Title: Single Tactile Afferents Outperform Human Subjects in a Vibrotactile Intensity

43 44 We simultaneously compared the sensitivity of single primary afferent neurons supplying 45 the glabrous skin of the hand and the psychophysical amplitude discrimination thresholds in 46 human subjects for a set of vibrotactile stimuli delivered to the receptive field. All recorded 47 afferents had a dynamic range narrower than the range of amplitudes across which the 48 subjects could discriminate. However, when the vibration amplitude was chosen to be 49 within the steepest part of the afferent’s stimulus-response function, the response of single 50 afferents, defined as the spike count over the vibration duration (500 ms), was often more 51 sensitive in discriminating vibration amplitude than was the perceptual judgment of the 52 participants. We quantified how the neuronal performance depended on the integration 53 window: for short windows the neuronal performance was inferior to the performance of 54 the subject. The neuronal performance progressively improved with increasing the spike 55 count duration and reached a level significantly above that of the subjects when the 56 integration window was 250 ms or longer. The superiority in performance of individual 57 neurons over observers could reflect a non-optimal integration window or be due to the 58 presence of noise between the sensory periphery and the cortical decision stage. 59 Additionally, it could indicate that the range of perceptual sensitivity comes at the cost of 60 discrimination through pooling across neurons with different response functions. 61 62 63 Introduction 64 65 The somatosensory system offers unique opportunities for making direct recordings of 66 peripheral neurons while concurrently obtaining perceptual judgments from awake, 67 neurologically normal human participants; this microneurography technique allows 68 recording of single impulses with percutaneously inserted tungsten microelectrodes (Vallbo 69 and Hagbarth, 1968). 70 71 Classic experiments on tactile sensitivity have identified a clear relationship between the 72 psychophysical performance of humans and the physiological properties of sensory 73 afferents (Werner and Mountcastle, 1965; Talbot et al., 1968; Harrington and Merzenich, 74 1970; Johansson and Vallbo, 1979a; but see Knibestöl and Vallbo, 1980). A high correlation 75 was found between the absolute detection threshold of the participant and that of the most 76 sensitive tactile afferents (Johansson and Vallbo, 1979a). Similar findings in other sensory 77 systems (Hawken and Parker, 1990; Vogels and Orban, 1990) suggest a ‘lower envelope 78 principle’, whereby the perceptual detection threshold is set by the most sensitive neurons 79 available (Parker and Newsome, 1998). 80 81 When responding to stimuli near the lowest end of detectable intensities, pooling the 82 activity of multiple neurons effectively results in the ’read out‘ of the most sensitive 83 neurons. While this pooling strategy may be optimal for stimulus detection, it may not be as 84 effective for discriminating between two stimuli at higher intensities. Figure 1 schematically 85 illustrates the difference that would arise from pooling across two neurons. In this example, 86 one neuron with a low threshold (gray curve) responds differentially to two low intensity 87 stimuli (L1 and L2), but responds equally strongly to two high intensity stimuli (H1 and H2). 88 The other neuron with a higher threshold (black curve) does not respond to L1 and L2, but 89 does respond differentially to H1 and H2. Participants attempting to discriminate between 90 these stimuli would be most efficient if they could identify the most appropriate neuron for 91 a given intensity and base the perceptual judgement on the activity of that neuron. Pooling 92 across both neurons would achieve equivalent performance when discriminating between 93 the low intensity stimuli because only the more sensitive neuron responds to those stimuli. 94 But pooling would reduce discrimination sensitivity between the higher intensity stimuli 95 because it would add uninformative input from the non-discriminating neuron. 96 97 The foregoing discussion indicates that pooling across peripheral inputs can reduce 98 perceptual sensitivity. However, other evidence indicates that pooling can lead to 99 perceptual sensitivity that is superior to that of any single peripheral unit. In a Vernier 100 judgement, human observers are able to discriminate the spatial offset between two line 101 segments when the offset is 5 arcsec, which is five to ten times smaller than the maximum 102 spatial resolution of individual photoreceptors (Westheimer, 1981). This ‘hyperacuity’ can 103 only be achieved by advantageous pooling across peripheral inputs. In summary, it is 104 difficult to provide any a priori estimation of perceptual sensitivity to a stimulus from the 105 sensitivity of individual peripheral neurons. Here, we compared the accuracy of perceptual 106 judgements by human subjects with the performance of their simultaneously recorded 107 single tactile afferents in a vibration amplitude discrimination task across a range of stimulus 108 amplitudes. 109 110 111 Methods 112 Subjects and recording procedure 113 In eight experiments, six healthy human subjects (five males and one female) participated 114 after providing written informed consent in accordance with the Declaration of Helsinki. 115 Three of the subjects were authors; the other three were naive to the purpose of the 116 experiment. Subjects sat comfortably in a dentist's chair with their right upper arm abducted 117 at ~30°, while their elbow rested on a horizontal extension of the chair. The arm was 118 immobilized by straps around the wrist. To stabilize the distal phalanges, the dorsal aspect 119 of the index, middle, and ring fingers was fixed into a plasticine mold. A tungsten 120 microelectrode was inserted into a cutaneous fascicle of the median nerve at the wrist and 121 neural activity amplified (2x10, 0.3-5kHz; ISO-80, World Precision Instruments, USA); an 122 uninsulated subdermal microelectrode served as the reference. Impulses were recorded 123 from single afferents that terminated in the palm, index, middle or ring fingers. For each 124 isolated fiber, calibrated nylon monofilaments (Semmes-Weinstein Esthesiometers, 125 Stoelting, USA) were used to determine the afferent's threshold force and to define the 126 receptive field. Neuronal activity was sampled continuously at 12.8 kHz and spikes were 127 sorted offline with a template matching protocol written in MATLAB (MathWorks, Natick, 128 MA). Microelectrode recordings from the median nerve have revealed four classes of 129 myelinated low-threshold mechanoreceptor (Johansson and Vallbo, 1983; Macefield and 130 Birznieks, 2008). 131 Single fibres were classified as slowly adapting type I or II (SA-I or SA-II), and fast adapting 132 type I or II (FA-I, or FA-II) by the established criteria of responses to static stimuli, responses 133 to rapidly changing stimuli, and receptive field size (Johansson and Vallbo, 1979b, 1983; 134 Vallbo and Johansson, 1984). Impulses were recorded from a total of 40 single tactile 135 afferents during 8 recording sessions. These included 21 slowly adapting type I (SA-I), 5 136 slowly adapting type II (SA-II), 11 fast adapting type I (FA-I), and 3 fast adapting type II (FA-II) 137 neurons. Out of these, 32 single afferents were successfully maintained throughout all 138 phases of the experiment (see below). 139 140 Characterizing neuronal response function 141 After classifying the afferent and determining its receptive field, the tip of a rod (1 mm in 142 diameter), attached to a custom made mechanical stimulator, was placed in the center of 143 the receptive field. The rod was used to present sinusoidal vibrations spanning a range of 144 amplitudes while we measured the afferent’s response. Each set of vibratory stimuli lasted 145 for 500 ms with a 500 ms interval between vibrations. Figure 1B illustrates the experimental 146 set-up. Stimuli were generated in MATLAB and played via a National Instruments (Austin, 147 TX) interface board. To estimate the amplitude response function, a frequency was selected 148 to drive the afferent effectively (between 10 and 60 Hz). The amplitude of the sinusoidal 149 stimulus systematically increased from zero to 60 μm with steps of 0.6 μm. We observed 150 the afferent response as the vibration amplitude progressively increased. This allowed us to 151 estimate the minimum amplitude that generated spiking in the neuron, and the maximum 152 amplitude for which spiking seemed to plateau. We selected one or more base amplitudes 153 (always multiples of 6 μm) that fell within the estimated minimum and maximum 154 amplitudes, and used these for the simultaneous psychophysical and neuronal 155 discrimination task. If the neuron responded to the smallest vibration then base amplitude 156 was set to zero. The amplitude response function was then saved for offline analysis to 157 verify the choice of the base amplitude relative to the dynamic range of the neuronal 158 response. The afferent response was characterized offline by fitting a piecewise linear 159 function to each amplitude response function (Figure 2). With the exception of one afferent, 160 all recordings were fitted well by the piecewise linear function, revealing the three distinct 161 response regions: the sub-threshold, the saturated, and the dynamic range. The offline 162 analyses showed that, out of the 64 selected base amplitudes, 52 were in the dynamic range 163 of the neuron. These sessions are analyzed to provide a direct comparison of psychophysical 164 and neuronal discrimination performances. 165 166 Simultaneous psychophysical and neuronal discrimination task 167 We used an adaptive staircase procedure (Kontsevich and Tyler, 1999), within a 2-interval, 168 2-alternative forced choice (2AFC) paradigm to measure each subject’s Just-Noticeable169 Difference (JND) for 500-ms sinusoidal vibrations presented to the center of the receptive 170 field of the simultaneously recorded neuron. Each trial comprised two 1-s intervals, each 171 marked by an auditory cue (Figure 1C). Subjects made a forced-choice judgment to indicate 172 which interval contained the stronger of two tactile stimuli. Using this 2AFC paradigm, we 173 measured each subject’s JNDs for vibrations of different base amplitudes. The amplitude of 174 the stronger vibration varied from trial-to-trial according to a Bayesian adaptive-staircase 175 method that optimized the information gain (in terms of measurement of the JND) on each 176 trial (Kontsevich and Tyler, 1999). The order of the base and higher-amplitude vibrations 177 varied randomly from trial-to-trial, and the JND was estimated at the end of each 30-trial 178 staircase. Subjects did not receive feedback on their responses. 179 180 181 Results 182 Figure 2A shows typical examples of the amplitude response function for a fast adapting (n1) 183 and a slowly adapting neuron (n2). Neurons showed a characteristic response profile 184 consisting of (i) a sub-threshold range for which they did not fire any spikes, (ii) a range of 185 amplitudes across which the response increased monotonically (we refer to this as the 186 neuron’s “dynamic range”), and (iii) a saturated range over which the response was 187 constant. In a few highly sensitive neurons, the response threshold was low and no distinct 188 sub-threshold range was identified while in a few other neurons the dynamic range 189 extended beyond the maximum amplitude (60 μm) and the saturated range was not 190 observed. In order to see how the full range of amplitudes was represented in the collective 191 response of afferents, we averaged their activity (n = 32). The average activity (Figure 2B) 192 was a linear function of amplitude, and covered the whole range of applied amplitudes (R 193 of the regression was 0.99). To quantify the response function of individual afferents, we 194 fitted a piecewise linear function to each amplitude response function, revealing the three 195 distinct response regions: the sub-threshold, the saturated, and the dynamic range. Figure 196 2C plots for each afferent the length of the dynamic range versus its slope. All recorded 197 afferents had a dynamic range narrower than that of the average response. Across afferents 198 the average slope was 0.41, while the slope of the average response was 0.17. 199 200 Figure 3 illustrates the difference in the number of spikes fired by the recorded afferents in 201 response to the two stimuli in each trial as a function of the amplitude difference between 202 the two stimuli. Dots in the upper-right or lower-left quadrants indicate trials in which the 203 neuronal response co-varied with vibration amplitude (i.e., a higher number of spikes for 204 the higher amplitude stimulus). For these trials, a hypothetical observer of the number of 205 spikes fired by this neuron would discriminate correctly between the two stimuli. On the 206 other hand, dots in the upper-left or lower-right quadrants indicate trials in which neuronal 207 response was lower for the higher amplitude stimulus. A decision based on the number of 208 spikes fired by such a neuron would therefore lead to an incorrect discrimination. Dots 209 falling on the x-axis indicate trials for which the neuronal response was identical for the 210 higher and lower amplitude stimuli. A decision based on the neuronal response in these 211 trials would lead to chance performance (50% correct). The subjects’ performance can be 212 assessed from the number of correct trials (black open circles) and incorrect trials (gray 213 filled circles). Performance was then compared to that of a hypothetical observer making a 214 decision on each trial based on the spike count of the simultaneously recorded neuron. For 215 the two examples illustrated in Figure 3, both subjects performed at 83.3% correct (accuracy 216 was similar between subjects because the staircase titrated the task difficulty across trials so 217 that each subject performed at about this level). The concurrent neuronal performance was 218 at 96.7% and 83.3%, respectively. 219 220 Figure 4A compares the performance of human subjects with those of single neurons across 221 all 52 recorded staircases where the base amplitude was selected within the dynamic range 222 of the neuron. This figure demonstrates that on the majority of cases the single neuron 223 outperformed the subject. Figure 4B shows the distribution of performances for subjects 224 and recorded neurons across the four afferent classes. The median performance was 93.3% 225 for neurons and 83.3% for subjects. A Wilcoxon signed rank test on the difference in 226 performance between neuron and subject on each of the 52 staircases showed that the 227 performance of the sample of individual neurons was significantly higher than the 228 corresponding performance of the human subjects (p < 0.001). 229 230 Our analyses up to here focused on the spike count measured across the whole vibration 231 interval. This is based on the assumption that the ideal observer can access the spike count 232 over the 500 ms vibration duration and decode a single spike difference between the counts 233 generated for each of the two vibrations. Previous research has indicated that the spike 234 count generated over a subsection of the vibration might be a more reliable predictor of 235 perceptual discrimination (Luna et al., 2005). Figure 5 quantifies how in our data set the 236 decoding performance depends on the integration time, and the precision with which the 237 decoder detects differences in spike counts across the two vibrations. Figure 5A illustrates 238 that for short integration windows the neuronal performance is inferior to the performance 239 of the subject. The average neuronal performance progressively improves as the spike count 240 duration increases and reaches a level significantly above that of the subjects (dashed 241 horizontal line) when the duration is 250 ms or longer. Figure 5B illustrates that the 242 superiority in neuronal performance over subjects critically depends on the precision with 243 which spike counts can be compared. When the ideal observer of the neuronal response can 244 detect a single spike difference across the two vibrations, it outperforms the subjects. As the 245 decoding precision decreases, the average neuronal performance rapidly drops to below 246 that of the subjects (dashed horizontal line in Figure 5B). 247 248 249 250 Discussion 251 252 We compared the sensitivity of single afferent fibres with the concurrently recorded 253 psychophysical performance of human subjects in a vibrotactile amplitude discrimination 254 task. The dynamic range for every afferent was narrower than the range across which 255 subjects could discriminate vibration amplitudes. However, when the base amplitude was 256 chosen to be within the afferent’s dynamic range, the spike count in individual neurons 257 could differentiate the amplitude significantly better than the human subjects could do 258 perceptually. We quantified how the superiority of neuronal performance critically depends 259 on the ability to integrate spikes over multiple cycles of the vibration (Figure 5A). 260 261 The superiority in performance of individual neurons over observers could indicate that the 262 range of perceptual sensitivity comes at the cost of discrimination through pooling across 263 neurons with different response functions. Furthermore, noise could be introduced 264 between the sensory periphery and the cortical decision stage. We quantified how a small 265 amount of noise added during synaptic transmission to the cortical decision areas can 266 reduce the single neuron information to levels compatible with human performance (Figure 267 5B). Consistent with this idea, previous simultaneous recordings of first-order and cortical 268 neurons in rat somatosensory system (barrel cortex) revealed that first-order neurons 269 carried more information about the kinetics of vibrotactile stimuli applied to whiskers than 270 did cortical neurons (Arabzadeh et al., 2005, 2006). 271 272 Previous experiments have described neuronal input-output functions with either a piece273 wise linear function or an S-shaped function, where input is the strength of the sensory 274 event (e.g. vibration amplitude) and output is spiking probability (Johansson and Vallbo 275 1979a; Knibestöl and Vallbo, 1980). The common finding that appears to generalize across 276 species is that neuronal firing rate increases as the amplitude of a vibrotactile stimulus 277 increases – for observations in primates see (Hernández et al., 2000; Luna et al., 2005; 278 Harvey et al., 2013) for rats see (Arabzadeh et al., 2003; Adibi and Arabzadeh, 2011). 279 Neurons recorded in the current study showed a response profile that was well 280 approximated by a piece-wise linear function but with variable thresholds (Figure 2). 281 Nonetheless, single neurons recorded across different subjects consistently outperformed 282 those same subjects in the discrimination task. 283 284 While previous neurophysiological experiments have indicated that the properties of the 285 peripheral sense organs determine the psychophysical threshold (Hecht et al., 1942), others 286 have argued that central mechanisms can limit detection sensitivity (Green and Swets, 287 1966). Focusing on tactile detection thresholds, Johansson and Vallbo (1979a) found a good 288 match between neuronal and psychophysical thresholds if the analyses were restricted to 289 the FA-I afferents. Similar results were found in a few cases with Pacinian corpuscles, but 290 not with any slowly adapting unit. Based on the expected density of FA-I afferents, their 291 average receptive field size, and the distribution of their thresholds, an argument is made 292 that a single impulse in a single unit could be enough to produce the sensation of touch 293 (Johansson and Vallbo 1979a,b). The estimates also indicated that the number of units 294 excited by stimuli at the minimal psychophysical thresholds is small. Indeed, using 295 intraneural microstimulation of single afferents innervating the human hand, Vallbo et al. 296 (1984) and Macefield et al. (1990) showed that activation of single SA-I, FA-I and FA-II 297 afferents evoked conscious experience (whereas activation of single SA-II and muscle 298 spindle afferents did not). Overall, these findings indicate an efficient read out mechanism 299 of afferent activity for stimuli close to detection threshold. 300 301 In order to determine how many sensory neurons are required to match the psychophysical 302 sensitivity to thermal changes, Darian-Smith and colleagues compared the temperature 303 sensitivity of afferents in anesthetized monkeys with that of human subjects (Darian-Smith 304 et al., 1973). The results indicated that the sensitivity of a single temperature sensitive 305 afferent is on average less than the psychophysical sensitivity of human subjects, and 16 306 afferents were required to achieve the perceptual temperature sensitivity. Similar findings 307 have been reported in higher sensory areas. In the visual system, Hawken and Parker (1990) 308 compared the psychophysical detection threshold of spatial contrast patterns in human with 309 the neuronal detection function of monkey V1 neurons. The slope of the neuronal detection 310 function correlated closely with that of the psychophysical detection function. Single 311 neurons in macaque somatosensory cortex exhibited orientation tuning with a degree of 312 sensitivity comparable to that measured in humans (Bensmaia et al., 2008). Similarly, in the 313 primary visual cortex neuronal discrimination thresholds for orientation are comparable 314 with monkeys’ psychophysical performance (Vogels and Orban, 1990). Recording from 315 single direction-selective neurons in the middle temporal (MT) and medial superior 316 temporal areas (MST) found trial-to-trial correlation between fluctuations in neural 317 responses and the perceptual judgment suggesting that performance was based on signals 318 pooled across a population of neurons (Newsome et al., 1990; Britten et al., 1996; Shadlen 319 et al., 1996). In the tactile domain, motion discrimination experiments involving moving 320 gratings and plaids presented to the monkey fingertips revealed populations of 321 somatosensory cortical neurons that exhibited motion integration properties similar to 322 neurons in visual area MT, with performances that matched ones obtained in human 323 psychophysics (Pei et al., 2010, 2011). Finally, recordings from primary and secondary 324 somatosensory cortices have shown correlation between neuronal activity and the 325 monkeys’ vibrotactile discrimination performance (Romo and Salinas, 2003; Romo et al., 326 2003). 327 328 The current experiment explored the relationship between neuronal and psychophysical 329 performance for stimuli that are well above detection threshold. There is substantial 330 evidence that stimulus intensity is represented in a neuronal population code where 331 different afferent types contribute with different weights (Muniak et al., 2007; Bensmaia, 332 2008). The fact that our subjects performed worse than their peripheral neurons is 333 consistent with the notion of an intensity code based on weighted pooling of the afferent 334 inputs. This type of intensity code ensures a continuum of amplitude perception for a wide 335 range of stimuli, but it reduces relative sensitivity (discrimination capacity) by including 336 ‘noise’ pooled from sensory units that do not discriminate between the stimuli being 337 compared (see Figure 1A). However, this conclusion is undermined by the observation that 338 the peripheral sensory neurons we and others have recorded show little variation in their 339 response to stimuli outside their dynamic range. 340 341 Another way that discrimination sensitivity could be reduced due to pooling across afferent342inputs is if the total input were subjected to a logarithmic compression at the central level.343This is in keeping with the Weber-Fechner Law, which states that perceptual discrimination344is based on the relative difference in neuronal responses, as a fraction of the mean response345to the stimuli. From classic psychophysical experiments with humans, various forms of346nonlinearity have been inferred in the amplitude response function (Knibestöl and Vallbo,3471980). These nonlinearities shift from accelerating at low vibration amplitudes to348decelerating at higher vibration amplitudes (Arabzadeh et al., 2008). Electrophysiological349recordings in monkeys reveal that many neurons in somatosensory cortex show350 decelerating response rates with increasing stimulus amplitude (Mountcastle et al., 1969). A351physiological rationale for this encoding principle is that it expands the psychophysical352dynamic range and filters biologically insignificant stimulus details. However, a better353understanding of this principle will depend upon identifying what the relevant intensity354code is and which afferent input it is based on. We conclude that, while stimulus detection355thresholds in human fingertip may be compatible with the lower envelope principle (Parker356and Newsome, 1998), amplitude discrimination displays characteristic features of pooled357coding (Bensmaia, 2008).358359360References361 Adibi M, Arabzadeh E. A comparison of neuronal and behavioral detection and362discrimination performances in rat whisker system. J. Neurophysiol. 105: 356–65, 2011.363 Arabzadeh E, Clifford CWG, Harris JA. Vision merges with touch in a purely tactile364discrimination. Psychol. Sci. 19: 635–41, 2008.365 Arabzadeh E, Panzeri S, Diamond ME. Deciphering the spike train of a sensory neuron:366counts and temporal patterns in the rat whisker pathway. J. Neurosci. 26: 9216–26, 2006.367 Arabzadeh E, Petersen RS, Diamond ME. Encoding of whisker vibration by rat barrel cortex368neurons: implications for texture discrimination. J. Neurosci. 23: 9146–54, 2003.369 Arabzadeh E, Zorzin E, Diamond ME. Neuronal encoding of texture in the whisker sensory370pathway. PLoS Biol. 3: e17, 2005.371 Bensmaia SJ, Denchev P V, Dammann JF, Craig JC, Hsiao SS. The representation of stimulus372orientation in the early stages of somatosensory processing. J. Neurosci. 28: 776–86, 2008.373 Bensmaia SJ. Tactile intensity and population codes. Behav. Brain Res. 190: 165–73, 2008.374 Britten K, Newsome W, Shadlen M, Celebrini S, Movshon J. A relationship between375behavioral choice and the visual responses of neurons in macaque MT. Vis. Neurosci. 13:37687–100, 1996.377 Darian-Smith IAN, Johnson KO, Dykes R. “ Cold” fiber population innervating palmar and378digital skin of the monkey: responses to cooling pulses. J. Neurophysiol. 36: 325, 1973.379 Green D, Swets J. Signal detection theory and psychophysics. Wiley New York. 1966.380 Harrington T, Merzenich MM. Neural coding in the sense of touch: human sensations of381skin indentation compared with the responses of slowly adapting mechanoreceptive382afferents innervating the hairy skin of monkeys. Exp. Brain Res. 10: 251–264, 1970.383 Harvey M a, Saal HP, Dammann JF, Bensmaia SJ. Multiplexing stimulus information through384rate and temporal codes in primate somatosensory cortex. PLoS Biol. 11: e1001558, 2013.385 Hawken M, Parker A. Detection and discrimination mechanisms in the striate cortex of the386old-world monkey. Vis. Coding Effic. 1990.387 Hecht S, Shlaer S, Pirenne MH. Energy, quanta, and vision. J. Gen. Physiol. 25: 819, 1942.388 Hernández A, Zainos A, Romo R. Neuronal correlates of sensory discrimination in the389somatosensory cortex. Proc. Natl. Acad. Sci. U. S. A. 97: 6191–6, 2000.390 Johansson RS, Vallbo ÅB. Detection of tactile stimuli. Thresholds of afferent units related to391ps ychophysical thresholds in the human hand. J Physiol, 297:405-422. 1979a.392 Johansson RS, Vallbo ÅB. Tactile sensibility in the human hand: relative and absolute393densities of four types of mechanoreceptive units in glabrous skin. J Physiol 286:283-300.3941979b.395 Johansson RS, Vallbo ÅB. Tactile sensory coding in the glabrous skin of the human hand.396Trends in Neurosci 6:27-31. 1983.397 Knibestöl M, Vallbo Å. Intensity of sensation related to activity of slowly adapting398mechanoreceptive units in the human hand [Online]. J. Physiol. 1980.399 Kontsevich LL, Tyler CW. Bayesian adaptive estimation of psychometric slope and threshold.400Vision Res. 39: 2729–2737, 1999.401 Luna R, Hernández A, Brody CD, Romo R. Neural codes for perceptual discrimination in402primary somatosensory cortex. Nat Neurosci 8: 1210–1219, 2005.403 Macefield VG, Birznieks I. Cutaneous mechanoreceptors, functional behaviour. In:404Windhorst U, Binder M & Hirokawa N (Editors): Encyclopedia of Neuroscience, Springer,405Heidelberg, Germany, 2008. pp 914-922406407Macefield G, Gandevia S, Burke D. Perceptual responses to microstimulation of single408afferents innervating joints, muscles and skin of the human hand. J. Physiol. : 113–129,4091990.410 Mountcastle V, Talbot W, Sakata H, Hyvarinen J. Cortical neuronal mechanisms in flutter-411vibration studied in unanesthetized monkeys. Neuronal periodicity and frequency412discrimination. J Neurophysiol 32: 452–484, 1969.413 Muniak MA, Ray S, Hsiao SS, Dammann JF, Bensmaia SJ. The neural coding of stimulus414intensity: linking the population response of mechanoreceptive afferents with415psychophysical behavior. J. Neurosci. 27: 11687, 2007.416 Newsome W, Britten K, Salzman C, Movshon J. Neuronal mechanisms of motion417perception. 1990, p. 697–705.418 Parker a J, Newsome WT. Sense and the single neuron: probing the physiology of419perception. Annu. Rev. Neurosci. 21: 227–77, 1998.420 Pei Y-C, Hsiao SS, Craig JC, Bensmaia SJ. Shape invariant coding of motion direction in421somatosensory cortex. PLoS Biol. 8: e1000305, 2010.422 Pei Y-C, Hsiao SS, Craig JC, Bensmaia SJ. Neural mechanisms of tactile motion integration in423somatosensory cortex. Neuron 69: 536–47, 2011.424 Romo R, Hernández A, Zainos A, Salinas E. Correlated neuronal discharges that increase425coding efficiency during perceptual discrimination. Neuron 38: 649–57, 2003.426 Romo R, Salinas E. Flutter discrimination: neural codes, perception, memory and decision427making. Nat. Rev. Neurosci. 4: 203–18, 2003.428 Shadlen M, Britten K, Newsome W, Movshon J. A computational analysis of the429relationship between neuronal and behavioral responses to visual motion. J. Neurosci. 16:4301486, 1996.431 Talbot WH, Darian-Smith I, Kornhuber HH, Mountcastle VB. The sense of flutter-vibration:432comparison of the human capacity with response patterns of mechanoreceptive afferents433from the monkey hand. J. Neurophysiol. 31: 301, 1968.434 Vallbo Å, Hagbarth KEK. Activity from skin mechanoreceptors recorded percutaneously in435awake human subjects. Exp. Neurol. 21: 270–289, 1968.436 Vallbo Å, Johansson R. Properties of cutaneous mechanoreceptors in the human hand437related to touch sensation. Hum Neurobiol : 3–14, 1984.438 Vogels R, Orban G. How well do response changes of striate neurons signal differences in439orientation: a study in the discriminating monkey. J. Neurosci. 10: 3543, 1990.440 Werner G, Mountcastle VB. Neural Activity in Mechanoreceptive Cutaneous Afferents:441Stimulus-Response Relations, Weber Functions, and Information Transmission. J.442Neurophysiol. 28: 359–97, 1965.443 Westheimer G. Progress in Sensory Physiology. Springer Berlin Heidelberg. 1981444