Algorithms of whisker-mediated touch perception -...
Transcript of Algorithms of whisker-mediated touch perception -...
Algorithms of whisker-mediated touch perceptionMiguel Maravall1 and Mathew E Diamond2
Available online at www.sciencedirect.com
ScienceDirect
Comparison of the functional organization of sensory
modalities can reveal the specialized mechanisms unique to
each modality as well as processing algorithms that are
common across modalities. Here we examine the rodent
whisker system. The whisker’s mechanical properties shape
the forces transmitted to specialized receptors. The sensory
and motor systems are intimately interconnected, giving rise to
two forms of sensation: generative and receptive. The sensory
pathway is a test bed for fundamental concepts in computation
and coding: hierarchical feature detection, sparseness,
adaptive representations, and population coding. The central
processing of signals can be considered a sequence of filters.
At the level of cortex, neurons represent object features by a
coordinated population code which encompasses cells with
heterogeneous properties.
Addresses1 Instituto de Neurociencias de Alicante UMH-CSIC, Campus de San
Juan, Apartado 18, 03550 Sant Joan d’Alacant, Spain2 Tactile Perception and Learning Lab, International School for Advanced
Studies-SISSA, Via Bonomea 265, 34136 Trieste, Italy
Corresponding author: Diamond, Mathew E ([email protected])
Current Opinion in Neurobiology 2014, 25:176–186
This review comes from a themed issue on Theoretical and
computational neuroscience
Edited by Adrienne Fairhall and Haim Sompolinsky
0959-4388/$ – see front matter, # 2014 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.conb.2014.01.014
IntroductionIn the process that culminates in sensing and identifying
an object, the starting point is the encoding of physical
parameters by sensory receptors. A growing set of inves-
tigations focuses on transformations along sensory path-
ways as a means to understand the conversion from raw
physical signals into sensations and percepts. A long-
standing hypothesis is that those transformations are built
up from a set of standard ‘canonical’ computations, imple-
mented repeatedly [1] and combined to generate
responses that are selective, specific and flexible [2].
Taking this hypothesis as a point of departure, this review
aims to identify canonical computations, or algorithms,
implemented in the rodent whisker system, an ‘expert’
system [3].
We focus on computations along the receptor-to-cortex
ascending pathway; nevertheless a complete picture of
Current Opinion in Neurobiology 2014, 25:176–186
tactile sensation will only be achieved by understanding
how sensory and motor computations are woven together
[4,5].
Mechanical forces in the follicleAs in any sensory pathway, transduction from physical
entities into action potentials constrains all later proces-
sing. Input signals — fluctuations in mechanical energy at
the whisker base — are shaped by the interaction be-
tween the whisker’s motion, its mechanical properties
(e.g., compliance) and properties of the contacted object.
The whisker-follicle junction is rigid, allowing robust
transmission and readout of the forces induced by whisker
motion [6��].
The form of whiskers (Fig. 1a) determines their mech-
anical behavior. Bending stiffness decreases from whisker
base to tip due to taper [7��], and the concomitant
increase in flexibility enables the slippage of whiskers
during object exploration [8,9�]. Additional flexibility is
achieved by their hollow structure [10].
New methods for tracking whisker motion have allowed
detailed analysis of how whiskers interact with objects
[11,12]. The combination of whisker measurements with
models of whisker deflection has begun to specify bend-
ing [9�,13] and changes in forces at the whisker base
[6��,7��,9�,14]. Contact-induced whisker deformations
can be decomposed into a slow bending component
and a transient vibrational component [15]; the relative
contributions of different components depend on the
specific interaction [6��,7��,16]. Algorithms involving
comparison of components require those components
to be effectively transduced by mechanoreceptors.
When whiskers sweep across a textured surface (Fig. 1b),
they are trapped and released by surface ridges and grains
[17–19,20��] (reviewed in [21]). These brief (�2 ms)
‘stick–slip’ events cause transient, high-frequency
vibrations [18]. The consequent sequence of fluctuations
in mechanical energy provides a signature of texture
[17,18]. Stick–slip events excite primary sensory neurons
and their targets [17,18,20��,22]. However, differences in
‘stick–slip’ events across trials are not well-correlated
with trial-to-trial choices in a texture discrimination task
[20��]; other features of whisker motion may also con-
tribute.
Transduction of touch into neuronal signalsTransduction is carried out at the terminals of neurons
whose cell body resides in the trigeminal ganglion (TG;
Fig. 1c). The many mechanoreceptor types, distributed
www.sciencedirect.com
Algorithms of touch perception Maravall and Diamond 177
Figure 1
(a) FF
LX
C
θ
T
(b)
(c)
BC
Cortex
VPMPOm
TG
TN
brain stemfollicle
whisker
Current Opinion in Neurobiology
Input forces to the sensory system and the ascending pathway. (a) The force acting upon a whisker during contact, and thus transmitted to the
receptors in the follicle, is illustrated. The object at position X strikes a whisker of length L at a distance C from the skin and at angle u away from the
whisker’s resting angle, inducing a force F. (b) Illustration of a single large stick–slip event. One frame from a high-speed (1000 frames/s) video is
shown in gray scale. The whisker traces have been enhanced to increase their visibility. While the rat palpated the surface to judge the groove spatial
frequency, one whisker was tracked through a sequence of frames and the traces, from violet to light blue, show the whisker position over 1 ms
timesteps. The whisker tip was blocked in a groove and then sprung free as the rat retracted the whisker shaft in the posterior direction. (c) Principal
sensory pathways to the cortex are illustrated schematically. TG neurons send a peripheral branch to the skin and a central branch into the trigeminal
nuclei (TN) of the brainstem. Axons from TN cross the midline to reach the thalamus, terminating in VPM and the posterior medial nucleus, POm.
Thalamic neurons project to BC. Blue, red, and green lines represent parallel pathways that carry different sorts of tactile information, as reviewed
elsewhere. (a) Adapted personal communication from A. Hires and K. Svoboda; (b) adapted from [20��]; (c) adapted from [5].
differentially across the follicle, have diverse response
properties and are best activated by distinct forces with
particular time courses. Given the diversity of receptor
types and spatial distributions, it is not surprising that
both of the principal functional classes of primary afferent
neurons, slowly and rapidly adapting (SA and RA), in fact
comprise a rich variety of feature combinations. Thus,
each neuron displays distinct sensitivity to the location,
direction and velocity of whisker displacements evoking
lateral forces [17,23–26], to the pattern of axial forces [25],
to whisking phase [27–29] and to contact, detachment or
their combinations [24,27].
As a population, TG neurons represent the space of
dynamical features of one whisker through a high-dimen-
sional code (�200, counting each neuron as a dimension)
(see Section ‘Feature selectivity’) [30��]. This permits
rapid information encoding: specific patterns of forces
(e.g., [7��]) may engage subsets of neurons to ‘label’ the
stimulus. Whisker motion patterns are richly represented
by the TG population, allowing several population-based
decoding schemes for any task. For example, information
present across the population probably permits instan-
taneous comparison of the relative magnitudes of
www.sciencedirect.com
different force components (see Section ‘Mechanical
forces in the follicle’). Intriguingly, each TG neuron
projects to multiple target neurons within a column of
the principal trigeminal nucleus (‘barrelette’), and indi-
vidual barrelette neurons receive convergent inputs of
different afferent types (SA, RA): thus, TG population
signals are decoded in a ‘one-to-many’ and ‘many-to-one’
manner [31�].
A requirement for TG to implement a fast population
code based on relatively small numbers of spikes is that
spike generation be precise. Indeed, TG neurons respond
with highly reliable firing patterns [17,23,32,33] and are
among the most temporally precise neurons yet discov-
ered in the animal kingdom [17,30��,32]. The minimal
sequential pathway from receptors to cortex (Fig. 1c)
requires just three synapses — very short compared to
other sensory systems.
To summarize, TG neurons convey information pack-
aged in a high-temporal precision code where different
neurons encode diverse physical properties. The speed of
peripheral encoding is reflected in the finding that cortical
neurons carry texture information within 20 ms after
Current Opinion in Neurobiology 2014, 25:176–186
178 Theoretical and computational neuroscience
Box 1 Self-generated motion and modes of sensing.
Touch entails putting sensors into contact with an object to
determine its identity, properties or location. Accordingly, the
intricate loops connecting ‘sensory’ and ‘motor’ circuits can be
understood in terms of the need to meld sensory and motor
information — motor output generates sensory input, sensory input
modulates motor output [5,37,38,39,40,41�].
Whisker-mediated sensation occurs through two modes of opera-
tion. In the generative mode (Fig. 2a), the animal moves its whiskers
to actively seek contact with objects and palpate them: the animal
causes the percept by its own motion [42]. Tasks involving the
generative mode include wall following [43], gap measurement [44],
texture discrimination [19,20��,21,35,45], and object localization
[46,47]. In the generative mode, whisker-mediated perception has
been put forward as a process where sensory and motor systems
dynamically converge, through repeated contacts, until both
neuronal representations reach a stable state [46,48].
In the receptive mode (Fig. 2b), the animal places its whiskers on an
object and is frequently observed to immobilize its whiskers to
optimize signal collection. Tasks include detection and discrimina-
tion of vibrations applied to whiskers [44,49–54] and discrimination of
the width of an aperture [55].
whisker contact [34,35]. In the rest of this review, we
highlight emerging themes concerning the transform-
ation of signals from TG to cortex.
Active sensingActive sensing systems are purposive and information-
seeking [36]: active sensing entails control of the sensor
apparatus in whatever manner best suits the task.
Although the concept of sensor apparatus control applies
to all modalities, it is perhaps most evident in the
modality of touch. Recent evidence (Box 1) indicates
Figure 2
Generative Sensing (a)
object motionless
head andwhiskersin motion
Two modes in which rats collect tactile information. Both panels are single
been enhanced to increase their visibility. (a) During generative sensing, the
the whisker tips and the object. This mode of sensing is critical when the obj
motor system. The image is taken as the rat judges the spatial frequency of
provides mechanical energy. Since the rat’s percept could be confounded b
image is taken as the rat perceives the vibration of the flat plate. (a) Adapte
Current Opinion in Neurobiology 2014, 25:176–186
that rats and mice select motor programs to interact with
objects according to the type of information to be
acquired — rodents can either generate sensations or
receive sensations.
Feature selectivityNeurons early in sensory processing are sensitive to the
presence of particular features in a stimulus. Feature
selectivity can be approximated by treating the neuron
as a device that (1) linearly filters its input and (2)
responds by applying a nonlinear threshold function to
the filtered stimulus [56]. Linear filtering is thus a candi-
date canonical algorithm [1].
Variants and generalizations of linear-nonlinear
models predict trains of action potentials in TG
[17,23,30��,57,58], the ventral posterior medial thalamic
nucleus (VPM) [57,59] and the barrel cortex (BC) [60,61��].Neurons are selective to temporal features of whisker
motion (Fig 3). For subcortical neurons, filtering is typically
simple in that a single, short-duration feature (e.g., instan-
taneous velocity) predicts a neuron’s response [30��,59].
These features are conserved from TG [30��] to VPM [59]
(Fig. 3b and c), though VPM processing differs in other
important aspects (discussed below). In contrast, BC
neurons are sensitive to temporal features (Fig. 3d and
e), but have more complex, high-dimensional selectivity:
multiple features, sometimes in combination, are necess-
ary to explain a neuron’s firing [60,61��]. Cortical neurons
respond to nonlinear dynamical features such as speed (the
absolute value of velocity) [17,60,61��] and are sensitive to
correlated motion between whiskers [61��] (Fig. 3f). Sub-
cortical tuning curves indicate faithful representation of
filtered stimulus magnitude, while cortical curves indicate
Receptive Sensing (b)
object inmotion
head and whiskersmotionless
Current Opinion in Neurobiology
frames from high-speed (1000 frames/s) video; the whisker traces have
rat moves its head and whiskers to create dynamic interaction between
ect is immobile and the mechanical energy must originate in the animal’s
grooves on the object surface. (b) During receptive sensing, the object
y its own motor output, the head and whiskers remain motionless. The
d from [20��]; (b) adapted from [129].
www.sciencedirect.com
Algorithms of touch perception Maravall and Diamond 179
Figure 3
(a) Ideal filters
position
Time (ms)
-40 10 10 10
Time (ms)
-40
Time (ms)
-40
Time (ms)
Neuron 1 Neuron 2 Neuron 3
-40 10 10 10
Time (ms)
-40
Time (ms)
-40
Time (ms)
Neuron 1 Neuron 2
Neuron 1
Filter 1 Filter 1
Pha
se (
rad)
Filter 2 Filter 2
Common input
Independent inputs “local”neurons
“global” neurons
Response to uncorrelated stimulation (z score)
Cel
l cou
nt
Neuron 2
Neuron 3
-40 10 10 10
Time (ms)
-40
Time (ms)
-40 10 10
0
-
2
π
π
πTime (ms)
-40
Time (ms)
-40 10 10
0
Time (ms)
Time (ms)
-40
40
1
2
24
c1-c + ×
20
01 10 100
-40
Time (ms)
-40
velocity acceleration
(b) Trigeminal ganglion
(c) VPM thalamic nucleus
(d) Barrel cortex (e) Cortical population filters
(f) Cortical selectivity for correlated vs uncorrelated stimulation
2π
Current Opinion in Neurobiology
Feature selectivity. (a) ‘Ideal’ filters for extraction of position, velocity and acceleration values. Note similarity between these waveforms and those in
the remaining panels. (b) Examples of filters for TG neurons. Each neuron extracts a distinct, rapid feature. (c) Examples for VPM neurons. (d) BC
neurons are sensitive to multiple features in combination. Features are of longer duration than those comprising subcortical filters, signifying longer
temporal integration. (e) Dataset of linear filters for barrel cortex neurons sensitive to single-whisker stimulation, sorted by the relative phase of the filter
waveform. (f) Distribution of barrel cortex sensitivity to stimuli with different degrees of interwhisker correlation. (a) and (c) Adapted from [59]; (b) from
[30��]; (d–f), from [61��].
www.sciencedirect.com Current Opinion in Neurobiology 2014, 25:176–186
180 Theoretical and computational neuroscience
Figure 4
(a) “L6”“L5”“L4”“L2/3”
60100
10-1
10-2
0.05
10-3
10-4
50
40
30
Ove
rall
spik
e ra
te (
Hz)
20
10
00 1 2 3 4 5 6 7 8 9418 588
Depth (μm) Session ranked by discrimination
Dis
crim
inat
ion
p va
lue
890 1154
(b)
Current Opinion in Neurobiology
Firing rate heterogeneity and population coding in BC. (a) Spike rates during pole localization in head-fixed mice. Each circle is one neuron and colored
bars indicate laminar boundaries. Median activity level differs across layers, yet within each layer there is marked variability. (b) Single-neuron and
population discrimination performance for 9 experimental sessions of a texture discrimination task in freely moving rats. Gray squares correspond to
individual neuronal clusters (single-unit or multi-unit); black circles, to the population of simultaneously recorded clusters (3–7 clusters per population).
p value is the probability that the discrimination performance of a neuron or population arose by chance; horizontal gray line, p = 0.05. Individual
neuronal clusters are variable and often perform below chance ( p > 0.05). However, every population performs well above chance. (a) Adapted from
[68��]; (b) from [78��].
the detection of events that exceed background by some
proportion [57,59,60]. Until there is evidence that can be
explained only by alternative models, our view is that the
high-dimensional selectivity of cortical neurons is, to a first
approximation, the result of convergence of neurons with
simpler properties.
Early studies described nonlinear integration of the
motion of multiple whiskers in cortex (reviewed in
[62]); such findings were given a new framework by
recent research on multiwhisker correlations [61��]. Mul-
tiwhisker selectivity allows neurons to construct ‘appar-
ent motion’ from sequential deflection of adjacent
whiskers [63]. Interestingly, a neuron’s directional sensi-
tivity to apparent motion is not predicted by its direc-
tional sensitivity to motion of its principal whisker.
Neurons in VPM can also display selectivity to apparent
global motion [64], although more weakly than in the
cortex [64,65].
Most studies of feature selectivity have relied on deliver-
ing controlled whisker stimulation to anesthetized
animals; as yet, no study has directly compared tuning
properties under anesthesia with response properties of
the same cells during behavior. However, feature selec-
tivity properties have strong parallels in animals perform-
ing discrimination tasks. Whisker contact modulates the
firing rate of BC neurons in behaving animals
[18,35,66,67,68��,69�] (S.A. Hires et al., abstract in SocNeurosci Abstr 2012, 677.18): responses correlate with
Current Opinion in Neurobiology 2014, 25:176–186
features including force magnitude, frequency of stick–slip events and maximum curvature.
SparsificationIn the whisker pathway, activity levels change system-
atically from stage to stage, a change that can be quanti-
fied as ‘sparseness’ — the fraction of neurons that are
active and encoding information at any moment
[68��,70,71,72] (Fig. 4a). The computational advantages
of sparse activity have been reviewed elsewhere [72–75].
Sparseness can reduce overlap (correlation) between
activity patterns, facilitate learning, discrimination and
categorization, and limit energy expenditure. A repres-
entation where individual neurons are selectively sensi-
tive to high-level features (Section ‘Feature selectivity’)
is a step toward the ‘concept’ representations found at the
final stage of cortical processing [2,76].
In whisker-mediated touch perception, sparseness
increases at more central stations of the sensory pathway,
peaking in layers 2/3 of BC (reviewed in [72]) (Fig. 4a).
For example, in behaving animals, stick–slip events
during texture exploration elicit low-probability, pre-
cisely timed cortical responses [18]. Mechanisms under-
lying sparsification are being examined [77].
Response heterogeneityA striking finding is that neurons responsive to the same
whisker diverge markedly in their responses to a given
peripheral event. Neurons within the same column can
www.sciencedirect.com
Algorithms of touch perception Maravall and Diamond 181
participate strongly, or not at all, in the encoding of
objects contacted by the corresponding whisker
[18,35,68��,69�,78��]. This heterogeneity occurs from
TG to all layers of BC, with two important properties.
First, neurons within each circuit are tuned to diverse
stimulus properties. Second, neurons have diverse
activity levels.
Diverse tuning properties
Tuning to stimulus features is heterogeneous at every
stage where it has been assessed (Section ‘Feature selec-
tivity’) [30��,59,61��] (Fig. 3). Since the preferred
stimulus features of neurons in VPM are similar to those
in TG, heterogeneity is partly inherited from stage to
stage [30��,59]. Higher-level features in cortex are also
heterogeneous across neurons, providing a rich repres-
entation of stimuli [61��]. Neurons in animals performing
sensory discrimination tasks also display diverse selectiv-
ity for sensory features, as do fibers projecting into BC
from motor cortex [41�,69�,79] (S.A. Hires et al., abstract
in Soc Neurosci Abstr 2012, 677.18). Neurons with different
feature selectivity are spatially intermingled within each
barrel column [80–82].
Diverse activity levels
Although mean activity levels differ systematically from
stage to stage (Section ‘Sparsification’), within each stage
there is a wide variability in activity and responsiveness
among individual neurons [72] (Fig. 4a). For example, in
the supragranular layers, where activity is sparsest, studies
under a range of conditions have nevertheless found a
subset of highly active neurons numbering �20–30% of
the population [68��,69�,81,82,83�,84].
The fraction of responsive neurons in a particular con-
dition could reflect selectivity to the stimuli presented. If
so, sparse, heterogeneous responses would reflect the
failure of stimuli to engage most neurons. However,
the finding that many neurons exhibit low firing rate
under a variety of behaviors and forms of stimulation
[18,68��,69�,84,85], even when stimulation systematically
explores a large region of parameter space [61��], suggests
that sparse firing is more a property of neurons than an
outcome of inadequate stimulus sets.
Responsiveness correlates with spontaneous activity
levels [84,85]. Further, the relative activity level of
strongly responding BC neurons is stable [84,85], despite
changes in tuning properties over time [84]. Thus, some
neurons appear intrinsically more active irrespective of
their specific tuning. At present it is unclear whether the
more active neurons constitute a separate category or one
extreme of a continuous distribution. In layer 2/3, more
active neurons receive more excitation and less inhibition
[83�,85]. Some of the heterogeneity in response proper-
ties correlates with differences in neuronal projection
pattern: layer 2/3 cells that project to motor cortex have
www.sciencedirect.com
different excitability than those projecting to secondary
somatosensory cortex [79] and respond differently during
tactile tasks [69�].
Spiking precision, timing-basedrepresentations, and synchronyAs described above, TG responses display reliable and
precise timing. Timing precision remains high across the
pathway: typical response ‘jitter’ (spike time standard
deviation across trials) is �0.2 ms in TG, �0.4 ms in
VPM and �1–2 ms in BC [17,30��,86].
The pressure driving temporal precision is unlikely to be
a need to limit variation in absolute timing, because jitter
is orders of magnitude smaller than the trial-to-trial
variability observed in naturalistic exploratory behavior
[3,17,20��]. The requirement, instead, is for sharp relativetiming, enabling precise temporal patterning of spikes
across neurons — either to permit sensory responses to be
referred to the animal’s own whisking motion, or to
facilitate activity propagation [87].
A sensory whisking signal is found in at least a subset of
TG neurons (Section ‘Transduction of touch into
neuronal signals’). Information about both whisking
amplitude and midpoint is relayed to BC
[40,41�,88,89]. BC neurons can show a small modulation
of membrane potential by whisking phase [66,79,83�,90],
allowing contact responses to vary with phase. The
whisking phase signal combined with the amplitude
and midpoint references could enable reconstruction of
whisker position and of the location of objects relative to
the animal’s face [39,67,91].
Sensory codes based on temporally patterned spikes can
convey more information than spike count-based codes
[92,93]. In BC, latency differences across populations can
code for whisker identity [94,95]. Spike patterns both in
TG and in BC can discriminate texture in a manner
decodable by an ideal observer [34].
Decoding spike timing information requires a temporal
reference. When the system is in generative mode, the
reference can be the whisking signal, as above. A refer-
ence also present in receptive mode, or when a stimulus
appears unpredictably, is the stimulus-evoked population
activity itself, both in the form of collective activity
oscillations and of synchronous spiking across subsets
of neurons [50,92,93,96]. Which neuronal subsets become
synchronously coactive in, e.g., VPM, depends on the
particular stimulus [97,98] (M. Bale et al., abstract in SocNeurosci Abstr 2011, 704.14). Therefore, the identity of the
synchronous subset can convey robust stimulus infor-
mation [99,100]. Dynamic changes in synchrony across
subpopulations of neurons can probably be decoded
downstream, as neurons are sensitive detectors of small
Current Opinion in Neurobiology 2014, 25:176–186
182 Theoretical and computational neuroscience
changes in synchrony across an afferent population
[100,101].
Synchronous activation allows activity to propagate reliably
in the thalamocortical pathway, where individual synaptic
connections are weak on average [87]. However, full,
precise synchrony across a large part of the presynaptic
population is unlikely to be required, as some thalamocor-
tical and intracortical connections can be effective enough
to help shape postsynaptic activation [102–104]. Instead,
activity propagation may occur via dynamically changing
partial synchrony across subsets of neurons. This enables
both robust activity propagation and rich population
representations, involving combinations of neurons that
become coactive based on shared feature selectivity
[99,100]. Indeed, selective, sparse synchronous activation
of subsets of cells is a hallmark of cortical responses during
texture exploration in awake rats [18].
Spike timing information and behavior
Is the information in spike timing used during behavior? In
a radial distance discrimination task, precise (ms) timing of
contacts within the whisking cycle was not crucial, because
uncertainty in contact time introduced by whisker azi-
muthal jitter had no effect on performance [7��]. In a pole
localization task, mice detected whether the pole fell in the
target region by maintaining their whiskers approximately
within the region and waiting for the object to strike the
whisker [47]. This maximized the difference in number of
contacts and in spike counts between positive and negative
trials [68��]. Consistent with use of a spike-count but not a
timing code, optogenetic stimulation in layer 4 increased
count and fooled animals into detecting a ‘phantom pole’;
jittering stimulation timing within a whisk caused no
appreciable effect [105�].
In texture discrimination tasks, as discussed above,
sequences of ms-scale fluctuations in whisker motion
differ according to texture, and neurons can carry texture
information by sequences of spikes [18,34]. Further data
from behaving rats are required to evaluate spike timing
as a candidate code for texture [21,106]. Future work on
the relevance of spike timing may test how animals solve
whisker-mediated tasks that require a timing-based
strategy. However, even more powerful would be the
demonstration that a task potentially solvable with a non-
timing scheme is actually solved using spike timing. This
test could be accomplished, for instance, if a behavioral
deficit were induced by jittering spike timing for two
stimuli that evoke different spike counts.
Response dynamics and adaptationAdaptation, the accommodation of response levels to
ongoing or repeated stimulation, occurs across species
and sensory modalities; the consequent change in
neuronal sensitivity is an instance of a canonical compu-
tation [1,107]. Responses evolve over time because of
Current Opinion in Neurobiology 2014, 25:176–186
variation in how the sensors encounter the object and
because of dynamics inherent to neuronal responses.
In one adaptation paradigm, repeated whisker stimulation
decreases responses to successive deflections (reviewed
in [62]), sharpening tuning properties of BC neurons,
including selectivity to whisker identity [108] and direc-
tion of motion [98,109]. This adaptation enhances dis-
crimination between stimuli of different amplitudes, at
the expense of detectability [110]. An enhanced ability of
neuronal responses to discriminate different stimulus
magnitudes is also found during adaptation to sinusoidal
whisker stimulation, suggesting a common functional
effect [111,112]. Intriguingly, regular-spiking (putative
excitatory) cortical neurons adapt more weakly to stimuli
repeated at irregular intervals than at a constant rate [113],
while putative inhibitory neurons do the opposite [114].
This behavior is echoed in the finding that human sub-
jects experience an increase in the perceived intensity of
a vibration when noise is introduced [114].
In another paradigm, BC neurons modify their firing rate
and sensitivity according to the ongoing statistical distri-
bution of stimuli: changes in the overall scale (variance or
‘contrast’) of a stimulus distribution elicit compensatory
adjustments in neuronal sensitivity (gain), thus maintain-
ing the information conveyed about the stimulus [60,115].
TG and VPM neurons also display adaptation and gain
adjustment under changes in variance [57]. Individual
neurons at each subcortical stage display variable adap-
tive behavior, ranging from fixed gain to full gain rescaling
[57]. This diversity may enable subcortical populations to
preserve representation of both relative and absolute
magnitudes. Because responses to dynamic stimuli adapt
over multiple timescales, adaptation allows neurons to
encode components of whisker stimuli with different
time courses (e.g., fast, high-frequency vibrations or slow
modulations in whisking waveform) [116].
Different forms of stimulation affect distinct biophysical
mechanisms, with overlapping roles. Adaptation to
repeated stimulation relies mostly on synaptic mechan-
isms, including short-term depression [117]. The under-
lying changes are complex and sensitive to specific
stimulation parameters: e.g., small-amplitude stimuli
sometimes elicit stronger adaptation than large-ampli-
tude stimuli [118]. Adaptation to noise stimulus statistics
probably recruits intrinsic membrane mechanisms
[119,120]. In summary, adaptation in the whisker system
has the effect of refining stimulus representations in a
context-dependent manner.
Coordination in BC populationsThe heterogeneity of functional properties across
neurons raises the issue of how populations collectively
represent information. Is activity coordinated? How many
neurons must an ‘observer’ (experimenter or downstream
www.sciencedirect.com
Algorithms of touch perception Maravall and Diamond 183
cell) sample to extract a robust sensory message? Is it
necessary to ‘listen’ to the ‘best’ neurons? In rats perform-
ing a texture discrimination task, some neurons, taken
alone, nearly match the animal’s performance, while
others carry no readily apparent texture signal [35]. Ran-
domly sampled clusters comprised of as few as 3–7
neurons can be linearly combined to robustly discriminate
texture even in cases where no individual neuron
represents the stimuli [78��] (Fig. 4b). The collective
signals arise from synergistic interactions between hetero-
geneous neurons, confirming long-standing theoretical
proposals [121–124]. Population representations of sen-
sory and motor parameters in motor cortex during a pole
localization task saturate for comparable-sized groups
(�2–6 neurons) [40]. Thus, downstream neurons receiv-
ing input from distinct subgroups of BC neurons can each
receive a robust message. Moreover, population repres-
entations of task parameters remain stable even when
individual neurons are plastic [40].
In sum, downstream neurons can extract texture identity
through a simple decoding scheme involving linear
synaptic weighting [78��]. The robustness of linear
decoding schemes applies across cortical areas and is a
candidate for a fundamental algorithm of cortical com-
putation [125–127].
ConclusionsThis review documents recent progress in understanding
the algorithms by which whisker shape and motion are
translated into patterns of neuronal activity distributed
across networks. We conclude by highlighting experimen-
tal paradigms that offer the opportunity to attack many
remaining questions. How does the feature detection
framework translate to the behaving sensory system?
Can sensory neurons integrate ‘local’ features over time
to generate a ‘global’ stimulus representation? Can such a
global representation be transferred to other brain regions
for storage and manipulation? In a delayed response task,
the mouse must use generative sensing to localize an
object, but delay its choice for a go cue. The flow of activity
from sensation to action has been mapped [128�]. In
another task, rats must use receptive sensing to perform
a tactile delayed comparison task [129��]. The rat receives
two stimuli, ‘base’ and ‘comparison,’ separated by a variable
delay. Each stimulus is a vibration, generated as a series of
velocity values sampled from a normal distribution. The rat
must judge which stimulus has greater velocity standard
deviation. The stimulus is exactly the sort of ‘noise’ used to
map neuronal feature selectivity through reverse corre-
lation methods: the design allows the investigator to
examine the algorithms of whisker-mediated perception
that underlie behavioral performance.
AcknowledgementsWe are grateful to Stuart Ingham for help with the artwork in Fig. 2 and toMichael Bale for comments on the manuscript. We apologize to our
www.sciencedirect.com
colleagues whose work could not be cited in entirety due to spaceconstraints. Funding was from the Spanish Ministry of Economy andCompetitiveness grant BFU2011-23049 (co-funded by the European Fundfor Regional Development); the Valencia Regional Government grantPROMETEO/2011/086; the Human Frontier Science Program grantNeuroscience of Knowledge (RG0015/2013); the European ResearchCouncil Advanced grant CONCEPT (294498); the European Union FETgrant CORONET (269459); the Italian Ministry for Universities andResearch grant HANDBOT, and the Compagnia San Paolo.
References and recommended readingPapers of particular interest, published within the period of review,have been highlighted as:
� of special interest�� of outstanding interest
1. Carandini M: From circuits to behavior: a bridge too far? NatNeurosci 2012, 15:507-509.
2. DiCarlo JJ, Zoccolan D, Rust NC: How does the brain solvevisual object recognition? Neuron 2012, 73:415-434.
3. Diamond ME, Arabzadeh E: Whisker sensory system — fromreceptor to decision. Prog Neurobiol 2013, 103:28-40.
4. Kleinfeld D, Ahissar E, Diamond ME: Active sensation: insightsfrom the rodent vibrissa sensorimotor system. Curr OpinNeurobiol 2006, 16:435-444.
5. Diamond ME, von Heimendahl M, Knutsen PM, Kleinfeld D,Ahissar E: ‘Where’ and ‘what’ in the whisker sensorimotorsystem. Nat Rev Neurosci 2008, 9:601-612.
6.��
Bagdasarian K, Szwed M, Knutsen PM, Deutsch D, Derdikman D,Pietr M, Simony E, Ahissar E: Pre-neuronal morphologicalprocessing of object location by individual whiskers. NatNeurosci 2013, 16:622-631.
A systematic examination of the transformation of object contact intowhisker shape. By high-speed videography and tracking, the authorsfound that angle and curvature can represent the coordinates of objectposition after contact in a reliable manner. Moreover, the follicle-whiskerjunction was found to be rigid, which enables direct readout of whiskermorphological coding by mechanoreceptors.
7.��
Pammer L, O’Connor DH, Hires SA, Clack NG, Huber D,Myers EW, Svoboda K: The mechanical variables underlyingobject localization along the axis of the whisker. J Neurosci2013, 33:6726-6741.
High-speed videography of whisker motion in head-fixed mice allowedestimation of the forces and bending moments at the base of the whiskeras a function of the radial distance of the contacted pole. Similarly to Ref.[6], the relative amplitudes of different stress variables change dramati-cally with radial object distance.
8. Williams CM, Kramer EM: The advantages of a tapered whisker.PLoS ONE 2010, 5:e8806.
9.�
Hires SA, Pammer L, Svoboda K, Golomb D: Tapered whiskersare required for active tactile sensation. Elife 2013, 2:e01350.
10. Voges D, Carl K, Klauer GJ, Uhlig R, Schilling C, Behn C, Witte H:Structural characterization of the whisker system of the rat.Sensor J IEEE 2012, 12:332-339.
11. Perkon I, Kosir A, Itskov PM, Tasic J, Diamond ME: Unsupervisedquantification of whisking and head movement in freelymoving rodents. J Neurophysiol 2011, 105:1950-1962.
12. Clack NG, O’Connor DH, Huber D, Petreanu L, Hires A, Peron S,Svoboda K, Myers EW: Automated tracking of whiskers invideos of head fixed rodents. PLoS Comput Biol 2012,8:e1002591.
13. Birdwell JA, Solomon JH, Thajchayapong M, Taylor MA, Cheely M,Towal RB, Conradt J, Hartmann MJ: Biomechanical models forradial distance determination by the rat vibrissal system. JNeurophysiol 2007, 98:2439-2455.
14. Quist BW, Hartmann MJ: Mechanical signals at the base of a ratvibrissa: the effect of intrinsic vibrissa curvature andimplications for tactile exploration. J Neurophysiol 2012,107:2298-2312.
Current Opinion in Neurobiology 2014, 25:176–186
184 Theoretical and computational neuroscience
15. Boubenec Y, Shulz DE, Debregeas G: Whisker encoding ofmechanical events during active tactile exploration. FrontBehav Neurosci 2012, 6:74.
16. Solomon JH, Hartmann MJ: Radial distance determination inthe rat vibrissal system and the effects of Weber’s law. PhilosTrans R Soc Lond B Biol Sci 2011, 366:3049-3057.
17. Arabzadeh E, Zorzin E, Diamond ME: Neuronal encoding oftexture in the whisker sensory pathway. PLoS Biol 2005, 3:e17.
18. Jadhav SP, Wolfe J, Feldman DE: Sparse temporal coding ofelementary tactile features during active whisker sensation.Nat Neurosci 2009, 12:792-800.
19. Morita T, Kang H, Wolfe J, Jadhav SP, Feldman DE: Psychometriccurve and behavioral strategies for whisker-based texturediscrimination in rats. PLoS ONE 2011, 6:e20437.
20.��
Zuo Y, Perkon I, Diamond ME: Whisking and whisker kinematicsduring a texture classification task. Philos Trans R Soc Lond BBiol Sci 2011, 366:3058-3069.
During judgment of spatial features on a grooved surface, distinct stick/slip events occurred during texture palpation and their frequencyincreased in relation to the spatial frequency of the grooves. The resultsallow a preliminary assessment of three candidate texture-codingmechanisms—the number of grooves encountered during each touch,the temporal difference between groove contacts and the spatial patternof groove contacts across the whiskers.
21. Jadhav SP, Feldman DE: Texture coding in the whisker system.Curr Opin Neurobiol 2010, 20:313-318.
22. Lottem E, Azouz R: Mechanisms of tactile informationtransmission through whisker vibrations. J Neurosci 2009,29:11686-11697.
23. Jones LM, Depireux DA, Simons DJ, Keller A: Robust temporalcoding in the trigeminal system. Science 2004, 304:1986-1989.
24. Szwed M, Bagdasarian K, Blumenfeld B, Barak O, Derdikman D,Ahissar E: Responses of trigeminal ganglion neurons to theradial distance of contact during active vibrissal touch. JNeurophysiol 2006, 95:791-802.
25. Stuttgen MC, Kullmann S, Schwarz C: Responses of rattrigeminal ganglion neurons to longitudinal whiskerstimulation. J Neurophysiol 2008, 100:1879-1884.
26. Bale MR, Petersen RS: Transformation in the neural code forwhisker deflection direction along the lemniscal pathway. JNeurophysiol 2009, 102:2771-2780.
27. Szwed M, Bagdasarian K, Ahissar E: Encoding of vibrissal activetouch. Neuron 2003, 40:621-630.
28. Leiser SC, Moxon KA: Responses of trigeminal ganglionneurons during natural whisking behaviors in the awake rat.Neuron 2007, 53:117-133.
29. Khatri V, Bermejo R, Brumberg JC, Keller A, Zeigler HP: Whiskingin air: encoding of kinematics by trigeminal ganglion neuronsin awake rats. J Neurophysiol 2009, 101:1836-1846.
30.��
Bale MR, Davies K, Freeman OJ, Ince RAA, Petersen RS: Low-dimensional sensory feature representation by trigeminalprimary afferents. J Neurosci 2013, 33:12003-12012.
Feature selectivity of TG neurons was analyzed using a generalized linearmodel that included linear filters acting on the stimulus and the spikehistory and could predict responses both to ‘noise’ stimuli and to replayednatural whisker motion on textures. The spectrum of TG feature selectivitydetermines the types of information provided to the rest of the whiskerpathway.
31.�
Sakurai K, Akiyama M, Cai B, Scott A, Han BX, Takatoh J,Sigrist M, Arber S, Wang F: The organization of submodality-specific touch afferent inputs in the vibrissa column. Cell Rep2013, 5:87-98.
32. Jones LM, Lee S, Trageser JC, Simons DJ, Keller A: Precisetemporal responses in whisker trigeminal neurons. JNeurophysiol 2004, 92:665-668.
33. Lottem E, Azouz R: A unifying framework underlyingmechanotransduction in the somatosensory system. JNeurosci 2011, 31:8520-8532.
Current Opinion in Neurobiology 2014, 25:176–186
34. Arabzadeh E, Panzeri S, Diamond ME: Deciphering the spiketrain of a sensory neuron: counts and temporal patterns in therat whisker pathway. J Neurosci 2006, 26:9216-9226.
35. von Heimendahl M, Itskov PM, Arabzadeh E, Diamond ME:Neuronal activity in rat barrel cortex underlying texturediscrimination. PLoS Biol 2007, 5:e305.
36. Prescott TJ, Diamond ME, Wing AM: Active touch sensing. PhilosTrans R Soc Lond B Biol Sci 2011, 366:2989-2995.
37. Lee S, Carvell GE, Simons DJ: Motor modulation of afferentsomatosensory circuits. Nat Neurosci 2008, 11:1430-1438.
38. Matyas F, Sreenivasan V, Marbach F, Wacongne C, Barsy B,Mateo C, Aronoff R, Petersen CC: Motor control by sensorycortex. Science 2010, 330:1240-1243.
39. Kleinfeld D, Deschenes M: Neuronal basis for object location inthe vibrissa scanning sensorimotor system. Neuron 2011,72:455-468.
40. Huber D, Gutnisky DA, Peron S, O’Connor DH, Wiegert JS, Tian L,Oertner TG, Looger LL, Svoboda K: Multiple dynamicrepresentations in the motor cortex during sensorimotorlearning. Nature 2012, 484:473-478.
41.�
Petreanu L, Gutnisky DA, Huber D, Xu NL, O’Connor DH, Tian L,Looger L, Svoboda K: Activity in motor-sensory projectionsreveals distributed coding in somatosensation. Nature 2012,489:299-303.
42. Grant RA, Mitchinson B, Fox CW, Prescott TJ: Active touchsensing in the rat: anticipatory and regulatory control ofwhisker movements during surface exploration. J Neurophysiol2009, 101:862-874.
43. Jenks RA, Vaziri A, Boloori AR, Stanley GB: Self-motion and theshaping of sensory signals. J Neurophysiol 2010,103:2195-2207.
44. Hutson KA, Masterton RB: The sensory contribution of a singlevibrissa’s cortical barrel. J Neurophysiol 1986, 56:1196-1223.
45. Carvell GE, Simons DJ: Biometric analyses of vibrissal tactilediscrimination in the rat. J Neurosci 1990, 10:2638-2648.
46. Knutsen PM, Pietr M, Ahissar E: Haptic object localization in thevibrissal system: behavior and performance. J Neurosci 2006,26:8451-8464.
47. O’Connor DH, Clack NG, Huber D, Komiyama T, Myers EW,Svoboda K: Vibrissa-based object localization in head-fixedmice. J Neurosci 2010, 30:1947-1967.
48. Yu C, Horev G, Rubin N, Derdikman D, Haidarliu S, Ahissar E:Coding of object location in the vibrissal thalamocorticalsystem. Cereb Cortex 2013, http://dx.doi.org/10.1093/cercor/bht241, first published online September 22, 2013.
49. Stuttgen MC, Ruter J, Schwarz C: Two psychophysical channelsof whisker deflection in rats align with two neuronal classes ofprimary afferents. J Neurosci 2006, 26:7933-7940.
50. Stuttgen MC, Schwarz C: Psychophysical and neurometricdetection performance under stimulus uncertainty. NatNeurosci 2008, 11:1091-1100.
51. Adibi M, Arabzadeh E: A comparison of neuronal and behavioraldetection and discrimination performances in rat whiskersystem. J Neurophysiol 2011, 105:356-365.
52. Adibi M, Diamond ME, Arabzadeh E: Behavioral study ofwhisker-mediated vibration sensation in rats. Proc Natl AcadSci U S A 2012, 109:971-976.
53. Mayrhofer JM, Skreb V, von der Behrens W, Musall S, Weber B,Haiss F: Novel two-alternative forced choice paradigm forbilateral vibrotactile whisker frequency discrimination inhead-fixed mice and rats. J Neurophysiol 2013, 109:273-284.
54. Miyashita T, Feldman DE: Behavioral detection of passivewhisker stimuli requires somatosensory cortex. Cereb Cortex2013, 23:1655-1660.
55. Krupa DJ, Matell MS, Brisben AJ, Oliveira LM, Nicolelis MA:Behavioral properties of the trigeminal somatosensory
www.sciencedirect.com
Algorithms of touch perception Maravall and Diamond 185
system in rats performing whisker-dependent tactilediscriminations. J Neurosci 2001, 21:5752-5763.
56. Sharpee TO: Computational identification of receptive fields.Annu Rev Neurosci 2013, 36:103-120.
57. Maravall M, Alenda A, Bale MR, Petersen RS: Transformation ofadaptation and gain rescaling along the whisker sensorypathway. PLoS ONE 2013, 8:e82418.
58. Theis L, Chagas AM, Arnstein D, Schwarz C, Bethge M: BeyondGLMs: a generative mixture modeling approach to neuralsystem identification. PLoS Comput Biol 2013, 9:e1003356.
59. Petersen RS, Brambilla M, Bale MR, Alenda A, Panzeri S,Montemurro MA, Maravall M: Diverse and temporally precisekinetic feature selectivity in the VPm thalamic nucleus. Neuron2008, 60:890-903.
60. Maravall M, Petersen RS, Fairhall AL, Arabzadeh E, Diamond ME:Shifts in coding properties and maintenance of informationtransmission during adaptation in barrel cortex. PLoS Biol2007, 5:e19.
61.��
Estebanez L, El Boustani S, Destexhe A, Shulz DE: Correlatedinput reveals coexisting coding schemes in a sensory cortex.Nat Neurosci 2012, 15:1691-1699.
Systematic application of different spatiotemporal patterns of stimuli tomultiple (24) whiskers, combined with reverse correlation analysis, pro-vided a unified framework capturing multiple response properties in barrelcortex. Several functional categories of neurons emerged, grouped bytheir responses to whisker dynamics and by their selectivity to the degreeof correlated stimulation of multiple whiskers.
62. Petersen RS, Panzeri S, Maravall M: Neural coding andcontextual influences in the whisker system. Biol Cybern 2009,100:427-446.
63. Jacob V, Le Cam J, Ego-Stengel V, Shulz DE: Emergentproperties of tactile scenes selectively activate barrel cortexneurons. Neuron 2008, 60:1112-1125.
64. Ego-Stengel V, Le Cam J, Shulz DE: Coding of apparent motionin the thalamic nucleus of the rat vibrissal somatosensorysystem. J Neurosci 2012, 32:3339-3351.
65. Hirata A, Castro-Alamancos MA: Cortical transformation ofwide-field (multiwhisker) sensory responses. J Neurophysiol2008, 100:358-370.
66. Crochet S, Petersen CC: Correlating whisker behavior withmembrane potential in barrel cortex of awake mice. NatNeurosci 2006, 9:608-610.
67. Curtis JC, Kleinfeld D: Phase-to-rate transformations encodetouch in cortical neurons of a scanning sensorimotor system.Nat Neurosci 2009, 12:492-501.
68.��
O’Connor DH, Peron SP, Huber D, Svoboda K: Neural activity inbarrel cortex underlying vibrissa-based object localization inmice. Neuron 2010, 67:1048-1061.
During a pole location discrimination task in mice, a small fraction ofneurons fired most of the spikes recorded in barrel cortex: the most active10% generated approximately 50% of spikes. A neuron’s activity levelcorrelated with its ability to discriminate trial type: both activity level anddiscriminative ability were lower in layers 2/3 and 6 than in layers 4 and 5.However, all layers contained neurons that responded strongly and werehighly discriminative.
69.�
Chen JL, Carta S, Soldado-Magraner J, Schneider BL,Helmchen F: Behaviour-dependent recruitment of long-rangeprojection neurons in somatosensory cortex. Nature 2013,499:336-340.
70. de Kock CP, Bruno RM, Spors H, Sakmann B: Layer- and cell-type-specific suprathreshold stimulus representation inrat primary somatosensory cortex. J Physiol 2007,581:139-154.
71. de Kock CP, Sakmann B: Spiking in primary somatosensorycortex during natural whisking in awake head-restrained ratsis cell-type specific. Proc Natl Acad Sci U S A 2009,106:16446-16450.
72. Barth AL, Poulet JF: Experimental evidence for sparse firing inthe neocortex. Trends Neurosci 2012, 35:345-355.
www.sciencedirect.com
73. Wolfe J, Houweling AR, Brecht M: Sparse and powerful corticalspikes. Curr Opin Neurobiol 2010, 20:306-312.
74. Ganguli S, Sompolinsky H: Compressed sensing, sparsity, anddimensionality in neuronal information processing and dataanalysis. Annu Rev Neurosci 2012, 35:485-508.
75. Harris KD, Mrsic-Flogel TD: Cortical connectivity and sensorycoding. Nature 2013, 503:51-58.
76. Quiroga RQ: Concept cells: the building blocks of declarativememory functions. Nat Rev Neurosci 2012, 13:587-597.
77. Petersen CC, Crochet S: Synaptic computation and sensoryprocessing in neocortical layer 2/3. Neuron 2013, 78:28-48.
78.��
Safaai H, von Heimendahl M, Sorando JM, Diamond ME,Maravall M: Coordinated population activity underlying texturediscrimination in rat barrel cortex. J Neurosci 2013,33:5843-5855.
This study used ideal observer analysis to determine the trial discrimina-tion performance of groups of neuronal single or multiple units recordedwhile a rat performed a texture discrimination task. The collective signalwas analyzed in terms of the net redundancy or synergy present and itsdependence on the amount of heterogeneity between neurons. Informa-tion in pooled populations saturated at �20 neurons, at which pointredundancy set in. The population size at which representations transitionfrom synergy to redundancy may depend on sensory modality, taskdifficulty, and degree of training.
79. Yamashita T, Pala A, Pedrido L, Kremer Y, Welker E, Petersen CC:Membrane potential dynamics of neocortical projectionneurons driving target-specific signals. Neuron 2013,80:1477-1480.
80. Andermann ML, Moore CI: A somatotopic map of vibrissamotion direction within a barrel column. Nat Neurosci 2006,9:543-551.
81. Kerr JN, de Kock CP, Greenberg DS, Bruno RM, Sakmann B,Helmchen F: Spatial organization of neuronal populationresponses in layer 2/3 of rat barrel cortex. J Neurosci 2007,27:13316-13328.
82. Sato TR, Gray NW, Mainen ZF, Svoboda K: The functionalmicroarchitecture of the mouse barrel cortex. PLoS Biol 2007,5:e189.
83.�
Crochet S, Poulet JF, Kremer Y, Petersen CC: Synapticmechanisms underlying sparse coding of active touch. Neuron2011, 69:1160-1175.
84. Margolis DJ, Lutcke H, Schulz K, Haiss F, Weber B, Kugler S,Hasan MT, Helmchen F: Reorganization of cortical populationactivity imaged throughout long-term sensory deprivation. NatNeurosci 2012, 15:1539-1546.
85. Yassin L, Benedetti BL, Jouhanneau JS, Wen JA, Poulet JF,Barth AL: An embedded subnetwork of highly active neurons inthe neocortex. Neuron 2010, 68:1043-1050.
86. Montemurro MA, Panzeri S, Maravall M, Alenda A, Bale MR,Brambilla M, Petersen RS: Role of precise spike timing in codingof dynamic vibrissa stimuli in somatosensory thalamus. JNeurophysiol 2007, 98:1871-1882.
87. Bruno RM, Sakmann B: Cortex is driven by weak butsynchronously active thalamocortical synapses. Science 2006,312:1622-1627.
88. Fee MS, Mitra PP, Kleinfeld D: Central versus peripheraldeterminants of patterned spike activity in rat vibrissa cortexduring whisking. J Neurophysiol 1997, 78:1144-1149.
89. Hill DN, Curtis JC, Moore JD, Kleinfeld D: Primary motor cortexreports efferent control of vibrissa motion on multipletimescales. Neuron 2011, 72:344-356.
90. Gentet LJ, Avermann M, Matyas F, Staiger JF, Petersen CC:Membrane potential dynamics of GABAergic neurons in thebarrel cortex of behaving mice. Neuron 2010, 65:422-435.
91. Knutsen PM, Ahissar E: Orthogonal coding of object location.Trends Neurosci 2009, 32:101-109.
92. Panzeri S, Diamond ME: Information carried by populationspike times in the whisker sensory cortex can be decoded
Current Opinion in Neurobiology 2014, 25:176–186
186 Theoretical and computational neuroscience
without knowledge of stimulus time. Front Synaptic Neurosci2010, 2:17.
93. Panzeri S, Ince RAA, Diamond ME, Kayser C: Reading spike timingwithout a clock: intrinsic decoding of spike trains. Phil Trans R SocB 2014, 369 20120467. http://dx.doi.org/10.1098/rstb.2012.0467.
94. Panzeri S, Petersen RS, Schultz SR, Lebedev M, Diamond ME:The role of spike timing in the coding of stimulus location in ratsomatosensory cortex. Neuron 2001, 29:769-777.
95. Petersen RS, Panzeri S, Diamond ME: Population coding ofstimulus location in rat somatosensory cortex. Neuron 2001,32:503-514.
96. Storchi R, Bale MR, Biella GE, Petersen RS: Comparison oflatency and rate coding for the direction of whisker deflectionin the subcortical somatosensory pathway. J Neurophysiol2012, 108:1810-1820.
97. Temereanca S, Brown EN, Simons DJ: Rapid changes inthalamic firing synchrony during repetitive whiskerstimulation. J Neurosci 2008, 28:11153-11164.
98. Khatri V, Bruno RM, Simons DJ: Stimulus-specific and stimulus-nonspecific firing synchrony and its modulation by sensoryadaptation in the whisker-to-barrel pathway. J Neurophysiol2009, 101:2328-2338.
99. Stanley GB, Jin J, Wang Y, Desbordes G, Wang Q, Black MJ,Alonso JM: Visual orientation and directional selectivitythrough thalamic synchrony. J Neurosci 2012, 32:9073-9088.
100. Brette R: Computing with neural synchrony. PLoS Comput Biol2012, 8:e1002561.
101. Rossant C, Leijon S, Magnusson AK, Brette R: Sensitivity of noisyneurons to coincident inputs. J Neurosci 2011, 31:17193-17206.
102. Song S, Sjostrom PJ, Reigl M, Nelson S, Chklovskii DB: Highlynonrandom features of synaptic connectivity in local corticalcircuits. PLoS Biol 2005, 3:e68.
103. Lefort S, Tomm C, Floyd Sarria JC, Petersen CC: The excitatoryneuronal network of the C2 barrel column in mouse primarysomatosensory cortex. Neuron 2009, 61:301-316.
104. Diaz-Quesada M, Martini FJ, Ferrati G, Bureau I, Maravall M:Diverse thalamocortical short-term plasticity elicited byongoing stimulation. J Neurosci 2014, 34:515-526.
105�
O’Connor DH, Hires SA, Guo ZV, Li N, Yu J, Sun QQ, Huber D,Svoboda K: Neural coding during active somatosensationrevealed using illusory touch. Nat Neurosci 2013, 16:958-965.
106. Diamond ME, von Heimendahl M, Arabzadeh E: Whisker-mediated texture discrimination. PLoS Biol 2008, 6:e220.
107. Maravall M: Adaptation and sensory coding. In Principles ofneural coding. Edited by Quian Quiroga R, Panzeri S. Boca Raton:Taylor & Francis/CRC Press; 2013:357-378.
108. Katz Y, Heiss JE, Lampl I: Cross-whisker adaptation of neuronsin the rat barrel cortex. J Neurosci 2006, 26:13363-13372.
109. Khatri V, Simons DJ: Angularly nonspecific responsesuppression in rat barrel cortex. Cereb Cortex 2007,17:599-609.
110. Wang Q, Webber RM, Stanley GB: Thalamic synchrony and theadaptive gating of information flow to cortex. Nat Neurosci2010, 13:1534-1541.
111. Adibi M, McDonald JS, Clifford CW, Arabzadeh E: Adaptationimproves neural coding efficiency despite increasingcorrelations in variability. J Neurosci 2013, 33:2108-2120.
Current Opinion in Neurobiology 2014, 25:176–186
112. Adibi M, Clifford CW, Arabzadeh E: Informational basis ofsensory adaptation: entropy and single-spike efficiency in ratbarrel cortex. J Neurosci 2013, 33:14921-14926.
113. Lak A, Arabzadeh E, Diamond ME: Enhanced response ofneurons in rat somatosensory cortex to stimuli containingtemporal noise. Cereb Cortex 2008, 18:1085-1093.
114. Lak A, Arabzadeh E, Harris JA, Diamond ME: Correlatedphysiological and perceptual effects of noise in a tactilestimulus. Proc Natl Acad Sci U S A 2010, 107:7981-7986.
115. Garcia-Lazaro JA, Ho SS, Nair A, Schnupp JW: Shifting andscaling adaptation to dynamic stimuli in somatosensorycortex. Eur J Neurosci 2007, 26:2359-2368.
116. Lundstrom BN, Fairhall AL, Maravall M: Multiple timescaleencoding of slowly varying whisker stimulus envelope incortical and thalamic neurons in vivo. J Neurosci 2010,30:5071-5077.
117. Chung S, Li X, Nelson SB: Short-term depression atthalamocortical synapses contributes to rapid adaptation ofcortical sensory responses in vivo. Neuron 2002, 34:437-446.
118. Ganmor E, Katz Y, Lampl I: Intensity-dependent adaptation ofcortical and thalamic neurons is controlled by brainstemcircuits of the sensory pathway. Neuron 2010, 66:273-286.
119. Diaz-Quesada M, Maravall M: Intrinsic mechanisms for adaptivegain rescaling in barrel cortex. J Neurosci 2008, 28:696-710.
120. Mease RA, Famulare M, Gjorgjieva J, Moody WJ, Fairhall AL:Emergence of adaptive computation by single neurons in thedeveloping cortex. J Neurosci 2013, 33:12154-12170.
121. Abbott LF, Dayan P: The effect of correlated variability on theaccuracy of a population code. Neural Comput 1999, 11:91-101.
122. Sompolinsky H, Yoon H, Kang K, Shamir M: Population coding inneuronal systems with correlated noise. Phys Rev E Stat NonlinSoft Matter Phys 2001, 64:051904.
123. Shamir M, Sompolinsky H: Implications of neuronal diversity onpopulation coding. Neural Comput 2006, 18:1951-1986.
124. Chelaru MI, Dragoi V: Efficient coding in heterogeneousneuronal populations. Proc Natl Acad Sci U S A 2008,105:16344-16349.
125. Hung CP, Kreiman G, Poggio T, DiCarlo JJ: Fast readout ofobject identity from macaque inferior temporal cortex. Science2005, 310:863-866.
126. Meyers EM, Freedman DJ, Kreiman G, Miller EK, Poggio T:Dynamic population coding of category information in inferiortemporal and prefrontal cortex. J Neurophysiol 2008, 100:1407-1419.
127. Nikolic D, Hausler S, Singer W, Maass W: Distributed fadingmemory for stimulus properties in the primary visual cortex.PLoS Biol 2009, 7:e1000260.
128�
Guo ZV, Li N, Huber D, Ophir E, Gutnisky D, Ting JT, Feng G,Svoboda K: Flow of cortical activity underlying a tactiledecision in mice. Neuron 2014, 81:179-194.
129��
Fassihi A, Akrami A, Esmaeili V, Diamond ME: Tactile perceptionand working memory in rats and humans. Proc Natl Acad Sci US A 2014 http://dx.doi.org/10.1073/pnas.1315171111. publishedahead of print January 21, 2014.
Using their whiskers to receive vibrations, rats show sensory acuity andworking memory capacities that rival those of human subjects using theirfingertips. The study will allow the study of neuronal mechanisms under-lying a new form of tactile perception in rats.
www.sciencedirect.com