Skip to main content
eLife logoLink to eLife
. 2015 Aug 6;4:e06619. doi: 10.7554/eLife.06619

Low-noise encoding of active touch by layer 4 in the somatosensory cortex

Samuel Andrew Hires 1,†,, Diego A Gutnisky 1,, Jianing Yu 1, Daniel H O'Connor 1,§, Karel Svoboda 1,*
Editor: Matteo Carandini2
PMCID: PMC4525079  PMID: 26245232

Abstract

Cortical spike trains often appear noisy, with the timing and number of spikes varying across repetitions of stimuli. Spiking variability can arise from internal (behavioral state, unreliable neurons, or chaotic dynamics in neural circuits) and external (uncontrolled behavior or sensory stimuli) sources. The amount of irreducible internal noise in spike trains, an important constraint on models of cortical networks, has been difficult to estimate, since behavior and brain state must be precisely controlled or tracked. We recorded from excitatory barrel cortex neurons in layer 4 during active behavior, where mice control tactile input through learned whisker movements. Touch was the dominant sensorimotor feature, with >70% spikes occurring in millisecond timescale epochs after touch onset. The variance of touch responses was smaller than expected from Poisson processes, often reaching the theoretical minimum. Layer 4 spike trains thus reflect the millisecond-timescale structure of tactile input with little noise.

DOI: http://dx.doi.org/10.7554/eLife.06619.001

Research organism: mouse

eLife digest

Cells called neurons connect to form large networks that process information in the brain. A region of the brain called the cerebral cortex receives information about touch from sensors in the skin. A series of neurons relay the touch information to the cerebral cortex as patterns of electrical activity called ‘spike trains’. Understanding how these spike trains represent information about the world around us is one of the greatest challenges facing neuroscience.

At first glance, the number and timing of the individual spikes within the trains appear to be random. It is possible that the irregularity within spike trains is ‘noise’ that is generated within the cortex itself. This noise could represent uncertainty about the nature of the stimulus from the sensors, or random fluctuations in brain activity. However, other findings have challenged this view and argued that these erratic spike trains actually carry hidden information.

Hires et al. investigated this possibility by recording how neurons within a region of the mouse brain called the somatosensory cortex responded to sensory information coming from the mouse's whiskers. Mice sweep their whiskers across objects to locate and identify them, much like how humans feel objects with their fingertips. Here, the mice used their whiskers to judge the location of an object by touch alone, while the electrical activity of the neurons was measured using electrodes. Importantly, the movements of the whiskers and contact with the object were tracked to one millisecond precision.

Similar to previous studies, sensory information from the whiskers triggered irregular spike trains in neurons within the somatosensory cortex. Hires et al. found that the apparently irregular spikes coincided precisely with the timing of when the whiskers contacted the object. Other spikes aligned perfectly with the movement of whiskers into particular positions. Furthermore, the patterns of electrical activity in the spike trains precisely predicted when and how the object was contacted, and which whisker was involved.

These findings suggest that the timing of individual spikes within spike trains carries important information to the brain. Future studies will develop our understanding of how the brain interprets and responds to the rich data contained in these spike trains to identify objects and decide how to interact with them.

DOI: http://dx.doi.org/10.7554/eLife.06619.002

Introduction

Variability in spike trains constrains how neural computations can be implemented (London et al., 2010; Renart and Machens, 2014). Measured cortical spike trains are often irregular in time, and spike counts vary over repeated presentations of identical sensory stimuli (Tolhurst et al., 1983; Shadlen and Newsome, 1998; Maimon and Assad, 2009). One view holds that this variability is irreducible and therefore represents noise (Shadlen and Newsome, 1998; London et al., 2010). Noisy spike trains are difficult to reconcile with the integrative properties of single neurons (Softky and Koch, 1993) and the high reliability of cortical neurons (Mainen and Sejnowski, 1995) and synapses (Stevens and Zador, 1998). This discrepancy has motivated models of cortical circuits that inherently produce noisy spike trains even with reliable neurons, by virtue of chaotic dynamics (van Vreeswijk and Sompolinsky, 1996; Shadlen and Newsome, 1998; Litwin-Kumar and Doiron, 2012). Irregular spike trains suggest that spike rates, but not spike timing, are used by the brain for computation (Mazurek and Shadlen, 2002; London et al., 2010).

Another view contends that variability in cortical spike trains is not noise but reflects fluctuating hidden states with possible behavioral significance and uncontrolled experimental factors (Gur et al., 1997; Kara et al., 2000; DeWeese et al., 2003; VanRullen et al., 2005; Amarasingham et al., 2006). Measuring and accounting for these hidden states may reveal the detailed structure of spike trains to be deterministic and predictable. For example, minimizing fixational eye movements in alert monkeys reduces spike count variability in the visual cortex (Gur et al., 1997). In the sensory periphery, spikes are often coupled to stimulus features with high temporal precision. This precision allows timing-based neural codes to be faster (Johansson and Birznieks, 2004) and more efficient (Gollisch and Meister, 2008) than rate-based codes.

Multiple factors can shape neural spike trains, only a subset of which are controlled or measured in typical experiments. Uncontrolled factors will add to measured variability (Masquelier, 2013; Renart and Machens, 2014). These factors can be external to the brain, such as sensory stimuli (Baudot et al., 2013), or internal to it, such as behavioral state (Mitchell et al., 2007, 2009; Churchland et al., 2010). They can be fundamentally irreducible, such as channel noise and chaotic dynamics of neural networks (van Vreeswijk and Sompolinsky, 1996), or potentially controllable, such as animal behavior (Gur et al., 1997) and fluctuating input from other brain areas (Gomez et al., 2013). Here we assessed the precision of spikes during active tactile behavior. We recorded from neurons in layer 4 (L4) in the mouse barrel cortex and measured tactile behaviors with high temporal and spatial resolution.

L4 contains a precise map, where individual barrels process information from single whiskers (Simons, 1978). Similar to other cortical circuits, connectivity within L4 is highly recurrent (Lefort et al., 2009). However, the only major long-range input into L4 ascends from sensory neurons via the posterior medial thalamus into L4 (Lu and Lin, 1993; Bureau et al., 2006; Hooks et al., 2011). L4 thus receives little uncontrolled input.

During natural behavior animals move their sensors to acquire information about the world (Deschenes et al., 2012). Measurement of spiking statistics in the barrel cortex is thus most meaningful during active sensation, when mice shape sensory input by moving their whiskers to solve a tactile task. In our experiments mice localized objects with their whiskers (Knutsen et al., 2006; O'Connor et al., 2010a; O'Connor et al., 2013). Activity within the barrel cortex is necessary for whisker-based sensation (Hutson and Masterton, 1986; O'Connor et al., 2010; Guo et al., 2014). Whisker deflections (Simons, 1978; Bruno and Sakmann, 2006; Jadhav et al., 2009) or active touch (Crochet and Petersen, 2006; Curtis and Kleinfeld, 2009; O'Connor et al., 2010b; O'Connor et al., 2013) trigger temporally sharp responses in barrel cortex neurons, which underlie the perception of object location (Diamond et al., 2008; O'Connor et al., 2013). Spike rates of barrel cortex are also modulated by whisker movements on multiple time scales (Fee et al., 1997; Crochet and Petersen, 2006; Curtis and Kleinfeld, 2009). Similar to visual cortical neurons (Tolhurst et al., 1983; Shadlen and Newsome, 1998; Maimon and Assad, 2009), barrel cortex neurons respond with high trial-to-trial variability to passive sensory stimulation (Wang et al., 2010; Adibi et al., 2013).

To minimize uncontrolled variability we thus recorded cortical spike trains in a well-characterized neural circuit, in mice engaged in active sensorimotor behavior, with precisely quantified sensory input. We show that spikes in L4 are temporally precise and have spike count variance close to the theoretical minimum. This precision allows efficient decoding of touch timing from small numbers of L4 neurons, supporting a role for temporal coding in cortical computation.

Results

L4 responses are temporally precise

We trained mice to locate an object by active touch with a single whisker (C2) (O'Connor et al., 2010a; O'Connor et al., 2013) (n = 21 mice, 52 sessions; fraction of trials correct, 0.740 ± 0.086; mean ± s.d.; Figure 1A; Figure 1—figure supplement 1). Single whisker experiments allowed us to track the relevant tactile variables with high precision during behavior. In each trial, during a sample epoch lasting a few seconds (1.54–4.05, mean 2.39 s), a pole appeared in one of two locations on the right side of the head. High-speed videography and automated whisker tracking quantified whisker movement (azimuthal angle, θ; whisking phase, ϕ), changes in curvature caused by the forces exerted by the pole on the whisker (change in curvature, ∆κ) (Birdwell et al., 2007; Pammer et al., 2013), and contact time, all with 1 millisecond temporal precision (Clack et al., 2012; O'Connor et al., 2013) (Figure 1A,B). Mice whisked in bouts (mean bout duration, 261 ms; peak-to-peak amplitude, 15.7°; frequency, 15.4 Hz) interspersed with periods of rest. Mice touched the pole multiple times (mean number of touches, 2.33) before reporting perceived object location with licking (mean reaction time 367 ± 234 ms; mean ± s.d.) (Figure 1—figure supplement 1).

Figure 1. Activity during tactile behavior in a layer 4 excitatory cell.

(A) Top, mice judged object location with a single whisker. Whisker position (azimuthal angle, θ), whisking phase (ϕ), and whisker curvature (κ) were measured from video recordings. Bottom, recordings were made from excitatory cells in the principal barrel (red). L4 excitatory neurons receive excitatory input from VPM and excite each other within individual barrels. (B) Behavioral and electrophysiological data (single trial). θ, whisker position (green); ϕ, whisking phase (green); Δκ, change in whisker curvature (blue), which is proportional to pressure on mechanoreceptors at the base of the whisker; Vex, extracellular spike waveform (black) recorded in loose-seal mode (blue crosses, touch onsets). The black horizontal bar indicates the time when the object was in reach. (C) Spike raster for one example neuron. Same data as in DH. Pole in reach for all trials (black bar) with variable exit time (grey bar). (D) Peri-stimulus time histogram aligned to the trial onset (bin size, 50 ms). (E) Spike raster aligned to first touch, and sorted according to last touch in the sample period (late on top). Trials without touch are not shown. (F) Peri-stimulus time histogram aligned to first touch (bin size, 1 ms). The grey line represents the proportion of touches with durations >= than time (max of 1). (G) Spikes aligned by whisking phase in a whisking bout (whisking amplitude >2.5° peak-to-peak). Only exploration periods excluding touch were used. (H) Spike histogram aligned to whisking phase (bin size, 30°) Best-fit spike modulation (grey). Average change in whisker position/bout (green).

DOI: http://dx.doi.org/10.7554/eLife.06619.003

Figure 1.

Figure 1—figure supplement 1. Behavior during cell-attached recordings.

Figure 1—figure supplement 1.

(A) Histogram of performing trials analyzed per session; x indicates mean value in all plots. (B) Histogram of behavioral performance for each session. (C) Histogram of the number of touches before the response time across all sessions. (D) Histogram of time between onset of first touch to first lick across all sessions. (E) Histogram of the maximum amplitude of each whisking bout across all sessions. (F) Histogram of the duration of each whisking bout across all sessions. (G) Histogram of the frequency of whisking (1/cycle length) for each whisk cycle across all sessions.
Figure 1—figure supplement 2. Targeting recordings.

Figure 1—figure supplement 2.

(A) Left, Intrinsic signal showing response to stimulated C2 whisker (630 nm illumination). Right, overlay of intrinsic signal (cyan) on skull image (530 nm illumination). (B) Left, example DiI marking of recording site. Right, cytochrome oxidase stain and DiI overlay for a recording in C2. (C) Same as B, for a recording on the D1/D2 border.

We targeted recordings to excitatory neurons in L4 of the principal barrel corresponding to the spared whisker (31 cells in C2; 10 additional outside C2) and L5 near the principal barrel (11 cells), guided by intrinsic signal imaging (number of trials per neuron 115 ± 60; Figure 1—figure supplement 2). We recorded single units via loose-seal juxtacellular methods to avoid potential artifacts of intracellular disruption from whole-cell patch or misassigned spikes from spike sorting (DeWeese et al., 2003). The dynamics of a typical neuron in L4 C2 is illustrated in Figure 1. The neuron had a low baseline firing (0.73 spk/s; 500 ms before start of sample epoch) that increased substantially during the sample epoch (2.6 fold; to 1.91 spk/s; p = 3.38e-7, Wilcoxon rank sum) (Figure 1C,D).

Neuronal variability can arise from external factors, such as trial-to-trial variations in behavior, or internal factors, such as synaptic noise and fluctuating motivation and arousal (Renart and Machens, 2014). The Fano factor (FF) is a widely used measure of variability in spike trains (Berry et al., 1997). FF is defined as the variance of the spike count divided by the mean spike count over some time window. For a Poisson process, FF = 1, independent of the window size. In our task, mice are free to explore the object differently in each trial. At a coarse scale, behavior and neural responses were irregular during object localization. FF computed by counting spikes over the entire sample epoch (stimulus presentation), was huge (FF = 7.51). Since each trial corresponds to different whisker movements and different patterns of touches this value of FF includes extrinsic variability due to behavior, in addition to intrinsic variability. Aligning spikes to the fine-scale structure of behavior revealed that spikes were mainly coupled to temporally irregular sensory input from object contact (Figure 1E). Spike rate was sharply elevated shortly after touch onset (Figure 1F). Each touch evoked on the order of one spike (1.51 spikes/first touch; 0.31/later touches) with short latency (onset, 8 ms) (Table 1; Figure 2—figure supplement 1). The large FF when computing spikes over the sample period is therefore at least in part due to trial-to-trial variability in active touch.

Table 1.

Spiking responses of recorded neurons

DOI: http://dx.doi.org/10.7554/eLife.06619.006

Area Spikes/touch first touch Spikes/touch later touch Spikes evoked (touch) Spikes evoked (touch and whisking) Phase modulation depth Non-whisking spike rate (spk/s) Whisking spike rate (spk/s) Onset latency (ms) Minimum ISI (ms)
L4 1.36 ± 1.32 0.74 ± 0.87 70.6 ± 20.9% 75.2 ± 19.0% 0.67 ± 0.32 1.31 ± 2.88 1.39 ± 2.33 7.8 ± 3.0 2.9 ± 1.9
Inside C2 0.87 0.33 74.7% 80.3% 0.70 0.32 0.33 8 2.3
(n=31) (0.08–5.77) (0.10–2.92) (24.5–96.0%) (35.3–96.8%) (0–1.00) (0.02–13.7) (0.02–10.9) (4–18) (1.2–11.0)
L4 0.08 ± 0.19 0.02 ± 0.04 9.2 ± 13.7% 33.0 ± 30.2% 0.58 ± 0.36 2.20 ± 2.91 2.41 ± 3.48 18.5 ± 6.0 3.3 ± 1.3
Outside C2 0 0 3.6% 23.8% 0.61 1.10 1.31 18 2.8
(n=10) (0–0.61) (0–0.12) (0–41.0%) (1.0–79.4%) (0.10–1.00) (0.03–8.86) (0.17–11.8) (12–26) (2.3–6.7)
L5 1.95 ± 2.96 1.28 ± 1.88 24.0–18.6% 31.5 ± 16.4% 0.18 ± 0.12 12.0 ± 9.87 16.2 ± 15.3 9.7 ± 5.2 3.7 ± 2.5
Near C2 0.98 0.42 24.4% 36.0% 0.13 12.8 13.5 8 2.7
(n=11) (0–8.68) (0–5.47) (0.5–55.5%) (1.4–56.1%) (0.06–0.46) (0.23–29.7) (1.00–54.0) (4–20) (1.6–9.1)

Mean ± standard deviation; median; (range).

Spike times outside of touch periods also appeared temporally irregular. Firing rates were low in both whisking (0.67 spikes/s) and non-whisking (0.23 spikes/s) periods. Yet during active exploration, the timing of spikes was coupled to the phase of rhythmic whisker movement (Fee et al., 1997; Curtis and Kleinfeld, 2009), with a modulation depth close to 1 (Figure 1G,H; Figure 2—figure supplement 1). For the example neuron, using whisker behavior it is possible to predict time windows where a neuron will fire a single or small number of spikes, as well as time periods when the spike probability is zero (Figure 1G,H; phase, 0), similar to spike trains measured in the salamander retina (Keat et al., 2001). In this sense the timing of each spike encodes a tactile feature (touch and whisking phase).

The temporally precise spiking after touch was restricted to L4 neurons in the principal barrel. Neurons recorded in the C2 barrel column showed brief responses to touch (Figure 2A–C; Figure 3—figure supplement 1; Table 1). In L4, but outside of C2, touch responses were much weaker (first touch, p = 1.4e-5; later touches, p = 6.32e-6; Wilcoxon rank sum). Layer 5 neurons near C2 had much higher firing rates, with touch responses that were more diverse (Figure 2B; Figure 3—figure supplement 2; Table 1). Modulation by whisking phase was not significantly different between L4 neurons inside and outside the C2 barrel (p = 0.68), but both were significantly more phase modulated than L5 (p = 4.3e-5 and p = 0.012 respectively, Wilcoxon rank sum) (Figure 2—figure supplement 1). There were no significant changes in firing rate between whisking and non-whisking across the L4 population (L4 inside C2, p = 0.75; L4 outside C2, p = 0.92), whereas L5 showed a modest, but significant increase with whisking (p = 0.019, Wilcoxon signed rank).

Figure 2. Neural responses to behavioral variables across three populations.

(A) Grand mean peri-stimulus time histogram. Top, L4 in the C2 barrel (31 neurons); middle, L4 outside of C2 (10 neurons); bottom, in L5 near C2 (11 neurons). Pole in reach for all trials (black bar) and some trials (grey bar). (B) Peri-stimulus time histograms aligned to first touch. Top, neurons in L4 C2 sorted by time of peak touch response; middle, L4 neurons outside of C2; bottom, neurons in L5 near C2. Arrow head points to the same neuron as Figure 1B–H. (C) Peak spike rate after touch (1 ms bin) vs spike rate in the absence of whisking, individual cells (o), population means (x). Red, L4 inside C2; grey, L4 outside C2; yellow L5 near C2. (D) Spike rate during whisking compared to spike rate in the absence of whisking. Symbols as in C.

DOI: http://dx.doi.org/10.7554/eLife.06619.007

Figure 2.

Figure 2—figure supplement 1. Some population characteristics of all recordings.

Figure 2—figure supplement 1.

(A) Latency from touch onset to onset of spiking response for 31 L4 cells in C2 (red), four L4 outside C2 (grey) and eleven L5 near C2 (yellow). Latencies could not be determined for six L4 outside cells and one L5 cell due to lack of touch response. Same colors in B, C. (B) Modulation of spike rate by whisking phase. Angle of the peak phase response is plotted vs modulation depth. (C) Histogram of the minimum interspike interval measured for each neuron.

We estimated the proportion of spikes that were temporally coupled to sensory input for each L4 neuron during whisking or touch in the sample epoch (exploration time). We counted the fraction of spikes falling into a small time window after touch (Figure 3A,B). This proportion increases with expanding window size, resulting in an initially steep curve (Figure 3C). At some window size, the curve levels off as the proportion of touch spikes increases no faster than expected for a random spike train at the neuron's mean spike rate. This transition point defines the proportion of touch-coupled spikes. For the neuron illustrated in Figure 3A, 74.4% spikes fall into a time window spanning 8 to 40 ms after touch onset, comprising 15.3% of overall exploration time. This implies temporally sparse spiking (Berry et al., 1997). Over the L4 C2 population, 70.6% ± 20.9% (mean, s.d.) of spikes were coupled to touch in an average of 14.9% of exploration time. L4 neurons outside of C2 had significantly less touch-evoked spikes, 8.4% ± 13.3% in 3.5% of exploration time (p = 6.53e-6, Wilcoxon rank sum). L5 neurons in C2 also showed significantly less touch-evoked spikes, 24.0% ± 18.6% touch spikes in 19.0% of exploration time (p = 2.30e-5, Wilcoxon rank sum) (Figure 3C).

Figure 3. The majority of spikes in L4 excitatory neurons encode tactile information.

(A) Top, peri-stimulus time histogram of example cell from Figure 1 for all session touches with overlaid touch window (8–40 ms) (blue). Bottom, peri-stimulus spike histogram (PSTH) aligned to preferred whisking phase with overlaid window (green) aligned with the peak of the PSTH. Bins are normalized by occupancy. (B) Raster plot aligned to the sample period. Overlays: Blue, touch windows from A. Green, whisking phase windows from A, centered on the maximum in the phase-aligned spike histogram. Phase windows can be truncated at the margins of whisk cycles resulting in variable window lengths. (C) Proportion of spikes in touch window as a function of time in trial. Red, L4 inside C2; grey, L4 outside C2; yellow L5 near C2. Lines show the evolution of touch spikes as touch windows expand from the onset latency of each cell. Circles indicate the proportion of spikes coupled to touch at the final touch window size for each cell. (D) Proportion of spikes coupled to touch and whisking phase vs to touch alone. Colors as in C.

DOI: http://dx.doi.org/10.7554/eLife.06619.009

Figure 3.

Figure 3—figure supplement 1. Response characteristics of all L4 recordings in C2.

Figure 3—figure supplement 1.

(A) Peri-stimulus time histogram aligned to first touch onsets (left) or all touch onsets (right) in a session. Touch spike windows computed from all touches for each cell (blue). Red scale bar 0.05 spikes/ms. (B) Peri-stimulus spike histogram aligned to preferred whisking phase with whisking spike window (green). (C) Average spike rates for 12 phase bins (black) for each cell with sinusoidal fit (grey). Red scale bar 0.5 spikes/s. (D) Histogram of interspike intervals during each recording session. Depth is manipulator reading from pia.
Figure 3—figure supplement 2. Response characteristics of all L4 recordings outside of C2 and L5 recordings near C2.

Figure 3—figure supplement 2.

(A) Peri-stimulus time histogram aligned to first touch onsets (left) or all touch onsets (right) in a session. Touch spike windows computed from all touches for each cell (blue). Red scale bar 0.05 spikes/ms. (B) Peri-stimulus spike histogram aligned to preferred whisking phase with whisking spike window (green). (C) Average spike rates for twelve phase bins (black) for each cell with sinusoidal fit (grey). Red scale bar 0.5 spikes/s for L4 cells, 2 spikes/s for L5 cells. (D) Histogram of interspike intervals during each recording session. Depth is manipulator reading from pia.

Using a similar approach we measured the proportion of the remaining spikes coupled to the phase of whisker movement (Figure 3A,B,D). Overall, for neurons in C2 75.2 ± 19.0% of spikes were coupled to touch or whisking phase in an average of 22.0% of exploration time, significantly more than outside of C2 (33.0 ± 30.2 in 16.9% of time; p = 2.88e-4), or in L5 (31.5 ± 16.4 in 26.1% of time; p = 2.96e-5, Wilcoxon rank sum). Subtraction of the touch spikes from touch and phase spikes reveals that 4.6 ± 8.7% of spikes were purely phase coupled in L4 C2, vs 23.8 ± 27.9% outside of C2 and 7.6 ± 14.9% in L5. We conclude that during object localization, the majority of spikes in L4 excitatory neurons encode aspects of touch of the primary whisker, with many of the remaining spikes encoding whisking phase.

Decoding touch

The temporal precision of spiking in L4 could be used to extract behaviorally relevant information. We assessed the ability of simulated populations of L4 neurons to detect touch and discriminate touch timing and whisking phase.

We used a simple model based on resampling the recorded spike trains measured in L4 neurons (‘Materials and methods’). Pooling activity from only fifteen L4 neurons (out of approximately 1600 [Lefort et al., 2009]) in C2 was sufficient to detect 95% of touches (integration time, 10 ms) (Figure 4A). Touch detection by neurons from surrounding barrels (>200 µm from the principal whisker) was poor (Figure 4A). Beyond detection, a group of 200 neurons in C2, integrating in 10 ms windows after touch, also allowed decoding of elapsed time from touch with high precision (minimum 0.55 ms uncertainty with 95% confidence at 10 ms post-touch onset using a naïve Bayes decoder [Duda et al., 2001]) (Figure 4B; Figure 4—figure supplement 1). This implies that a decoder reading L4 activity can determine which whisker makes contact with millisecond temporal precision (Panzeri et al., 2014).

Figure 4. Decoding of touch and phase from L4 spikes.

(A) Decoding of touch by a linear decoder of pooled activity. We randomly selected a variable number of neurons in two separate sets of neurons corresponding to L4 barrels inside C2 (red) and outside of C2 (gray). We pooled the activity of all the neurons in each of the sets and integrated the neural activity for 10 milliseconds. A one-dimensional linear decoder was trained to discriminate neural activity during touch and non-touch epochs. Sets of neurons inside C2 decode touch presence with high confidence. Shading indicates 95% bounds. (B) Decoding of time of touch using a naïve Bayes classifier. A decoder was trained to classify neural responses occurring at different times from touch onset (the decoder assumes that the touch onset is known). Mean prediction (red line), 95% bounds (light red), true time (black dash). Inset, precision of time of touch decoding as a function of population size and time from touch. (C) Decoding of whisking phase using a naïve Bayes classifier. Median phase resolution from 100 decoding runs (dark lines), 95% bounds (light bands). Performance of d′ = 1 is equivalent to 0.76 of estimates falling within the resolution width. Decoding performance of whisking phase is poor even with N = 1000 neurons. Inside C2, red; outside C2, grey.

DOI: http://dx.doi.org/10.7554/eLife.06619.012

Figure 4.

Figure 4—figure supplement 1. Temporal decoding error.

Figure 4—figure supplement 1.

Heatmap of touch time decoding RMS error with respect to both population size and temporal position of analysis window.

In contrast, a large population of neurons (1000) was required to provide even a rough estimate of whisking phase (24° in C2, 33° outside at d′ = 1 performance, equivalent to 76% correct discrimination) in C2 and the surround columns (Figure 4C). The poor performance of the decoder is related to the low spike rate during whisker movements. Thus, the timing of spikes in L4 barrel cortex provides only a coarse representation of whisking phase.

Low spike count variability

A hallmark of cortical spike trains is high variability in spike count over repeated presentations of identical stimuli (Renart and Machens, 2014). Variability in the number of spikes evoked reflects both variability in behavior (e.g., the number and quality of touches) and irreducible noise intrinsic to the cortical circuit. We thus analyzed variability aligned to individual touches while accounting for differences in the properties of individual touches.

In our active object localization task, mice produce tactile input through whisker movements, which varies greatly across trials and individual touches within a trial (O'Connor et al., 2010). To explore how sensory responses depended on different tactile stimulus features we sorted touches and the L4 neuron responses by one of three behavioral variables: the order of touch within each trial, to account for adaptation; whisker velocity just before touch, to account for rate of change of impact forces; maximum whisker curvature during touch, which is proportional to peak touch force (Figure 5A,B). Touch response magnitude was highly modulated by each of the three variables (mean modulation index: touch order, 0.71 ± 0.25; velocity, 0.71 ± 0.27; max curvature, 0.72 ± 0.23; mean ± std; Figure 5A–C). Responsiveness to pretouch velocity and max curvature covaried strongly (pairwise correlation coefficient 0.80, p = 4.3e-8), whereas touch order response was somewhat less correlated with velocity and curvature (correlation coefficients 0.70, 0.65, p = 1.0e-5, p = 8.2e-5). The deep modulation index of sensory responses to tactile stimulus features indicates that stimulus variability likely accounts for a significant component of variability in the touch response.

Figure 5. L4 spike count varies with touch properties.

Figure 5.

(A) Left, touch aligned spike rasters for a single cell, sorted by one of three touch properties: Order of touch in trial (top), whisker velocity at touch onset (middle), maximum whisker curvature during touch (bottom). Value of touch property corresponding to the touch (red line). Spike integration window for binned touch response (pink), first touches in trials highlighted (grey). Right, average spikes per touch for a binned range of touch property (10 bins with equal number of touches) (black line), 95% confidence interval (grey line). Same example cell as in Figures 1, 3. (B) Heatmap of the response of each L4 excitatory cell inside C2 (n = 31) to the three touch characteristics across 10 equal element bins. Responses normalized to peak for each cell. Cells are ordered by the mean tuning to maximum touch curvature. Example cell highlighted by black arrow. (C) Heatmap of the modulation index of the same cells and touch characteristics (max bin − min bin)/(max bin + min bin).

DOI: http://dx.doi.org/10.7554/eLife.06619.014

To put an upper bound on how much spike count variation in our recordings derives from irreducible noise intrinsic to the cortical circuit, it is critical to sort touch events by stimulus characteristics and compute the FF across touches with similar features. A density-based clustering algorithm (Ester et al., 1996; Ankerst et al., 1999) was used to search for sets of touches with similar characteristics (see ‘Materials and methods’). To minimize effects of adaptation (Wang et al., 2010) we considered only touches that occurred after long inter-touch intervals (>250 ms). We binned the remaining touch events into five groups with similar touch strength and velocity at touch.

Since the vast majority of spikes occur in a narrow time window after touch, we computed FF in a sliding window around touch (10 ms; Figure 6A,B). The smallest possible FF is not zero in general because spike counts are whole numbers, whereas mean spike rates are continuous. For example, for a mean spike count equal to 0.5/touch, the minimum FF is produced with one spike in half of the trials (DeWeese et al., 2003). For mean spike count <1 the minimum FF corresponds to binomial spiking, with FF = 1- mean spike count (Berry et al., 1997). Across the population of neurons within C2 the FF following touch dropped below one, with the FF lying close to the binomial limit for most cells (Figure 6C,D).

Figure 6. L4 responses show minimal spike count variance.

(A) Raster plot of an example neuron aligned to touch onset. Example integration window (dashed lines) in which the neuron elicits 0 or 1 spike (black dots) per touch. (B) Fano factor computed over a sliding window of 10 ms (red; same neuron as A). Fano factor is ∼1 before touch occurs because the mean spike count is very low (∼0.03), which implies a minimum possible Fano factor of ∼1. Error bars, bootstrap s.e.m. Theoretical minimum Fano factor (green). PSTH aligned to touch, 1 ms bins (black). (C) Fano factor as a function of mean spike count for all L4 neurons in C2. For each cell we calculated the Fano factor in sliding windows of 10 ms and for each of the five similar touch groups. Fano factor for each 10 ms sliding window starting from touch onset up to 20 ms post touch in 1 ms increments (grey dots). The minimum Fano factor between 0–20 ms post touch (five touch clusters per cell; red circles). The minimum Fano factor across the five groups (blue circles). Theoretical minimum Fano factor (green line). (D) Fano factor averaged across the population of L4 neurons (black). Fano factor expected for Poisson neurons with equivalent spike rates (dashed grey) and with a 2.3 ms refractory period (dark grey). Error bars represent s.e.m. (E) Comparison of Fano factors. ‘1’, counting spikes during the sample period when the pole is within reach. ‘2’, counting spikes in random windows of 38 ms duration. The number of epochs per trial was matched to the number of touches in each trial. ‘3’ counting spikes in 38 ms windows after touch plus a latency of 6 ms. ‘4’, minimum FF using a sliding window of 10 ms after touch (between 0–20 ms after touch). ‘5’, same as ‘4’, except touches were divided into five groups (as in panels AD). Bars are s.e.m. (F) Fano factor as a function of mean spike count for the five conditions shown in panel E. Bars are s.e.m.

DOI: http://dx.doi.org/10.7554/eLife.06619.015

Figure 6.

Figure 6—figure supplement 1. Grouping of touch events using density-based clustering (OPTICS algorithm; see ‘Materials and methods’) for an example neuron.

Figure 6—figure supplement 1.

(A) Separation into five groups of touches by successively removing touches with similar attributes. In the first panel, blue dots represents the 20% of touches with similar velocity at touch and maximum curvature change (maximum between 0–20 ms after touch). Black dots represent all the other touches. The second group of touches (red) is obtained by repeating the clustering with the blue dots removed. This process is repeated until obtaining the five touch groups. (first group = blue; second group = red; third group = green; fourth group = cyan; fifth group = magenta). (B) Standard deviation in each of the five groups compared to random sampling. The procedure finds subset of points with significantly less behavioral variability than in the full dataset. The black curve represents the standard deviation for velocity at touch (z-scored), the curve for maximum curvature is similar (not shown).

Why is the FF close to 1 before and after touch (Figure 6B)? A brief (10 ms) sliding window was used to compute the FF, ensuring that mostly touch-related spikes were counted after touch. Since the spike rate is very low outside of the touch window, the mean spike count is also very low (∼0.03), implying FF of ∼1 for both the binomial and Poisson models. In contrast, around the peak of the touch response the mean spike count is high (∼0.7), allowing us to detect spiking statistics that differ from the Poisson distribution.

One possible explanation for the low FF after touch is a Poisson process with a refractory period (Berry II and Meister, 1998). A Poisson spike train with deletion of spikes that occur during the refractory period exhibits a FF less than one. We calculated FFs for simulated spike trains with Poisson rates matched to each recorded neuron and a median refractory period of 2.3 ms (Figure 2—figure supplement 1). Comparing the simulations with actual spike trains revealed that L4 neuron response precision could not be explained by the refractory period alone (Figure 6D).

We calculated the FF with different degrees of alignment to fine-scale behavior (Figure 6E,F). Response variability was very large (FF = 6.57 mean ± 5.41 s.d.) when we ignored the millisecond time scale of behavior (Figures 6E and 1). Integrating activity in time windows as wide as average touch response periods (38 ms) gave a FF of 1.59 ± 0.49 (Berry II and Meister, 1998) (Figures 6E, 2) when randomly sampling from the trial, and FF of 1.37 ± 0.74 (Figures 6E, 3) when aligned to touch onset. Using a narrower touch window of 10 ms reduced the FF to 0.73 ± 0.19 (Figures 6E, 4). The FF was further reduced to 0.54 ± 0.27 when calculated only across similar touches (Figures 6E, 5). Together, these results show that excitatory neurons in L4 of barrel cortex exhibit almost noiseless responses to touch.

Discussion

We measured the encoding of information by L4 neurons in the somatosensory cortex during active tactile sensation. Spike rates were low except for several milliseconds after touch onset (Figures 1, 2). During object localization, the majority (>70%) of spikes were temporally coupled to touch onset (Figure 3). Whisker movements organized the remaining spikes so that they aligned with particular phases of the whisk cycle. The time-scale of temporal modulation (approximately 10 ms) was much shorter than the mean inter-spike interval (approximately 1 s). Based on observations of whisker behavior it is possible to predict brief time windows when a neuron will fire a single or small number of spikes, as well as time periods when the spike probability is zero. Touch times could be reliably and precisely decoded by pooling activity from a handful of L4 neurons (Figure 4) (Panzeri et al., 2014). Spike count variance after touch (Figure 5), measured using the FF, was close to the binomial limit, the theoretical minimum (Figure 6). Based on these criteria we conclude that L4 responses encode touch with millisecond timescale precision and minimal noise.

This picture of low noise, rapidly modulated responses differs from the conclusions based on recordings from the cortex of behaving non-human primates (Tolhurst et al., 1983; Shadlen and Newsome, 1998; Maimon and Assad, 2009). Even in cases with time-varying stimuli and corresponding cortical responses with rapid modulation, spike counts vary greatly across trials, resulting in large FFs (Bair and Koch, 1996; Buracas et al., 1998). Four experimental factors might contribute to the large spike rate modulation and small FFs observed in our experiments: First, mice solved a discrimination task using active sensation (Knutsen et al., 2006; O'Connor et al., 2010; Kleinfeld and Deschenes, 2011; O'Connor et al., 2013). In contrast to passive presentation of stimuli (Tolhurst et al., 1983; Shadlen and Newsome, 1998; Maimon and Assad, 2009), in our task mice control sensory input by palpating the object with their whiskers. It is likely that mice tune their movements to achieve high signal-to-noise ratio encoding of tactile information. Second, by employing loose-seal cell-attached recordings we targeted neurons independent of activity, permitting accurate sampling of the spike trains produced by the L4 neuron population (DeWeese et al., 2003; O'Connor et al., 2010). These recording methods could be critical because standard extracellular recordings can have problems detecting neurons with low spike rate and high synchrony (Lewicki, 1998; DeWeese et al., 2003). During highly synchronous events, such as the population volley after touch, the probability is high that simultaneously recorded units in an extracellular electrode will be missed or misassigned (Cotton et al., 2013). Close to the binomial limit relatively few misassigned spikes can have a large impact on calculations of the FF (DeWeese et al., 2003). Third, compared to measurements from other cortical layers and regions, L4 spike trains in barrel cortex are more easily interpreted because extrinsic input arises mainly from the sensory periphery, rather than other cortical layers or higher cortical areas that might provide input with unobserved dynamics (Figure 1A).

Fourth, we track sensory input and whisker movements with temporal resolutions that are high compared to the inter-spike intervals of L4 cells. This is necessary to uncover possible influences of behavior on individual spikes. Indeed, L4 spikes appear irregular when aligned to the sample epoch (coefficient of variation >> 1, Figure 1C,D; Figure 6E,F). Only after aligning spikes to the fine-scale structure of the sensory input does the meaning of individual spikes become clear (Berry et al., 1997; Baudot et al., 2013).

In contrast to L4, only a small fraction of spikes in L5 neurons in the barrel cortex can be interpreted in terms of somatosensory behavior (Figure 3). This is likely because these neurons receive multiple types of input with unobserved dynamics representing hidden states (Petreanu et al., 2009). The irregularity of the L5 spike trains, as well as spike trains in other cell types and brain areas, may reveal themselves as deterministic fine-scale structure once the multitude of their inputs can be simultaneously monitored (Gomez et al., 2013).

The irregularity of cortical spike trains has often been interpreted as an irreducible feature of cortical discharges, or noise (Renart and Machens, 2014). This in turn has led to the view that only spike rates averaged across neuronal populations, but not precisely timed spikes, can be used to perform computations in cortical circuits (Mazurek and Shadlen, 2002; London et al., 2010). However, because a large number of inputs converge on single cortical neurons, noisy discharges are difficult to achieve in most models of neural networks (Softky and Koch, 1993). One exception is balanced networks, which have gained prominence in part because they produce irregular discharges as a result of chaotic dynamics (van Vreeswijk and Sompolinsky, 1996; Litwin-Kumar and Doiron, 2012). L4 neurons have variability close to the theoretical minimum. This implies that models of cortical networks should not explicitly aim to produce intrinsically noisy activity.

Other measurements in the barrel cortex have found highly variable responses to passive whisker deflection in L4 of anesthetized rats (Wang et al., 2010; Bale et al., 2013). Several key differences between this study and ours could underlie this discrepancy. During anesthesia cortical activity is modulated by slow rhythms (e.g., up and down states), which are not observed during active behavior (Crochet and Petersen, 2006). These rhythms are expected to increase spike count variability. In addition, the movements underlying active sensation might recruit circuit mechanisms that reduce variability. In our experiments mice ‘choose’ temporally sharp touches, which are expected to drive strong and rapid feedforward inhibition within L4 (Gabernet et al., 2005). The inhibition shortens L4 responses after touch and thus might produce effectively binary (0 or 1 spike) responses and low variability (DeWeese et al., 2003). Attentional mechanisms are also known to reduce neuronal variability (Mitchell et al., 2007, 2009) and may contribute to the low FFs we report here.

Neural coding of the timing of touch onset and whisking phase are key components of models of object localization (Curtis and Kleinfeld, 2009; Kleinfeld and Deschenes, 2011). We show that touch onset is reliably decodable from a small number of L4 neurons. In contrast, although many L4 cells are modulated by whisking phase, their low spike rates during whisker movements (mean <0.15 spikes/whisk cycle) hinders efficient decoding of whisking phase from population activity (Figure 4). The poor encoding of whisking phase is consistent with the observation that mice do not use the timing of L4 activity relative to whisking phase to measure object location, at least in the context of the simple task used in our experiments (O'Connor et al., 2013). How the brain uses temporally precise and low variance coding of touch in L4 neurons for tactile sensation remains to be discovered.

Materials and methods

Animals

All procedures were in accordance with protocols approved by the Janelia Farm Research Campus Institutional Animal Care and Use Committee. We report recordings from a total of 21 mice. 52 loose-seal cell-attached recordings were made in the following mice (7 recordings were reported in [O'Connor et al., 2013]): 21 recorded neurons from 7 C57BL/6 mice, 6 recorded neurons from 3 VGAT-ChR2(H134R) mice (i.e., Slc32a1-COP4*H134R/EYFP) (Zhao et al., 2011), 17 recorded neurons from 8 PV-ires-cre mice (i.e., Pvalbtm1(cre)Arbr) (Hippenmeyer et al., 2005), 8 recorded neurons from 3 PV-ires-cre X Ai32 mice (i.e., Gt(ROSA)26Sortm32.1(CAG-COP4*H134R/EYFP)Hze) (Madisen et al., 2012).

Behavior and videography

A detailed description of the behavioral apparatus, headplate installation, water-restriction schedule and training paradigm has been described (O'Connor et al., 2010; Guo et al., 2014). Mice were trained on a whisker-based go/nogo object localization task (O'Connor et al., 2010; O'Connor et al., 2013; Guo et al., 2014). A 0.5 mm diameter pole (class ZZ gage pin, Vermont Gage) was presented in one of two locations, 4–8mm apart on the anteroposterior axis. Mice licked the spout of an optical or electrical lickport to receive water reward if the pole was located in the posterior position, and withheld licking in the anterior position. No airpuff punishment or active removal of residual water from the lickport was used. Mice were trimmed to a single C2 whisker to perform the discrimination.

For each behavioral trial whisker video was recorded for 4–5 s, spanning the period prior to pole movement and following the response window (Clack et al., 2012; Pammer et al., 2013). Video frames were acquired in Streampix 3 software (Norpix, Canada) at 1000fps with 90–200 µs exposure times (Edmunds Optics #58-257). Videography was with a 0.36× telecentric lens and a Basler 504k camera under 940 nm LED illumination (Roithner Laser). Whisker trajectories and shapes were automatically quantified using the Janelia Whisker Tracker (https://openwiki.janelia.org/wiki/display/MyersLab/Whisker+Tracking; [Clack et al., 2012]). Contact periods with the pole were automatically determined by pole proximity and whisker curvature using custom Matlab routines (https://github.com/hireslab/HLab_Whiskers) and manually curated to ensure 1 millisecond accuracy.

Whisker analysis

The behavioral time-series were separated into touch, whisking, and non-whisking epochs. Touch epochs were periods where the whisker was in contact with the pole. Whisker curvature was measured, with K = 1/radius of the osculating circle tangent to the point of curvature measurement (Pammer et al., 2013). For other epochs, time-series of the azimuthal angle (theta) were bandpassed between 6–60 Hz (Butterworth fourth order) followed by decomposition by the Hilbert transform (Hill et al., 2011). Whisking epochs corresponded to periods where the zero-crossing phase of the Hilbert transform had amplitudes >2.5°, during which individual whisking cycles spanning–π to π were extracted. Whisking cycles during licking or within 70 ms after touch were excluded. Whisking epochs used for phase analysis had monotonic phase for a complete whisking cycle. Non-whisking periods were defined as contiguous periods of at least 100 ms with no touch or licking and with whisking amplitude <1.25°. Reaction time is the time between the first touch onset and the first lick calculated for every trial where licks occur (Figure 1—figure supplement 1).

Electrophysiology

On the day of the first recording, a small craniotomy (∼200 µm diameter) was made over the C2 barrel column determined by transcranial intrinsic signal imaging (O'Connor et al., 2010). The dura was left intact. Recordings targeting cortical L4 were obtained with patch pipettes pulled from borosilicate tubing (Sutter instrument, CA) and an Axopatch 700B amplifier (Molecular Devices). Loose-seal juxtacellular pipettes were filled with ACSF or cortex buffer (in mM): 125 NaCl, 5 KCl, 10 dextrose, 10 HEPES, 2 CaCl2, 2 MgSO4, pH 7.4, osmolality ∼272 mmol/kg. The manipulator depth was zeroed upon pipette tip contact with the dura (O'Connor et al., 2010). After contact, the craniotomy was covered by cortex buffer or 2% agar in cortex buffer. Aided by positive pressure (1 PSI), the pipette was advanced through the dura. When searching for cells, the pipette pressure was reduced to 0.1–0.3 PSI. Two pipette shapes were used, with a thicker shank (3.5–5 MΩ) (O'Connor et al., 2010) or a thinner shank (6–9 MΩ). Cells were recorded blindly (DeWeese et al., 2003). Recordings using thick shank pipettes caused cortical dimpling of ∼100 µm (O'Connor et al., 2010). Cells recorded with thick shanks and raw manipulator depths of 505–665 µm (405–565 µm corrected; 11 cells) were considered L4 and depths of 727–944 µm (627–844 µm corrected; 11 cells) were considered L5. Cells recorded with thin shank pipettes with raw manipulator depths of 444–560 µm (uncorrected; 30 cells) were considered L4 (Figure 3—figure supplements 1, 2). This depth range was consistently in L4 based on juxtacellular cell fills (unpublished observations). Data acquisition was controlled by Ephus (Suter et al., 2010). The sampling rate was 10 kHz.

Following the final recording, a DiI coated pipette was inserted into the craniotomy. Recordings in L4 with the midpoint of the DiI track <160 µm from the center of C2 were considered in C2 (31 recordings) and those >200 µm but still in the barrel field (10 recordings) were considered outside of C2 (Figure 1—figure supplement 2). L5 recordings ranged 98–324 µm from the center of C2 (near C2). Recordings with unstable spike rate across the behavioral session were excluded. Recordings were repeated for 1–5 days per animal.

Histology

Mice were deeply anesthetized with 5% isoflurane then perfused with 0.1 M sodium phosphate buffer followed by 4% paraformaldehyde (PFA, in 0.1 M phosphate buffer, pH 7.4). The brain was immersed in fixative for at least 24 hr before sectioning. The fixed cortex was flattened and fixed, and 100 µm slices were cut tangentially. Cytochrome oxidase (CO) staining was performed to reveal the barrel field (Land and Simons, 1985) and fluorescence imaging to determine the DiI pipette track relative to the field (Figure 1—figure supplement 2).

Spike analysis

For touch analyses, peri-contact time histograms of spikes (PCTHs), aligned to either first touch or later touch onset, were constructed (Figure 3—figure supplements 1, 2). The peak touch rate (Figure 2) was the highest mean spikes per bin (1 ms) of the cell's PCTHs, typically corresponding to the first touch. Spike onset latency is the time between the touch onset and where the rise of the PCTH exceeds the pretouch mean (−50 to 0 ms pre-onset) by two standard deviations (Figure 2—figure supplement 1). Spikes evoked per touch is the mean spike count within the touch-onset coupled spike window, defined below (Figure 5). For phase analysis, spikes were aligned to the whisking phase by linearly interpolating spike time to phase. For each neuron we built a histogram with 12 bins and fitted to a cosine function: A[1+cos(θφpref)]+B with A, B > 0 (Figure 3—figure supplements 1, 2). From the fit we extracted the preferred phase (φpref) and the modulation depth A/(A+B) (Figure 2—figure supplement 1).

Spikes accounted by active sensation

We estimated the proportion of spikes attributed to touch onset and whisking phase during active exploration (Figure 3). Active exploration epochs are whisking epochs, including touch epochs (excluding prolonged touches, > 100 ms). For touch onset, each spike was indexed by the time elapsed since the closest touch onset (taking into account the latency of each cell). Each neuron was characterized by a curve, yton(ti), representing the total number of spikes that occurred within a given time window (tiwithi=0,1etc). In cells where most spikes occurred shortly after touch onset these curves rise steeply and then plateau. For each neuron we determined the touch response cutoff, the first time point ti=ton in which the rate of increase Δyton(ti)=yton(ti)yton(ti1) is below chance level of the overall firing rate (p < 0.05; bootstrap method). The chance level was estimated by shuffling the spike times (1000 repetitions) to create surrogate curves ytonk(ti). The derivative was calculated from these curves smoothed with a polynomial fit (degree 11). For each ti we estimated the 95% percentile of the slope of the surrogate curves. For whisking phase, each spike that was not attributed to touch onset was indexed by the time elapsed between the spike and the closest whisk cycle preferred phase (taking into account phase circularity and including whisking epochs with non-monotonic phase). As with touch, we built ywhisk(ti) representing the number of spikes that occurred around the preferred whisking phase. We determined the whisking spike cutoff, the first time point ti=tw in which the rate of increase of explained spikes by whisking was below chance levels.

Population decoding

To decode touch based on spikes for each neuron ‘i’ (Figure 4) we extracted the spiking response ri,j(t) of all touch events ‘j’ from t = 30 ms before touch up to t = 50 ms after touch (with j=1Ntouch(i); excluding events with a second touch in the 50 ms post-touch window). We randomly selected a set of neurons (up to N = 200) and for each neuron we randomly sampled (with replacement) 1000 of its touch aligned responses ri,j(t). Each spike train was causally integrated with a defined window (i.e., w = 10 ms). At each time point tj we built a different decoder that was trained to discriminate between the pooled population response at tj and the population response at epochs without touch (i.e., a one-dimensional decoder). We used half of the data to find the optimal threshold and the other half of the data to predict the performance of the decoder. Other decoders, such as naïve Bayes classifier and Fisher linear discriminant, produced similar results (not shown). The decoders assume implicit knowledge of when the touch occurred, since at each time point a different decoder was employed.

To decode the time elapsed since touch onset we used the same sampling procedure as described above and integrated the response with a sliding window with duration of 10 ms. We trained a multinomial naïve Bayes classifier (Duda et al., 2001) to report for every time point what was the most likely time elapsed from touch onset, with a root mean square time error derived from boostrapping (100 runs) (Figure 4—figure supplement 1). To decode the whisking phase we assumed that the preferred phase was uniformly distributed. For each neuron we aligned the spikes to the whisking phase of each whisking cycle (minimum peak whisking amplitude >2.5°) and binned the response in 120 bins between π and π. To simulate the population response we randomly picked a set of neurons (with replacement) and circularly shifted their response to obtain a uniform distribution of preferred phases. We trained a naïve Bayes classifier to discriminate the population response to two different whisking phases. We tested the performance of the decoders in function of the difference of whisking phase and determined the phase difference that achieved 76% correct discrimination in function of the number of neurons.

Under the assumption of independent neural responses, our simulations show several-fold more information about touch in C2 than non-C2 and poor performance in decoding whisking phase (Figure 4). Other factors, such as correlated activity among neurons, can impact the accuracy of the population code (Dayan and Abbott, 2001; Moreno-Bote et al., 2014) and the exact information carried by touch and whisking phase, but this is unlikely to alter the qualitative picture presented here.

Analysis of spike count variance

To sort out external and intrinsic contributions to spike count variability, we took several steps to reduce external variability due to differences in sensorimotor variables. Touch responses in L4 neurons are modulated by adaptation, pretouch velocity and curvature of the whisker (which is proportional to touch force) (Figure 5). To reduce adaptation effects we selected touch epochs in which the inter contact interval (ICI) was longer than 250 ms. We divided the remaining touch events into five groups (N, number of points per group; minimum, 20), clustered by touch characteristics. We z-scored pretouch velocity and the maximum curvature change shortly after touch (0–20 ms). For each neuron we clustered the touch events using a density-based clustering method (OPTICS; [Cunningham and Yu, 2014]). The output of OPTICS gives an ordered list li of points sorted by similarity. We searched for the set of consecutive sorted points li,li+1,,li+N1 that mimimized the sum over all pairwise distances. After obtaining the set of touch events for the first bin, we removed those points and proceeded in the same manner to obtain the second data bin. We repeated the procedure until obtaining five data bins (Figure 6—figure supplement 1). We calculated the FF by counting spikes in sliding windows of 10 ms for each cell and each of the five bins (Figure 6). Confidence intervals for the FFs were obtained by resampling 1000 times. We also computed FFs using a Poisson process with absolute refractory period, with average touch response matched to the data (Berry II and Meister, 1998) (Figure 2—figure supplement 1).

Acknowledgements

We thank Mike DeWeese, Shaul Druckmann, David Golomb, Judith Hirsch, Máté Lengyel, Nuo Li, Simon Peron, Alex Pouget, Sandro Romani, Nick Sofroniew, and Fritz Sommer for comments on the manuscript.

Funding Statement

The funder had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Funding Information

This paper was supported by the following grant:

  • Howard Hughes Medical Institute (HHMI) to Samuel Andrew Hires, Diego A Gutnisky, Jianing Yu, Daniel H O'Connor, Karel Svoboda.

Additional information

Competing interests

The authors declare that no competing interests exist.

Author contributions

SAH, Conception and design, Acquisition of data, Development of Whisker analysis code, Drafting the article.

DAG, Conception and design, Development of Whisker analysis code, Drafting the article.

JY, Acquisition of data, Analysis and interpretation of data, Final approval of article.

DHO, Conception and design, Development of Whisker analysis code, Final approval of article.

KS, Conception and design, Development Whisker analysis code, Drafting the article.

Ethics

Animal experimentation: All procedures were in accordance with protocols approved by the Janelia Farm Research Campus Institutional Animal Care and Use Committee (#11-71).

References

  1. Adibi M, Clifford CW, Arabzadeh E. Informational basis of sensory adaptation: entropy and single-spike efficiency in rat barrel cortex. The Journal of Neuroscience. 2013;33:14921–14926. doi: 10.1523/JNEUROSCI.1313-13.2013. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Amarasingham A, Chen TL, Geman S, Harrison MT, Sheinberg DL. Spike count reliability and the Poisson hypothesis. The Journal of Neuroscience. 2006;26:801–809. doi: 10.1523/JNEUROSCI.2948-05.2006. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Ankerst M, Breunig MM, Kriegel HP, Sander J. OPTICS: Ordering points to identify the clustering structure. Sigmod Record. 1999;28:49–60. doi: 10.1145/304181.304187. [DOI] [Google Scholar]
  4. Bair W, Koch C. Temporal precision of spike trains in extrastriate cortex of the behaving macaque monkey. Neural Computation. 1996;8:1185–1202. doi: 10.1162/neco.1996.8.6.1185. [DOI] [PubMed] [Google Scholar]
  5. Bale MR, Davies K, Freeman OJ, Ince RA, Petersen RS. Low-dimensional sensory feature representation by trigeminal primary afferents. The Journal of Neuroscience. 2013;33:12003–12012. doi: 10.1523/JNEUROSCI.0925-13.2013. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Baudot P, Levy M, Marre O, Monier C, Pananceau M, Frégnac Y. Animation of natural scene by virtual eye-movements evokes high precision and low noise in V1 neurons. Frontiers in Neural Circuits. 2013;7:206. doi: 10.3389/fncir.2013.00206. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Berry MJ, II, Meister M. Refractoriness and neural precision. The Journal of Neuroscience. 1998;18:2200–2211. doi: 10.1523/JNEUROSCI.18-06-02200.1998. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Berry MJ, Warland DK, Meister M. The structure and precision of retinal spike trains. Proceedings of the National Academy of Sciences of USA. 1997;94:5411–5416. doi: 10.1073/pnas.94.10.5411. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Birdwell JA, Solomon JH, Thajchayapong M, Taylor MA, Cheely M, Towal RB, Conradt J, Hartmann MJ. Biomechanical models for radial distance determination by the rat vibrissal system. Journal of Neurophysiology. 2007;98:2439–2455. doi: 10.1152/jn.00707.2006. [DOI] [PubMed] [Google Scholar]
  10. Bruno RM, Sakmann B. Cortex is driven by weak but synchronously active thalamocortical synapses. Science. 2006;312:1622–1627. doi: 10.1126/science.1124593. [DOI] [PubMed] [Google Scholar]
  11. Buracas GT, Zador AM, DeWeese MR, Albright TD. Efficient discrimination of temporal patterns by motion-sensitive neurons in primate visual cortex. Neuron. 1998;20:959–969. doi: 10.1016/S0896-6273(00)80477-8. [DOI] [PubMed] [Google Scholar]
  12. Bureau I, von Saint Paul F, Svoboda K. Interdigitated Paralemniscal and Lemniscal Pathways in the mouse barrel cortex. PLOS Biology. 2006;4:e382. doi: 10.1371/journal.pbio.0040382. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Churchland MM, Yu BM, Cunningham JP, Sugrue LP, Cohen MR, Corrado GS, Newsome WT, Clark AM, Hosseini P, Scott BB, Bradley DC, Smith MA, Kohn A, Movshon JA, Armstrong KM, Moore T, Chang SW, Snyder LH, Lisberger SG, Priebe NJ, Finn IM, Ferster D, Ryu SI, Santhanam G, Sahani M, Shenoy KV. Stimulus onset quenches neural variability: a widespread cortical phenomenon. Nature Neuroscience. 2010;13:369–378. doi: 10.1038/nn.2501. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Clack NG, O'Connor DH, Huber D, Petreanu L, Hires A, Peron S, Svoboda K, Myers EW. Automated tracking of whiskers in videos of head fixed rodents. PLOS Computational Biology. 2012;8:e1002591. doi: 10.1371/journal.pcbi.1002591. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Cotton RJ, Froudarakis E, Storer P, Saggau P, Tolias AS. Three-dimensional mapping of microcircuit correlation structure. Frontiers in Neural Circuits. 2013;7:151. doi: 10.3389/fncir.2013.00151. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Crochet S, Petersen CC. Correlating whisker behavior with membrane potential in barrel cortex of awake mice. Nature Neuroscience. 2006;9:608–610. doi: 10.1038/nn1690. [DOI] [PubMed] [Google Scholar]
  17. Cunningham JP, Yu BM. Dimensionality reduction for large-scale neural recordings. Nature Neuroscience. 2014;17:1500–1509. doi: 10.1038/nn.3776. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Curtis JC, Kleinfeld D. Phase-to-rate transformations encode touch in cortical neurons of a scanning sensorimotor system. Nature Neuroscience. 2009;12:492–501. doi: 10.1038/nn.2283. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Dayan P, Abbott LF. Theoretical neuroscience: computational and mathematical modeling of neural systems. MIT Press; 2001. [Google Scholar]
  20. Deschenes M, Moore J, Kleinfeld D. Sniffing and whisking in rodents. Current Opinion in Neurobiology. 2012;22:243–250. doi: 10.1016/j.conb.2011.11.013. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. DeWeese MR, Wehr M, Zador AM. Binary spiking in auditory cortex. The Journal of Neuroscience. 2003;23:7940–7949. doi: 10.1523/JNEUROSCI.23-21-07940.2003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Diamond ME, von Heimendahl M, Knutsen PM, Kleinfeld D, Ahissar E. 'Where' and 'what' in the whisker sensorimotor system. Nature Reviews. Neuroscience. 2008;9:601–612. doi: 10.1038/nrn2411. [DOI] [PubMed] [Google Scholar]
  23. Duda RO, Hart PE, Stork DG. Pattern classification, Edn. 2nd. Wiley, New York: 2001. [Google Scholar]
  24. Ester M, Kriegel HP, Sander J, Xu X. Association for the Advancement of Artificial Intelligence. 1996. A density-based algorithm for discovering clusters in large spatial databases with noise. in Proceedings of the Second International Conference on Knowledge Discovery and Data Mining (KDD’96) pp. 226–231. [Google Scholar]
  25. Fee MS, Mitra PP, Kleinfeld D. Central versus peripheral determinants of patterned spike activity in rat vibrissa cortex during whisking. Journal of Neurophysiology. 1997;78:1144–1149. doi: 10.1152/jn.1997.78.2.1144. [DOI] [PubMed] [Google Scholar]
  26. Gabernet L, Jadhav SP, Feldman DE, Carandini M, Scanziani M. Somatosensory integration controlled by dynamic thalamocortical feed-forward inhibition. Neuron. 2005;48:315–327. doi: 10.1016/j.neuron.2005.09.022. [DOI] [PubMed] [Google Scholar]
  27. Gollisch T, Meister M. Rapid neural coding in the retina with relative spike latencies. Science. 2008;319:1108–1111. doi: 10.1126/science.1149639. [DOI] [PubMed] [Google Scholar]
  28. Gomez C, Smolyanskaya C, Kreiman G, Born RT. Silencing V2/V3 reduces spiking variability in MT: implications for excitatory/inhibitory balance CoSyNe Meeting. 2013. [Google Scholar]
  29. Guo ZV, Hires SA, Li N, O'Connor DH, Komiyama T, Ophir E, Huber D, Bonardi C, Morandell K, Gutnisky D, Peron S, Xu NL, Cox J, Svoboda K. Procedures for behavioral experiments in head-fixed mice. PLOS ONE. 2014;9:e88678. doi: 10.1371/journal.pone.0088678. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Guo ZV, Li N, Huber D, Ophir E, Gutnisky D, Ting JT, Feng G, Svoboda K. Flow of cortical activity underlying a tactile decision in mice. Neuron. 2014;81:179–194. doi: 10.1016/j.neuron.2013.10.020. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Gur M, Beylin A, Snodderly DM. Response variability of neurons in primary visual cortex (V1) of alert monkeys. The Journal of Neuroscience. 1997;17:2914–2920. doi: 10.1523/JNEUROSCI.17-08-02914.1997. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Hill DN, Curtis JC, Moore JD, Kleinfeld D. Primary motor cortex reports efferent control of vibrissa motion on multiple timescales. Neuron. 2011;72:344–356. doi: 10.1016/j.neuron.2011.09.020. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Hippenmeyer S, Vrieseling E, Sigrist M, Portmann T, Laengle C, Ladle DR, Arber S. A developmental switch in the response of DRG neurons to ETS transcription factor signaling. PLOS Biology. 2005;3:e159. doi: 10.1371/journal.pbio.0030159. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Hooks BM, Hires SA, Zhang YX, Huber D, Petreanu L, Svoboda K, Shepherd GM. Laminar analysis of excitatory local circuits in vibrissal motor and sensory cortical areas. PLOS Biology. 2011;9:e1000572. doi: 10.1371/journal.pbio.1000572. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Hutson KA, Masterton RB. The sensory contribution of a single vibrissa's cortical barrel. Journal of Neurophysiology. 1986;56:1196–1223. doi: 10.1152/jn.1986.56.4.1196. [DOI] [PubMed] [Google Scholar]
  36. Jadhav SP, Wolfe J, Feldman DE. Sparse temporal coding of elementary tactile features during active whisker sensation. Nature Neuroscience. 2009;12:792–800. doi: 10.1038/nn.2328. [DOI] [PubMed] [Google Scholar]
  37. Johansson RS, Birznieks I. First spikes in ensembles of human tactile afferents code complex spatial fingertip events. Nature Neuroscience. 2004;7:170–177. doi: 10.1038/nn1177. [DOI] [PubMed] [Google Scholar]
  38. Kara P, Reinagel P, Reid RC. Low response variability in simultaneously recorded retinal, thalamic, and cortical neurons. Neuron. 2000;27:635–646. doi: 10.1016/S0896-6273(00)00072-6. [DOI] [PubMed] [Google Scholar]
  39. Keat J, Reinagel P, Reid RC, Meister M. Predicting every spike: a model for the responses of visual neurons. Neuron. 2001;30:803–817. doi: 10.1016/S0896-6273(01)00322-1. [DOI] [PubMed] [Google Scholar]
  40. Kleinfeld D, Deschenes M. Neuronal basis for object location in the vibrissa scanning sensorimotor system. Neuron. 2011;72:455–468. doi: 10.1016/j.neuron.2011.10.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Knutsen PM, Pietr M, Ahissar E. Haptic object localization in the vibrissal system: behavior and performance. The Journal of Neuroscience. 2006;26:8451–8464. doi: 10.1523/JNEUROSCI.1516-06.2006. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Land PW, Simons DJ. Cytochrome oxidase staining in the rat SmI barrel cortex. The Journal of Comparative Neurology. 1985;238:225–235. doi: 10.1002/cne.902380209. [DOI] [PubMed] [Google Scholar]
  43. Lefort S, Tomm C, Floyd Sarria JC, Petersen CC. The excitatory neuronal network of the C2 barrel column in mouse primary somatosensory cortex. Neuron. 2009;61:301–316. doi: 10.1016/j.neuron.2008.12.020. [DOI] [PubMed] [Google Scholar]
  44. Lewicki MS. A review of methods for spike sorting: the detection and classification of neural action potentials. Network. 1998;9:R53–R78. doi: 10.1088/0954-898X/9/4/001. [DOI] [PubMed] [Google Scholar]
  45. Litwin-Kumar A, Doiron B. Slow dynamics and high variability in balanced cortical networks with clustered connections. Nature Neuroscience. 2012;15:1498–1505. doi: 10.1038/nn.3220. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. London M, Roth A, Beeren L, Hausser M, Latham PE. Sensitivity to perturbations in vivo implies high noise and suggests rate coding in cortex. Nature. 2010;466:123–127. doi: 10.1038/nature09086. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Lu SM, Lin RCS. Thalamic afferents of the rat barrel cortex: a light- and electron-microscopic study using Phaseolus vulgaris leucoagglutinin as an anterograde tracer. Somatosensory & Motor Research. 1993;10:1–16. doi: 10.3109/08990229309028819. [DOI] [PubMed] [Google Scholar]
  48. Madisen L, Mao T, Koch H, Zhuo JM, Berenyi A, Fujisawa S, Hsu YW, Garcia AJ, III, Gu X, Zanella S, Kidney J, Gu H, Mao Y, Hooks BM, Boyden ES, Buzsáki G, Ramirez JM, Jones AR, Svoboda K, Han X, Turner EE, Zeng H. A toolbox of Cre-dependent optogenetic transgenic mice for light-induced activation and silencing. Nature Neuroscience. 2012;15:793–802. doi: 10.1038/nn.3078. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Maimon G, Assad JA. Beyond Poisson: increased spike-time regularity across primate parietal cortex. Neuron. 2009;62:426–440. doi: 10.1016/j.neuron.2009.03.021. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Mainen ZF, Sejnowski TJ. Reliability of spike timing in neocortical neurons. Science. 1995;268:1503–1506. doi: 10.1126/science.7770778. [DOI] [PubMed] [Google Scholar]
  51. Masquelier T. Neural variability, or lack thereof. Frontiers in Computational Neuroscience. 2013;7:7. doi: 10.3389/fncom.2013.00007. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Mazurek ME, Shadlen MN. Limits to the temporal fidelity of cortical spike rate signals. Nature Neuroscience. 2002;5:463–471. doi: 10.1038/nn836. [DOI] [PubMed] [Google Scholar]
  53. Mitchell JF, Sundberg KA, Reynolds JH. Differential attention-dependent response modulation across cell classes in macaque visual area V4. Neuron. 2007;55:131–141. doi: 10.1016/j.neuron.2007.06.018. [DOI] [PubMed] [Google Scholar]
  54. Mitchell JF, Sundberg KA, Reynolds JH. Spatial attention decorrelates intrinsic activity fluctuations in macaque area V4. Neuron. 2009;63:879–888. doi: 10.1016/j.neuron.2009.09.013. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Moreno-Bote R, Beck J, Kanitscheider I, Pitkow X, Latham P, Pouget A. Information-limiting correlations. Nature Neuroscience. 2014;17:1410–1417. doi: 10.1038/nn.3807. [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. O'Connor DH, Clack NG, Huber D, Komiyama T, Myers EW, Svoboda K. Vibrissa-based object localization in head-fixed mice. The Journal of Neuroscience. 2010a;30:1947–1967. doi: 10.1523/JNEUROSCI.3762-09.2010. [DOI] [PMC free article] [PubMed] [Google Scholar]
  57. O'Connor DH, Hires SA, Guo ZV, Li N, Yu J, Sun QQ, Huber D, Svoboda K. Neural coding during active somatosensation revealed using illusory touch. Nature Neuroscience. 2013;16:958–965. doi: 10.1038/nn.3419. [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. O'Connor DH, Peron SP, Huber D, Svoboda K. Neural activity in barrel cortex underlying vibrissa-based object localization in mice. Neuron. 2010b;67:1048–1061. doi: 10.1016/j.neuron.2010.08.026. [DOI] [PubMed] [Google Scholar]
  59. Pammer L, O'Connor DH, Hires SA, Clack NG, Huber D, Myers EW, Svoboda K. The mechanical variables underlying object localization along the axis of the whisker. The Journal of Neuroscience. 2013;33:6726–6741. doi: 10.1523/JNEUROSCI.4316-12.2013. [DOI] [PMC free article] [PubMed] [Google Scholar]
  60. Panzeri S, Ince RA, Diamond ME, Kayser C. Reading spike timing without a clock: intrinsic decoding of spike trains. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences. 2014;369:20120467. doi: 10.1098/rstb.2012.0467. [DOI] [PMC free article] [PubMed] [Google Scholar]
  61. Petreanu L, Mao T, Sternson SM, Svoboda K. The subcellular organization of neocortical excitatory connections. Nature. 2009;457:1142–1145. doi: 10.1038/nature07709. [DOI] [PMC free article] [PubMed] [Google Scholar]
  62. Renart A, Machens CK. Variability in neural activity and behavior. Current Opinion in Neurobiology. 2014;25:211–220. doi: 10.1016/j.conb.2014.02.013. [DOI] [PubMed] [Google Scholar]
  63. Shadlen MN, Newsome WT. The variable discharge of cortical neurons: implications for connectivity, computation, and information coding. The Journal of Neuroscience. 1998;18:3870–3896. doi: 10.1523/JNEUROSCI.18-10-03870.1998. [DOI] [PMC free article] [PubMed] [Google Scholar]
  64. Simons DJ. Response properties of vibrissa units in rat SI somatosensory neocortex. Journal of Neurophysiology. 1978;41:798–820. doi: 10.1152/jn.1978.41.3.798. [DOI] [PubMed] [Google Scholar]
  65. Softky WR, Koch C. The highly irregular firing of cortical cells is inconsistent with temporal integration of random EPSPs. The Journal of Neuroscience. 1993;13:334–350. doi: 10.1523/JNEUROSCI.13-01-00334.1993. [DOI] [PMC free article] [PubMed] [Google Scholar]
  66. Stevens CF, Zador AM. Input synchrony and the irregular firing of cortical neurons. Nature Neuroscience. 1998;1:210–217. doi: 10.1038/659. [DOI] [PubMed] [Google Scholar]
  67. Suter BA, O'Connor T, Iyer V, Petreanu LT, Hooks BM, Kiritani T, Svoboda K, Shepherd GM. Ephus: multipurpose data acquisition software for neuroscience experiments. Frontiers Neuroscience. 2010;4 doi: 10.3389/fncir.2010.00100. [DOI] [PMC free article] [PubMed] [Google Scholar]
  68. Tolhurst DJ, Movshon JA, Dean AF. The statistical reliability of signals in single neurons in cat and monkey visual cortex. Vision Research. 1983;23:775–785. doi: 10.1016/0042-6989(83)90200-6. [DOI] [PubMed] [Google Scholar]
  69. van Vreeswijk C, Sompolinsky H. Chaos in neuronal networks with balanced excitatory and inhibitory activity. Science. 1996;274:1724–1726. doi: 10.1126/science.274.5293.1724. [DOI] [PubMed] [Google Scholar]
  70. VanRullen R, Guyonneau R, Thorpe SJ. Spike times make sense. Trends in Neurosciences. 2005;28:1–4. doi: 10.1016/j.tins.2004.10.010. [DOI] [PubMed] [Google Scholar]
  71. Wang Q, Webber RM, Stanley GB. Thalamic synchrony and the adaptive gating of information flow to cortex. Nature Neuroscience. 2010;13:1534–1541. doi: 10.1038/nn.2670. [DOI] [PMC free article] [PubMed] [Google Scholar]
  72. Zhao S, Ting JT, Atallah HE, Qiu L, Tan J, Gloss B, Augustine GJ, Deisseroth K, Luo M, Graybiel AM, Feng G. Cell type-specific channelrhodopsin-2 transgenic mice for optogenetic dissection of neural circuitry function. Nature Methods. 2011;8:745–752. doi: 10.1038/nmeth.1668. [DOI] [PMC free article] [PubMed] [Google Scholar]
eLife. 2015 Aug 6;4:e06619. doi: 10.7554/eLife.06619.017

Decision letter

Editor: Matteo Carandini1

eLife posts the editorial decision letter and author response on a selection of the published articles (subject to the approval of the authors). An edited version of the letter sent to the authors after peer review is shown, indicating the substantive concerns or comments; minor concerns are not usually shown. Reviewers have the opportunity to discuss the decision before the letter is sent (see review process). Similarly, the author response typically shows only responses to the major concerns raised by the reviewers.

Thank you for sending your work entitled “Low-noise encoding of active touch by layer 4 in the somatosensory cortex” for consideration at eLife. Your article has been favorably evaluated by a Senior editor and three reviewers, one of whom is a member of our Board of Reviewing Editors.

The Reviewing editor and the other reviewers discussed their comments extensively before we reached this decision, and the Reviewing editor has assembled the following comments to help you prepare a revised submission.

1) This revised submission does not require new experiments, but it does require new analyses and ideally new data from other cortical layers, which the authors may already have in their database.

This is potentially a solid study that sheds some light on the factors that determine what makes cortical neurons fire. The paper reports on the variability of spike trains in layer 4 (L4) neurons in mouse somatosensory cortex during a touch task (pole localization) involving active whisking. It makes a convincing argument that variability is very low, once spikes are aligned properly based on time of touch and on whisk phase. Specifically, touch (more than whisking phase) seems to be the primary feature encoded by the timing of the spikes. Indeed, most spikes can be explained by whisker curvature and related kinematics, together with (to a smaller degree) whisking itself. Around the time of the touch the noise in these neurons is near the theoretical minimum. Thus what might appear as trial-to-trial variability or noise is not. Spiking appears to be noisy only if the observer does not or cannot take into account the true determinants of spiking.

2) This paper, however, doesn't really change our view of sensory processing in L4 of barrel cortex. It shows that L4 responses can be predicted rather closely if one knows a few fundamentals about the sensory and motor time courses. This is hardly surprising: it would have been surprising if one could have predicted spike trains without knowledge of those fundamentals. Still, it is interesting to see that most spikes can be accounted for, once one has those simple pieces of information.

3) This said, the study also suffers from major limitations. Principally, the main finding is not novel – it is made to appear novel only by adopting strawman hypotheses, ignoring much of the relevant and recent literature.

4) Indeed, it is widely agreed that individual cortical neurons are not noisy: if cortical responses are variable, the variability is due to integration of inputs, which is mostly seen in central neurons.

5) Moreover, as the paper points out, layer 4 in this system is close to the sensory periphery, with little input from the rest of cortex – and we already know that in the sensory periphery there is very little noise. These and other criticisms are laid out below.

6) Premise of the paper. The Introduction begins with “Cortical spike trains are often considered to be noisy”, and casts much of the paper in the context of this view. But after decades of work, this idea of cortical neurons being “noisy” has become a strawman. It is out of step with crucial pieces of the literature. We already know that cortical neurons are not intrinsically noisy; see e.g. Mainen and Sejnowski, “Reliability of spike timing in neocortical neurons”. Science 1995, and Deweese and Zador, “Shared and private variability in the auditory cortex”. J Neurophysiol 2000. Also, the apparent Poisson nature of variability can be explained by a deterministic spike threshold (Carandini, Plos Biol 2000). Similarly, the statement that “In cortex, irregular spike trains suggest that only rate codes can be used to perform reliable computations” is misleading, and based on a reference that is over 10 years old.

7) Vision literature. The paper uses the literature from visual cortex to make the point that cortical responses are noisy. But even in that literature there is much that contradicts this point. In a paper titled “Low response variability in simultaneously recorded retinal, thalamic, and cortical neurons” (Neuron, 2000), Kara, Reinagel, and Reid have already shown that responses in layer 4 of visual cortex can be remarkably repeatable. And these results are not limited to anesthetized preparations: similar results have been obtained in awake monkeys, where (as one could have imagined, and similarly to what is done here for whiskers) it is essential to know eye position on a moment by moment basis: Gur, Beylin, and Snodderly, “Response variability of neurons in primary visual cortex (V1) of alert monkeys” (J Neurosci, 1997).

8) Somatosensory literature. If we now turn to the recent literature on somatosensory cortex, we see that it makes precisely the opposite case relative to neurons being “noisy”. Indeed, the paper might just as well have begun with: “Cortical spike trains are often considered to be made up of precisely timed spikes, whereby the timing of each spike carries some meaning about the stimulus.” See for example Arabzadeh, Zorzin and Diamond “Neuronal encoding of texture in the whisker sensory pathway” (PLoS Biol 2005) and “Deciphering the spike train of a sensory neuron: counts and temporal patterns in the rat whisker pathway” (J Neurosci 2006). Additional contributions that should be considered in rethinking how to cast these results include those from G. Stanley (admittedly anesthetized, as noted in this study), E. Arabzadeh, G. Foffani, R. Petersen, and S. Panzeri. The latter include “Information carried by population spike times in the whisker sensory cortex can be decoded without knowledge of stimulus time” (Frontiers, 2010) – which is related to the authors' conclusion that “Pooling activity from only fifteen L4 neurons […] in C2 was sufficient to detect 95 % of touches” – and “Complementary Contributions of Spike Timing and Spike Rate to Perceptual Decisions in Rat S1 and S2 Cortex” (Current Biology, 2015). The latter is admittedly too recent to have shaped the present paper, but it is now in the literature.

9) Poisson spike generators. Given the literature, the Poisson model with a refractory period has become a strawman. Therefore, the simulation of Poisson firing with refractory period (in the sixth paragraph of the subsection “Low spike count variability”) is unsatisfying. We already know that the Poisson model is not a reasonable model, even with refractory periods. We know this in retina, in LGN, in V1, and in A1. It would be more interesting to test a more plausible model. For instance, would an integrate-and-fire model with Gaussian noise in Vm give the observed statistics?

10) Fano Factor. Related to the last point: is it sensible to calculate Fano Factor (FF) for such a radically non-stationary process? Fano Factor is at best awkward for non-stationary processes and the main message of the paper is that spiking is non-stationary, e.g. extremely variable over large time scales and extremely precise when whiskers kinematics dictate so. For a Poisson process, FF = 1, independent of the window size. Since the observed FF for full stimulus window was large, evidently a Poisson process is not underlying spikes. At best it is a Poisson process driven by a variable mean rate. Indeed, Poisson statistics are meant to model stationary processes, that is, processes where the probability of an event per unit time is constant. Obviously firing probability increases in the tens of ms after touch, compared to the rest of the trial.

11) Further, having demonstrated that the neurons of interest in barrel cortex do not resemble a Poisson process, the modeled neurons do: “We used a simplified model based on independent Poisson neurons with rapidly modulated spike rates”. It is puzzling to model Poisson neurons once one knows that the model is not correct.

12) Factors accounting for spike trains. In the present analysis (in the second paragraph of the subsection “Low spike count variability”), 3 factors are used to account, one at a time, for spike trains: order of touch, whisker velocity, and whisker curvature. Why use them one at a time? Why not use all 3 at once and see how much better one does? Or use only the two main ones? Are there any spikes unaccounted for once one does this? This could be done with a GLM (perhaps enlisting the help of colleague Jeremy Freeman) or with similar approaches. Indeed, since the main message of the paper is that spike trains can be explained by whisker curvature and related kinematics, it would make sense to place less emphasis on the Poisson analysis and more emphasis on showing whisker state as a determining factor.

13) Other layers. Rather than focusing solely on L4, it would have been much more interesting to compare L4 with other cortical layers, and investigate the sources of variability there. Indeed, a major missing component is that neurons outside L4 are not recorded or reported. This is set up in the Introduction, but not really addressed. It's great to show L4 does one thing, but more interesting to show that it does something different than other layers. Given other publications from this group, it seems likely that the authors have other data that addresses this. Analyzing these data would make the paper much stronger.

eLife. 2015 Aug 6;4:e06619. doi: 10.7554/eLife.06619.018

Author response


1) This revised submission does not require new experiments, but it does require new analyses and ideally new data from other cortical layers, which the authors may already have in their database.

The reviewers are presumably referring to the database from O’Connor et al. 2010. Unfortunately, in this previous study and all other previous studies from our lab and other labs, the behavioral tracking (videography) was not sufficiently accurate to extract touches with millisecond precision and low error rates.

However, we agree that the paper would be stronger by comparing our results with other neuron types that are known to receive long-range cortico-cortical inputs. One of the main points of our work is that virtually all L4 spikes can be explained based on behavior. The layer 4 circuit in one S1 barrel receives external inputs mainly from ∼200 neurons of the corresponding barreloid in VPM. Other neuron types, such as layer 5 pyramidal neurons, receive in addition inputs from diverse sources that we do not observe (motor cortex, S2 etc); this can cause apparently variable neural responses.

To enable a comparison between microcircuit types, we have extended our analysis to a new set of L5 neurons. In contrast to L4, only a small percentage of spikes in L5 can be explained by whisking and touch. The new L5 data has been included in Figure 2C, D and Figure 3C, D.

We have extended our L4 data set as well, adding 14 new recordings in L4 C2 bringing the total from 17 neurons to 31 neurons.

2) This paper, however, doesn't really change our view of sensory processing in L4 of barrel cortex. It shows that L4 responses can be predicted rather closely if one knows a few fundamentals about the sensory and motor time courses. This is hardly surprising: it would have been surprising if one could have predicted spike trains without knowledge of those fundamentals. Still, it is interesting to see that most spikes can be accounted for, once one has those simple pieces of information.

We respectfully disagree with this point. The experimental literature overwhelmingly reports highly variable neural responses (e.g. references in our manuscript; Carandini, 2004; Churchland et al. 2010; Goris et al. 2014). This has influenced theoretical modeling in such a way that variable (Poisson-like) spiking statistics are a measure of success for networks models of the brain (e.g. van Vreeswijk & Sompolinsky, 1996; Litwin-Kumar & Doiron, 2012).

Although it may not be ‘surprising’ to the reviewers that L4 spike rates and spiking statistics can be ‘accounted for’ by behavioral variables, the quantitative extent in behaving animals is new (and surprising to us).

In fact, we see little published prior art on these key points made in our paper:

A) Our measurements are in defined neurons in an active sensory perception task. Most of what we know about L4 responses in mice and rats comes from experiments in anesthetized experiments with passive stimulation. Even responses in the ganglion cells are very different with piezo stimulation compared to actively sensing rodents (Rasmus Petersen, Thalamus meeting Janelia 2015). Furthermore, anesthesia puts the brain in a non-natural state. Studies of neural coding have to be ultimately performed in a well-controlled behavioral task (Renart & Machens, 2014).

B) The existing literature clearly does not predict our results, even for simple metrics such as spike rate. For example, in the only other paper reporting recordings in L4 in awake animals (Curtis & Kleinfeld, 2009) baseline spike rates are high and modulation is relatively low – implying that a relatively small fraction of spikes would be explained by behavior. The difference might be due to recording techniques (cell-attached vs. wire electrodes), behavioral state, or tracking of the behavior.

C) Our quantitative estimate of the fraction of spikes that are behavior-related is a new analysis and observation. Our analysis goes well beyond qualitative assessments (i.e. classification as cells responding to touch or not). Importantly, our analysis could not have been done without millisecond timescale tracking of behavior.

D) The observed Fano factors are much smaller than any reported before, even though our measurements were done in active behavior. Careful tracking of behavior and best-of-class single unit isolation are critical for this analysis. There is a great interest in understanding and explaining variability in neural responses. For instance Goris et al. (2014) found that a stimulus-independent parameter can account for the super-Poissonian variability observed in multiple visual areas. Ecker at al. (2014) found that under anesthesia correlated variability is modulated by slowly fluctuating input that also significantly reduces the correlation among neurons. Our work highlights that the neural code can be close to noiseless ‘under battleground conditions’.

We have rewritten parts of the paper to make these points clearer (Introduction).

3) This said, the study also suffers from major limitations. Principally, the main finding is not novel – it is made to appear novel only by adopting strawman hypotheses, ignoring much of the relevant and recent literature.

We respectfully disagree with the statement that our study is not novel. To the best of our knowledge, the major take-home messages in our paper are new (also see response to Point 2).

A) The vast majority of spikes, in many cases >90%, can be accounted for in terms of tactile behavior.

B) The intrinsic variability of responses was substantially lower than expected for Poisson spike trains and often reached the theoretical minimum.

We are not aware of any studies that showed (A) and (B) in behaving animals in any cortical area.

We understand that ‘strawman’ refers to our Fano factor (FF) calculation, where we compare FF in the vicinity of touch (low) to the FF integrated over the full sampling period. This comparison could be easily removed without changing any of our conclusions. However, we feel that this comparison serves as an important didactic point.

The ‘strawman’ shows that FF can be affected dramatically if behavior is not measured with precision. The same would hold if neural dynamics is influenced by the hidden dynamics of unobserved brain areas (as is the case in most experiments where this sort of analysis has been attempted).

We have reworded the offending sections to dismantle the strawman (Results):

“Neuronal variability can arise from external factors […] at least in part due to trial-to-trial variability in active touch.”

We reply below to the point on recent literature on cortical noise.

4) Indeed, it is widely agreed that individual cortical neurons are not noisy: if cortical responses are variable, the variability is due to integration of inputs, which is mostly seen in central neurons.

Please note that we do not write nor imply nor state that ‘individual cortical neurons’ are noisy. In fact, in our Introduction we write:

“…noisy discharges are difficult to reconcile with the observed high reliability of cortical neurons (Mainen and Sejnowski, 1995) and groups of synapses (Stevens and Zador, 1998)”.

It is well established that intrinsically noisy networks can be constructed with reliable neurons.

We state that trial-to-trial variability in neuronal responses (e.g. spike count) has been observed in visual cortex, motor cortex, somatosensory cortex, both in behaving and anesthetized preparations (e.g. Churchland et al., 2010). Contrary to the reviewers’ assertions, the causes of this variability are unknown and an intense area of study (for reviews see Renart & Machens, 2014). Our point is that if extrinsic factors are taken into account cortical neurons are remarkably precise (as precise as they can be!). One important corollary is that we do not need network models that are intrinsically noisy.

5) Moreover, as the paper points out, layer 4 in this system is close to the sensory periphery, with little input from the rest of cortex – and we already know that in the sensory periphery there is very little noise. These and other criticisms are laid out below.

In terms of number of synapses L4 is as far from the periphery as L5 and L6; all of these layers receive direct input from VPM. The key point is that L4 receives long-range input almost exclusively from VPM (and recurrent input from other L4 neurons); it’s therefore a relatively simple cortical circuit and by monitoring sensory information we monitor a lot of the input to L4. In contrast, L5 and L6 receive inputs from other long-range sources.

We cite papers in which low response variability was recorded (below; however, none during active behavior):

“However, some studies in the sensory cortex suggest that the apparent irregularity in spike trains does not reflect noise but the hidden states of a temporally rich signal (DeWeese, Wehr and Zador 2003; Gur, Beylin and Snodderly 1997; VanRullen, Guyonneau and Thorpe 2005; Amarasingham et al. 2006; Kara, Reinagel and Reid 2000).”

Many recent high profile papers report high FF in multiple brain areas:

Churchland et al. (2010). FF: 1-2 in V1, V4, MT, LIP, PRR, PMd, OFC.

Carandini (2004). FF ∼= 1.

Goris et al. (2014). From LGN, V1, V2, MT. FF typically in 1-10 range.

Mitchell, Sundberg, & Reynolds (2007). V4, FF ∼=1.4.

Cohen and Maunsell (2009). V4, FF ∼=1.

Wang, Webber, & Stanley (2010). L4 in S1 FF∼=0.8.

Adibi, et al., Arabzadeh (2013). S1. FF = 0.8-1.2.

Bale & Petersen (2013). FF in S1 = 1.3, but close to minimum FF in ganglion (∼0.3), and still low in VPM (0.5). These results contradict the reviewers comment that as L4 is close to the periphery it is expected to exhibit also low noise.

Theoretical papers that are modeling high FF:

Litwin-Kumar & Doiron (2012). They show that slightly non-random networks cause cortical variability and stimulus-dependent quenching of neural variability. The idea that cortical neurons have FF>1 motivates this paper.

van Vreeswijk & Sompolinsky (1996). Influential paper in which a balanced state is proposed that can explain large FF of cortical neurons. This type of model has been the standard when trying to do simulations to generate poisson-like firing statistics.

Beck, et al., Pouget (2012). They suggest that Poisson noise actually reflects suboptimal inference.

By simplifying the problem (choosing a well-known, relatively simple circuit; controlled behavior; gold-standard recording techniques and appropriate analyses) we discovered very low variability. Our paper thus calls into question that cortical circuits inherently generate noisy discharges.

Recent review papers call for reassessment of the idea of cortical noise (Renairt & Machens 2014; Masquelier, 2013).

Masquelier: “High trial-to-trial variability in response to repeated presentation of a same stimulus has been reported in every modality. It is often quantified in terms of reliability and precision (Box 1), and both are usually poor in vivo (e.g. Fano factors ∼1 and precision ∼tens of ms or above). The origin of this variability, and its implication for information processing, has been much debated (Stein et al., 2005; Ermentrout et al., 2008; Faisal et al., 2008; Tiesinga et al., 2008; Rolls and Deco, 2010), yet a consensus has not emerged. Here we argue that most of the observed variability could come from uncontrolled variables, or the use of inappropriate reference times, rather than from intrinsic sources of noise (“intrinsic” meaning that they cannot be eliminated).”

We believe that our study sheds light onto these issues. The citations above clearly show that many researchers in the field feel that this is important and unsettled business.

6) Premise of the paper. The Introduction begins withCortical spike trains are often considered to be noisy, and casts much of the paper in the context of this view. But after decades of work, this idea of cortical neurons beingnoisyhas become a strawman. It is out of step with crucial pieces of the literature. We already know that cortical neurons are not intrinsically noisy; see e.g. Mainen and Sejnowski,Reliability of spike timing in neocortical neurons. Science 1995, and Deweese and Zador,Shared and private variability in the auditory cortex. J Neurophysiol 2000. Also, the apparent Poisson nature of variability can be explained by a deterministic spike threshold (Carandini, Plos Biol 2000). Similarly, the statement thatIn cortex, irregular spike trains suggest that only rate codes can be used to perform reliable computationsis misleading, and based on a reference that is over 10 years old.

Please see the responses above.

We further disagree that we are misleading with our statement that it has been argued that irregular spike trains suggest that only rate codes can be used to perform reliable computations. London et al. (2010) and others make exactly that argument. The title of their paper is: “Sensitivity to perturbations in vivo implies high noise and suggests rate coding in cortex”. And they take variability in cortical discharges for granted: “It is well known that neural activity exhibits variability, in the sense that identical sensory stimuli produce different responses”.

Mainen & Sejnowski (1995; cited) is slice work and we have no disagreements with them. Neural networks can generate noisy discharges with deterministic neurons (there are literally hundreds of papers on this). In the last 10 years nothing has happened as far as we can tell that has challenged the status quo.

We feel that we present the status quo fairly in our revised Introduction.

7) Vision literature. The paper uses the literature from visual cortex to make the point that cortical responses are noisy. But even in that literature there is much that contradicts this point. In a paper titledLow response variability in simultaneously recorded retinal, thalamic, and cortical neurons(Neuron, 2000), Kara, Reinagel, and Reid have already shown that responses in layer 4 of visual cortex can be remarkably repeatable. And these results are not limited to anesthetized preparations: similar results have been obtained in awake monkeys, where (as one could have imagined, and similarly to what is done here for whiskers) it is essential to know eye position on a moment by moment basis: Gur, Beylin, and Snodderly,Response variability of neurons in primary visual cortex (V1) of alert monkeys(J Neurosci, 1997).

We cite Gur et al. (1997). They show that responses in V1 are more reliable when fixational eye movements are taken into account. This paper is indeed important prior art and is discussed as such. However, this study does not show that responses are close to the theoretical minimum (which would be FF ∼ 0 for their high spike counts). Kara et al. (2000; also cited) show that responses can be sub-Poissonian in anesthetized cats.

8) Somatosensory literature. If we now turn to the recent literature on somatosensory cortex, we see that it makes precisely the opposite case relative to neurons beingnoisy. Indeed, the paper might just as well have begun withCortical spike trains are often considered to be made up of precisely timed spikes, whereby the timing of each spike carries some meaning about the stimulus.See for example Arabzadeh, Zorzin and DiamondNeuronal encoding of texture in the whisker sensory pathway(PLoS Biol 2005) andDeciphering the spike train of a sensory neuron: counts and temporal patterns in the rat whisker pathway(J Neurosci 2006). Additional contributions that should be considered in rethinking how to cast these results include those from G. Stanley (admittedly anesthetized, as noted in this study), E. Arabzadeh, G. Foffani, R. Petersen, and S. Panzeri. The latter includeInformation carried by population spike times in the whisker sensory cortex can be decoded without knowledge of stimulus time(Frontiers, 2010) – which is related to the authors' conclusion thatPooling activity from only fifteen L4 neurons […] in C2 was sufficient to detect 95 % of touches– andComplementary Contributions of Spike Timing and Spike Rate to Perceptual Decisions in Rat S1 and S2 Cortex(Current Biology, 2015). The latter is admittedly too recent to have shaped the present paper, but it is now in the literature.

It’s been known since the classic studies of Dan Simons (cited) that after passive whisker deflection L4 neurons spike with short latencies and little timing jitter. We cite this and other papers in our Introduction.

We show similar temporal precision in L4 neurons during active sensation. Furthermore we show that spike rates are extremely low except in a narrow time window after touch. This picture is very different compared to the only other L4 recordings during active behavior that we are aware of (Curtis and Kleinfeld, 2009); this study shows high baseline spike rates and thus relatively small fraction of spikes explained.

Importantly, rapidly changing spike rates do not imply low variability! Bair and Koch (1996) have already shown temporally precise responses with large FF. Where it’s been looked at variability has been relatively high in anesthetized barrel cortex (FF ∼ 1; e.g. Adibi et al., 2013).

We thank the reviewers for pointing out the papers from Diamond’s lab. We now discuss one of them in our revised manuscript (Panzeri et al. 2014; in the subsection “Decoding touch”).

We note in passing that Diamond too acknowledges the need to identify the sources of cortical variability (Arabzadeh et al., 2005):

“In a number of sensory modalities, first-order neuron responses can be remarkably reliable when a stimulus is presented repeatedly, whereas cortical responses vary across trials. It is of interest to elucidate the mechanisms that permit reliable first-order neuron responses and, by the same token, to identify the sources of trial-to-trial variability among cortical neurons.”

“From these observations we conclude that, under our experimental conditions, the trial-to-trial response variability of first-order neurons is caused exclusively by stimulus jitter, whereas that of cortical neurons results mainly from variations across time in sensory integration, and must emerge at some integration site between the trigeminal ganglion and cortex. A question of current interest is whether the variability in cortical responses results from noise and imprecision in neuronal integration, or else reflects functionally significant modulations in responsiveness.”

9) Poisson spike generators. Given the literature, the Poisson model with a refractory period has become a strawman. Therefore, the simulation of Poisson firing with refractory period (in the sixth paragraph of the subsection “Low spike count variability”) is unsatisfying. We already know that the Poisson model is not a reasonable model, even with refractory periods. We know this in retina, in LGN, in V1, and in A1. It would be more interesting to test a more plausible model. For instance, would an integrate-and-fire model with Gaussian noise in Vm give the observed statistics?

We respectfully disagree with this characterization of Poisson models. Spike trains are almost always modeled as Poisson trains.

Recent papers (e.g. Goris, Movshon, & Simoncelli, 2014) insist on the Poisson model with some variations. All GLM models we are aware of have an output that is a Poisson generator (refractoriness is obtained with a post-spike filter). Given that we observe variability below Poisson, the next simplest model would be to add a refractive period. This is a statistical, not a mechanistic model (as the reviewers propose). Our analysis says that the low variability we report is not due to very high spike rate changes followed by a refractory period (which was the first possible explanation we considered).

10) Fano Factor. Related to the last point: is it sensible to calculate Fano Factor (FF) for such a radically non-stationary process? Fano Factor is at best awkward for non-stationary processes and the main message of the paper is that spiking is non-stationary, e.g. extremely variable over large time scales and extremely precise when whiskers kinematics dictate so. For a Poisson process, FF = 1, independent of the window size. Since the observed FF for full stimulus window was large, evidently a Poisson process is not underlying spikes. At best it is a Poisson process driven by a variable mean rate. Indeed, Poisson statistics are meant to model stationary processes, that is, processes where the probability of an event per unit time is constant. Obviously firing probability increases in the tens of ms after touch, compared to the rest of the trial.

We respectfully disagree. FF is appropriate for analysis of non-stationary processes. There are two types of non-stationarity that have different impact on the FF. First, a time-varying firing rate that is the same in every trial, referred to as an inhomogenous Poisson process. The Fano factor is often used to characterize non-stationary processes, precisely because FF = 1 even for an inhomogeneous Poisson process (e.g. Berry and Meister, 1997). Second, the other type of non-stationarity corresponds to changes in mean firing rate across trials. In this case, for a neuron that fires with Poisson statistics according to a stochastic rate in each trial, the Fano factor will be larger than 1 (i.e. it is a doubly stochastic process).

11) Further, having demonstrated that the neurons of interest in barrel cortex do not resemble a Poisson process, the modeled neurons do:We used a simplified model based on independent Poisson neurons with rapidly modulated spike rates. It is puzzling to model Poisson neurons once one knows that the model is not correct.

We now use spike trains sampled from recorded spike trains for modeling (new Figure 4).

12) Factors accounting for spike trains. In the present analysis (in the second paragraph of the subsection “Low spike count variability”), 3 factors are used to account, one at a time, for spike trains: order of touch, whisker velocity, and whisker curvature. Why use them one at a time? Why not use all 3 at once and see how much better one does? Or use only the two main ones? Are there any spikes unaccounted for once one does this? This could be done with a GLM (perhaps enlisting the help of colleague Jeremy Freeman) or with similar approaches. Indeed, since the main message of the paper is that spike trains can be explained by whisker curvature and related kinematics, it would make sense to place less emphasis on the Poisson analysis and more emphasis on showing whisker state as a determining factor.

The point of this figure is to show that touch-evoked spikes vary with the quality of the touch in multiple, correlated dimensions. This sets up the message of Figure 6, which is that spike count variability during touch is minimal once these touch characteristics are considered. Because of the low spike rates and spike counts, number of sensory variables, and limited experiment duration, it is difficult to fit models with more parameters.

The use of GLM models for this work is problematic for the following reasons:

A) Amount of data and overfitting. In a recent imaging study with a GLM-like model (Peron et al., 2015) we observed that we needed at least ∼100 trials to fit GLMs with only two variables and nine time bins (this was an imaging study with frame rate ∼ 7.5Hz). At a resolution of 1ms and much more temporal richness than in imaging we’d need many more time bins to describe the data, which increases the risk of overfitting. Regularization and carefully chosen temporal basis sets may overcome this problem under some conditions, but this is far from trivial. In general we are limited by trials and spikes.

B) GLMs typically use linear kernels and a static non-linearity. However, phase preference extracted from whisking cannot be linearly mapped. This means that standard GLMs cannot be applied to extract modulation with whisking.

C) Adaptation, to a first-order, can be accounted using a post-spike filter. But this significantly increases the number of model parameters and given limited data makes fitting impossible.

D) It is not possible in the classical GLM to generate near-binomial responses and still capture the overall ISI distribution.

E) Ideally every input variable should have a white-noise spectrum to fully sample the system. Temporally and spatially inhomogeneous sampling (across different input variables), as is the case in active behavior, can give unreliable estimates of the GLM parameters.

F) Adaptation of GLMs for trial-based, spiking neurons, with multiple correlated features were only recently developed (Park…Pillow, 2014). Even in this case, most of the inputs are discretized (saccade direction, stimulus onset, coherence level), implying relatively low dimensionality and few parameters.

This is just to say that we simply don’t know how to apply ‘GLM or with similar approaches’ (even with Dr. Freeman). To show that we have seriously considered this approach we show GLMs to fit the time-course of touch-responses (see Author response image 1).

Author response image 1.

Author response image 1.

GLM modeling of L4 touch responses. (a-b) Two example neural responses (blue) aligned to touch onset and their corresponding GLM prediction (red). (c-d) Same as (a-b) but aligned to touch offset. (e-f) Touch adaptation in function of touch number. (g-h) Proportion of spikes explained as a function of the exploration time (similar to Figure 3 in the paper). The blue curve (data) was obtained as described in the paper. For the GLM we started counting spikes ranked by the maximum GLM prediction (i.e. the moments with highest probability of a spike occurring according to the GLM model).

DOI: http://dx.doi.org/10.7554/eLife.06619.019

The GLM models touch onset and touch offset. To deal with adaptation, first touch and subsequent touches were treated separately. As expected, the GLM captures the touch-aligned responses well (constraints for smoothness mean that the GLM doesn’t capture sharp temporal responses). We tested if the GLM can help predict a larger fraction of spikes than the approach presented in the paper (Figure 3). Indeed, the GLM does slightly better (i.e. the proportion explained curves rise faster than with the approach we adapted) but the difference was small. We therefore chose to stick with the approach that was more closely related to the primary data.

13) Other layers. Rather than focusing solely on L4, it would have been much more interesting to compare L4 with other cortical layers, and investigate the sources of variability there. Indeed, a major missing component is that neurons outside L4 are not recorded or reported. This is set up in the Introduction, but not really addressed. It's great to show L4 does one thing, but more interesting to show that it does something different than other layers. Given other publications from this group, it seems likely that the authors have other data that addresses this. Analyzing these data would make the paper much stronger.

L4 is the core of this study. As discussed in point 5), other layers receive multiple long-range inputs that can make the activity more variable (since the input is not observed). However, we agree with the reviewers that adding L5 data would strength our paper by showing that the L4 is distinct from other layers. We repeated several of our analysis with a new set of L5 neurons. Whereas the vast majority of spikes in L4 can be explained by touch and whisking, only a small percentage of spikes can be explained in L5.

The new L5 data has been included in Figure 2C, D and Figure 3C, D and Figure 3–figure supplement 2.


Articles from eLife are provided here courtesy of eLife Sciences Publications, Ltd

RESOURCES