The Vision of Hsiao on Somatosensation

Articles in PresS. J Neurophysiol (November 12, 2014). doi:10.1152/jn.00670.2014
1
2
3
The Vision of Hsiao on Somatosensation
4
Martha Flanders and John F. Soechting
5
Laboratory of Neurophysiology, University of Minnesota, Minneapolis, MN 55455
6
7
8
9
10
11
Running title: Research contributions of S.S. Hsiao
12
Keywords: somatosensory, haptics, touch, Pacinian corpuscle, neuroprosthetics
13
14
15
16
Corresponding author:
17
18
19
20
21
22
23
24
25
26
Martha Flanders
6-145 Jackson Hall
321 Church St SE
Minneapolis, MN 55455
Phone (612) 624-6601
Fax (612) 626-5009
[email protected]
1
Copyright © 2014 by the American Physiological Society.
27
ABSTRACT
28
29
The goal of this review is to start to consolidate and distill the substantial body of
30
research that comprises the published work of the late Professor Steven S. Hsiao. The
31
studies of Hsiao began by demonstrating the receptive field properties of somatosensory
32
neurons, progressed to describing cortical feature selectivity, and then eventually elevated
33
the field to hopes of tapping into natural neural codes with artificial somatosensory
34
feedback. With ongoing analogies to contemporaneous studies in visual neuroscience,
35
the research results and writings of Hsiao have provided the fields of haptics and
36
somatosensory neurophysiology with the conceptual tools needed to allow profound
37
progress. Specifically, Hsiao suggested that slowly-adapting tactile form perception
38
could be restored with cortical microstimulation, rapidly-adapting slip reflexes should be
39
relegated to low-level, hard-wired prosthetic components, and Pacinian-corpuscle
40
spatiotemporal population responses could potentially be decoded/encoded to provide
41
information about interactions of hands and hand-held instruments with external objects.
42
Future studies will be guided by these insightful reports from Hsiao.
43
44
2
45
BEGINNINGS
46
47
Twenty-two years ago, we (Soechting and Flanders 1992) were neighbors with Steve
48
Hsiao and his mentor in a volume of the Annual Review of Neuroscience (Johnson and
49
Hsiao 1992). In their review, our contemporaries put forward the most lucid description
50
of tactile receptor sensitivities that we have encountered to date, outlining the
51
complementary nature of slowly-adapting type one (SAI) afferents (from Merkel
52
complexes), rapidly-adapting type one (RAI) afferents (from Meissner corpuscles), and
53
the rapidly-adapting signals from Pacinian corpuscles (PC). They explained the
54
preference of SAIs for isolated curves of skin indentation (LaMotte and Srinivasan 1987;
55
Phillips and Johnson 1981) as a mechanical surround-suppression effect (implicitly
56
analogous to the center-surround organization in the visual system), and went on to
57
speculate that the SAI receptor density may be matched to the skin’s low-pass filtering to
58
achieve the best possible spatial resolution. Whereas SAIs specialize in spatial
59
resolution, encapsulated receptor structures make RAIs and PCs selective for higher
60
frequency vibrations, and these latter two groups are complementary to each other in their
61
superficial location/small receptive field size (RA) compared to deep location/large
62
receptive field size (PC). With the example of tool use, Johnson and Hsiao (1992)
63
concisely summarized that SAIs would give a maintained representation of the detailed
64
form of the tool while RAs would, with high spatiotemporal fidelity, signal any slippage
65
of the instrument within the hand. The authors envisioned that the population of PCs
66
would transduce spatiotemporal patterns of transmitted vibrations to represent “transient
67
mechanical events that occur at the working surfaces of the instrument” (p. 246).
3
68
Regardless of whether these words were written by Hsiao or his mentor, this is the vision
69
of somatosensation with which Steven S. Hsiao began his career.
70
71
Hsiao’s first publication was momentous (Phillips et al. 1988). In it, the authors used
72
their new technique of characterizing the response of an individual neuron (peripheral or
73
cortical) by generating a spatial event plot (SEP). Figure 1 expands on Fig. 1 from
74
Phillips et al. (1988) in order to explain the approach. Immobilized monkeys were
75
anesthetized for peripheral recordings but alert for cortical recordings. As shown
76
schematically in Fig 1A, a raised letter was moved across a stationary finger pad. The
77
letters were raised above the background by 500 m, moved across the skin at 50 mm/s,
78
and applied with a servo-controlled normal force of 60 g. After each proximal to distal
79
movement, the letter stimulus was shifted 0.2 mm along its long axis (i.e., perpendicular
80
to the direction of the scanning movement) so that the neuron’s receptive field (dashed
81
circle) now scanned across the letter on a different row. The SEP in Fig. 1B is
82
constructed so that within each row, recorded action potentials can be positioned
83
according to the location of the receptive field relative to a raised letter stimulus at the
84
time when they occurred. In this study, there were 64 sweeps across the letter by the
85
receptive field of a somatosensory neuron, resulting in 64 rows of action potentials in the
86
SEP.
87
88
Figure 1 of Phillips et al. (1988) exhibited the action potentials of a cortical area 3b SA
89
neuron that always fired when the leading edge of the letter “K” was in its receptive field,
90
and fired with high fidelity whenever a raised portion occupied its receptive field. Figure
4
91
2 of Phillips et al. (1988) then revealed that a similarly high spatial resolution was
92
observed in recordings from SAI afferents. We exhibit this phenomenon in Fig. 2 (row
93
2), which is reprinted from a subsequent study where raised dot patterns (350 m) were
94
slowly scanned (20 mm/s) and applied with a servo-controlled normal force of 30 g
95
(Connor et al. 1990). As first shown by Phillips et al. (1988), Fig. 2 (row 3) demonstrates
96
that RA afferents also had sufficiently small receptive fields to provide an isomorphic
97
representation of the stimulus pattern. In contrast, PC afferents (Fig. 2, bottom row)
98
failed to do so because of their large receptive fields and preferred vibration rates (100-
99
300 Hz) above the speed of the scanned stimuli.
100
101
Hsiao and colleagues (Phillips et al. 1988) explained that these SEPs truly represented the
102
response of a single neuron to a stimulus moved across the skin, but also, especially in
103
the case of dense and uniform SA1 and RA afferent populations, they could be
104
interpreted as representing the population response, or the “neural image” of the stimulus.
105
They went on to discuss the sensitivities of various somatosensory cortical neurons as a
106
progression from an isomorphic to an abstracted touch representation.
107
108
As a sign of the times, this 1988 paper was entitled “Spatial pattern representation and
109
transformation…,” similar to our own titles involving sensorimotor transformations (e.g.,
110
Soechting and Flanders 1989a,b), where we conceptualized neural processing as
111
occurring in successive stages with information being serially transformed from one stage
112
to the next. The framework laid by Phillips et al. (1988) also broke transformations into
113
sequential stages. They hypothesized that cortical area 3b RA neurons and area 1
5
114
neurons may exhibit intermediate representations – much like the hidden units in Zipser
115
and Anderson’s (1988) model of transformation in parietal visual cortex.
116
117
FEATURES OF SENSORY RESPONSES
118
119
In the 1990’s and beyond, Hsiao and colleagues continued to develop sophisticated
120
methods for precise stimulation (e.g., Lane et al. 2010) and analysis of receptive fields,
121
while their conceptual framework evolved toward feature selectivity and perceptual
122
constancy. The early work on SAI and RA afferents was extended and elaborated by
123
showing the precise features of the stimuli encoded by the afferents, and by deducing that
124
roughness can be encoded by the spatial variations in the firing rate of a population of
125
SAIs (Connor et al. 1990; Blake et al. 1997, see Fig 2). More recently, Hsiao’s former
126
coworker, Bensmaia, further extended this interpretation by showing that the finer
127
features of texture stimuli are represented by spike timing in RA and PC afferents, and
128
has emphasized the central convergence of these different types of peripheral afferent
129
inputs (Weber et al. 2013; Saal and Bensmaia 2014).
130
131
Just as visual cortical neurons are tuned to the orientations of bars, neurons in primary
132
(S1) and secondary (S2) somatosensory cortex are tuned to the orientation of edges
133
impressed on the hand, a feature extracted from the response of populations of SAI
134
afferents (Hsiao et al. 2002). S2 neurons have receptive fields extending over multiple
135
digits, with markedly similar tuning properties for each finger (Fitzgerald et al. 2004,
136
2006), facilitating the haptic discrimination of the shapes of large objects. Hsiao and
6
137
colleagues noted the similarities in orientation selectivity in primary visual and
138
somatosensory cortical areas and suggested that these similarities arose from analogous
139
neural mechanisms in these areas (Bensmaia et al. 2008).
140
141
In the visual system, it is well known that the representation of motion is transformed
142
from a representation of the components of a pattern (e.g., the stripes of a plaid) to one of
143
the motion of the entire pattern (the plaid) as one ascends the hierarchy of visual cortical
144
areas (Movshon et al. 1985). Furthermore, this “dorsal stream” transformation evolves
145
gradually over time (Masson and Stone 2002). Hsiao and colleagues discovered an
146
analogous phenomenon in the tactile perception of motion (Pei et al. 2008) and
147
subsequently demonstrated a similar hierarchy in the somatosensory cortical areas (Pei et
148
al. 2011). Neurons in S1 area 3b responded to local (component) motion (analogous to
149
neurons in V1 area 17). However in S1 area 1, some neurons encoded the pattern motion
150
while others were tuned to component motion or exhibited intermediate characteristics
151
(analogous to the medial temporal visual area MT).
152
153
Ascending the hierarchy of “ventral stream” areas, cortical neurons are tuned to
154
progressively more complex features of contours and objects (cf., Brincat and Connor
155
2004). In a collaboration of the Hsiao and Connor laboratories, Yau et al. (2009)
156
explicitly tested the idea that there are analogous coding schemes in visual and
157
somatosensory areas. They recorded activity from intermediate visual area (V4) for
158
visual stimuli, and somatosensory area (S2) for tactile stimuli. In each case, monkeys
159
were presented with contours whose directions of curvature were systematically varied.
7
160
Indeed, they found neurons in both areas that responded briskly to such stimuli, and
161
found very similar patterns of tuning in both areas. Thus the concept of hierarchies of
162
feature representation holds well for both vision and somatosensation.
163
164
165
GOALS FOR THE FUTURE
166
In a recent paper on neuroprosthetics, Hsiao et al. (2011) gave their forecast and
167
recommendations on the prospect of providing artificial somatosensory feedback.
168
Building on decades of his work on feature selectivity, Hsiao predicted that it will
169
eventually be possible, using electrical or optical stimulation in cortex, to find the
170
“natural neural codes” (p. 76) to provide SA-type perception of form. Regarding the
171
restoration of RA-type input, Hsiao et al. (2011) recommended hard-wiring slip reflexes
172
into the mechanical hand itself. The authors reasoned that the user need not be
173
consciously aware of slippage, as long as it is automatically corrected. They compared
174
this to the case of Dr. Strangelove, where the movie character’s funny right hand had a
175
mind of its own. The authors also saw promise in the prospect of tapping into the PC
176
channel, if we can “mimic the temporal pattern of what a PC afferent would have
177
experienced under the same circumstances” (p. 73). The idea is to use the spatiotemporal
178
patterns that naturally arise when a person uses a tool to interact with external objects. In
179
this case the authors mentioned examples of a blind person using a cane and, more
180
dramatically, Luke Skywalker using his prosthetic hand to wield his lightsaber weapon!
181
182
Another current topic that will extend beyond the work of Hsiao is research on spinal
183
circuits, where we are beginning to gain an appreciation for the processing of tactile
8
184
feedback in the dorsal horn. At the end of their detailed review of peripheral and spinal
185
“touch” representations, Abraira and Ginty (2013) thanked their close colleague Steven
186
Hsiao for his helpful comments on the manuscript. Remarkably, these authors analogized
187
the spinal dorsal horn to the retina, for its complexity and importance in early sensory
188
processing. Indeed, individual peripheral afferent neurons may serve as the first stage of
189
directional feature selectivity (Pruszynski and Johansson 2014). Furthermore, this dorsal-
190
horn spinal circuit may be essential for comparing cortical motor commands with
191
somatosensory feedback during tool use and tactile exploration (Flanders 2011; Weiss
192
and Flanders 2011).
193
194
Whereas the pioneering approach of the spatial event plot (Johnson and Hsiao 1992;
195
Phillips et al. 1988) was necessarily done with the hand immobilized, a challenge for
196
future experimentation is to put tools in the hands and let them move freely along
197
surfaces. This future research will be guided by the vision of Hsiao.
198
199
200
Acknowledgements
201
The authors were supported by a grant for the National Institute of Neurological
202
Disorders and Stroke NS027484. We thank the three reviewers for their very helpful
203
suggestions.
204
205
9
206
References
207
Abraira VE, Ginty DD. The sensory neurons of touch. Neuron 79:618-639, 2013.
208
209
Bensmaia SJ, Denchev PV, Dammann JF 3rd, Craig JC, Hsiao SS. The representation of
210
stimulus orientation in the early stages of somatosensory processing. J Neurosci 28:776-
211
786, 2008.
212
213
Blake DT, Hsiao SS, Johnson KO. Neural coding mechanisms in tactile pattern
214
recognition: the relative contributions of slowly and rapidly adapting mechanoreceptors
215
to perceived roughness. J Neurosci 17:7480-7489, 1997.
216
217
Brincat SL, Connor CE, Underlying principles of visual shape selectivity in posterior
218
inferotemporal cortex. Nat Neurosci 7: 880-886, 2004.
219
220
Connor CE, Hsiao SS, Phillips JR, Johnson KO. Tactile roughness: Neural codes that
221
account for psychophysical magnitude estimates. J Neurosci 10:3823-3836, 1990.
222
223
Fitzgerald PJ, Lane JW, Thakur PH, Hsiao SS. Receptive field properties of the macaque
224
second somatosensory cortex: evidence for multiple functional representations. J
225
Neurosci 24:11193-11204, 2004.
226
227
Fitzgerald PJ, Lane JW, Thakur PH, Hsiao SS. Receptive field properties of the macaque
228
second somatosensory cortex: representation of orientation on different finger pads. J
229
Neurosci 26:6473-6484, 2006.
230
231
Flanders M. What is the biological basis of sensorimotor integration? Biol Cybern 104:1-
232
8, 2011.
233
234
Hsiao SS, Fettiplace M, Darbandi B. Sensory feedback for upper limb prostheses. Prog
235
Brain Res 192:69-81, 2011.
236
10
237
Hsiao SS, Lane J, Fitzgerald P. Representation of orientation in the somatosensory
238
system. Behav Brain Res 135:93-103, 2002.
239
240
Johnson KO, Hsiao SS. Neural mechanisms of tactual form and texture perception. Ann
241
Rev Neurosci 15:227-250, 1992.
242
243
LaMotte RH, Srinivasan MA. Tactile discrimination of shape: Responses of slowly
244
adapting mechanoreceptive afferents to a step stroked across the monkey fingerpad. J
245
Neurosci 7:1655-1671, 1987.
246
247
Lane JW, Fitzgerald PJ, Yau JM, Pembeci I, Hsiao SS. A tactile stimulator for studying
248
passive shape perception. J Neurosci Methods. 185:221-229, 2010
249
250
Masson GS, Stone LS. From following edges to pursuing objects. J Neurophysiol
251
88:2869-2873, 2002.
252
253
Movshon JA, Adelson EH, Gizzi G, Newsome WT. The analysis of visual moving
254
patterns. In: Pattern Recognition Mechanisms, edited by. Chagas C, Gattass R, and Gross
255
C. New York: Springer, 1985.
256
257
Pei YC, Hsiao SS, Bensmaia SJ. The tactile integration of local motion cues is analogous
258
to its visual counterpart. Proc Natl Acad Sci U S A. 105:8130-8135, 2008.
259
260
Pei YC, Hsiao SS, Craig JC, Bensmaia SJ. Neural mechanisms of tactile motion
261
integration in somatosensory cortex. Neuron 69:536-547, 2011.
262
263
Phillips JR, Johnson KO. Tactile spatial resolution: III. A continuum mechanics model of
264
skin predicting mechanoreceptive responses to bars, edges, and gratings. J Neurophysiol
265
46:1204-1225, 1981.
266
11
267
Phillips JR, Johnson KO, Hsiao SS. Spatial pattern representation and transformation in
268
monkey somatosensory cortex. Proc Natl Acad Sci. USA 85:1317-1321, 1988.
269
270
Pruszynski JA, and Johansson RS. Edge-orientation processing in first-order tactile
271
neurons. Nat Neurosci 17:1404-1409, 2014.
272
273
Saal HP, Bensmaia SJ. Touch is a team effort: interplay of submodalities in cutaneous
274
sensibility. Trends Neurosci (in press) 2014.
275
276
Soechting JF, Flanders M. Sensorimotor representations for pointing to targets in three-
277
dimensional space. J Neurophysiol 62: 582-594, 1989.
278
279
Soechting JF, Flanders M. Errors in pointing are due to approximations in sensorimotor
280
transformations. J Neurophysiol 62: 595-608, 1989.
281
282
Soechting JF, Flanders M. Moving in three-dimensional space: frames of reference,
283
vectors, and coordinate systems. Ann Rev Neurosci 15:167-191, 1992.
284
285
Weber AI, Saal HP, Lieber JD, Cheng JW, Manfredi LR, Dammann JF 3rd, Bensmaia SJ.
286
Spatial and temporal codes mediate the tactile perception of natural textures. Proc Natl
287
Acad Sci USA 110:17107-17112, 2013.
288
289
Weiss EJ, Flanders M. Somatosensory comparison in haptic tracing. Cereb Cortex
290
21:425-434, 2011.
291
292
Zipser D, Anderson RA. A back-propagation network that simulates response properties
293
of a subset of posterior parietal neurons. Nature 331: 679-684, 1988.
294
295
Yau JM, Pasupathy A, Fitzgerald PJ, Hsiao SS, Connor CE. Analogous intermediate
296
shape coding in vision and touch. Proc Natl Acad Sci U S A. 106:16457-16462, 2009
297
12
298
Figure Captions
299
300
FIG. 1.
Panel A shows how moving a raised letter (K) toward the distal
301
end of a finger pad, from right to left in this figure, advances the stationary skin receptive
302
field (dashed circle) from left to right across the letter. The receptive field was advanced
303
by 1 mm every 20 ms (50 mm/s). Panel B shows the conventions of a spatial event plot
304
(SEP) for the action potentials recorded at each location.
305
306
FIG. 2.
Example spatial event plots (SEPs) gathered while grids of 0.5 mm
307
diameter raised dots were scanned across the distal finger pad of an anesthetized macaque
308
monkey. The top row shows the dot spacing (1.3 – 6.2 mm) of the stimuli. The middle
309
two rows show that SAI and RA peripheral afferents can provide nearly isomorphic
310
neural images of the pattern; the bottom row suggests that PC afferents cannot.
311
Reproduced with permission from Connor et al., J Neurosci 10:3826 (Fig. 4), 1990.
13
A. Schematic of receptive field
B. Spatial event plot to mark
scanning across a raised letter
action potentials at place of
occurrence on letter stimulus
K
8 mm
high
receptive field swept across
64 rows
raised 500 µm
upper
corner of
embossed
letter “K”
2 mm wide
1 mm/tick (20 ms/tick at 50 mm/s)
ce.
eter and dot spacinghad statistically significant
essestimates(ANOVA p < 0.000 1). Perceived
asedsomewhat with increasingdot size, espeerdot spacings.Dot spacinghad a much larger
t diameter, roughnessincreasedwith increasing
o 3.2 mm, then declined. Thus, the roughness
ectionsfor typical SA,
s.The corresponding
stimuluspattern are
w (dot diameter,0.5
omprises
a setof rasm repetitive parallel
ulusacrossthe recepstimulusshiftedverbetweeneachscan.
. . a.-.
. . .......
..........
...........
.*-,
:.-
.
Figure 3B shows, for example, that dot spacingsof approximately 2 and 5 mm produce the sameroughnessmagnitude
even though they evoke very different neural responses,as described in the next section. The remainder of the paper is concerned with discovering the coding mechanismthat extracts a
constant factor in different neural responses
that evoke the same
subjective senseof roughness.
.
..........
......
.....
.
......
......
‘.-.-.-.-*-,
......... .
......... Z.’
. . l :
.’
PC
Dot spacing (mm)