0.98 and 0.97 correlation coefficients, respectively). The relationships were linear as well as consistent, even though nothing in the analysis predisposed the relationships to linearity; the putative neural measures were computed without reference to the psychophysical outcome. Comparable measures of RA impulse rates were more poorly correlated, but they cannot be said to have failed the consistency test in any clear and unambiguous way in this study. This study failed to distinguish between temporal and spatial measures because the temporal and spatial structures of the stimuli did not vary independently.
A second study in this series aimed to distinguish between temporal and spatial coding measures by varying the temporal and spatial properties independently; the surfaces were designed to produce results that could be consistent with either a spatial mechanism or a temporal mechanism but not both (Connor and Johnson 1992
). The result was that subjects’ roughness judgments were positively correlated with temporal measures of variation in firing rates for half the surfaces (those in which dot spacing varied in the scanning direction) and negatively correlated for the other half (those in which dot spacing varied in the direction orthogonal to the scanning direction). Thus, there was no consistent relationship between any measure of temporal variation in the firing of either SA1 or RA afferents and roughness judgments. The same result was obtained for all measures of mean impulse rate, which eliminated measures based on mean impulse rate for the second time. This left spatial variation in either the SA1 or RA afferents as candidate mechanisms.
A third study aimed to distinguish between codes based on SA1 and RA afferent responses by varying dot height (Blake and others 1997a
). Previous studies (Blake and others 1997b
) had shown that RA responses saturate at dot heights greater than about 300 microns, whereas SA1 afferents respond with impulse rates proportional to dot height for dot heights greater than 600 microns (the saturation limit is not known). If roughness perception depends on RA responses, then roughness judgments should be independent of dot height, but they were not. The roughness judgments, like the SA1 responses, were proportional to dot height. The relationship between roughness judgments and SA1 spatial variation was linear, and the correlation was greater than 0.97 as in the previous two studies.
The fourth study was designed as a challenge to the SA1 spatial variation hypothesis (Yoshioka and others 2001
). The SA1 innervation density is about 100 afferents per cm2
of skin area. Many fine surfaces that are very rough (e.g., fine sandpapers) have feature densities much higher than 100 per cm2
. The question was whether a mechanism based on spatial variation in SA1 firing rates can account for the perceived roughness of surfaces whose spatial variation is much finer even than the afferent fiber spacing. This fourth study used 20 gratings with spatial periods ranging from 0.1 to 2.0 mm. As in previous studies, there was no consistent relationship between PC responses and roughness judgments, because the PCs were activated so strongly and uniformly by the fine and coarse gratings alike. SA1 afferents responded to the finely textured surfaces in a graded manner, and the mean absolute difference in SA1 firing rates between afferents (spatial variation) was correlated strongly (0.97) with subjective roughness estimates.
The correlation between SA1 spatial variation and roughness perception in all four studies is shown in and in two forms. The correlation is shown as consistency plots in . The linear correlation is 0.97 or greater for all four sets of data but, as well, the roughness judgments are proportional to SA1 spatial variation. Again, nothing about the analysis predisposed the outcome to linearity; a linear relationship emerged the first time SA1 spatial variation was computed. shows the comparison in a more conventional form. The psychophysical roughness judgments are shown on the left ordinate of each graph; the corresponding SA1 spatial variation is shown on the right ordinate. Because both ordinates are linear, the correspondence implies a linear relationship between SA1 spatial variation and roughness perception.
Fig. 4 Consistency plots of slowly adapting type 1 (SA1) spatial variation versus subjective roughness from four studies with different stimulus patterns. The ordinate in each graph is the mean subjective report across all subjects in a single study. The abscissa (more ...)
Fig. 5 Roughness magnitude and spatial variation in slowly adapting type 1 (SAI) firing rates from four studies (see text). The left ordinate and filled circles in each graph represent mean roughness judgments for individual surfaces. The right ordinate and (more ...)
The mean absolute difference in firing rates between SA1 afferents with receptive field centers separated by 2 to 3 mm may seem abstract, but it corresponds, in fact, to a simple neurophysiological mechanism (Connor and others 1990
; Yoshioka and others 2001
). Every neuron within the central nervous system whose receptive field includes regions of inhibition and excitation (which is, as far as is known, virtually all neurons in somatosensory cortex) computes a measure of the spatial variation in skin deformation. More precisely, the neuron’s discharge rate is proportional to the difference in discharge rates between afferents arising from the excitatory and inhibitory receptive field subregions. The principal importance of the mechanism is that it confers selectivity for particular stimulus features and their orientations. But the summed discharge rate of a population of such neurons can form the basis for roughness perception. In fact, neurons with exactly the properties hypothesized to account for roughness perception have been demonstrated in somatosensory cortex (DiCarlo and Johnson 2000
). Such a mechanism has several things to recommend it. Like roughness perception, it is unidimensional and it is affected only secondarily by factors such as scanning velocity (DiCarlo and Johnson 1999
). The combined psychophysical studies described above and the existence of neurons with the hypothesized properties suggest that roughness perception is based on the mean firing rate of a population of cortical neurons that compute SA1 spatial variation.