What single-cell stimulation has told us about neural coding

Abstract

In recent years, single-cell stimulation experiments have resulted in substantial progress towards directly linking single-cell activity to movement and sensation. Recent advances in electrical recording and stimulation techniques have enabled control of single neuron spiking in vivo and have contributed to our understanding of neuronal coding schemes in the brain. Here, we review single neuron stimulation effects in different brain structures and how they vary with artificially inserted spike patterns. We briefly compare single neuron stimulation with other brain stimulation techniques. A key advantage of single neuron stimulation is the precise control of the evoked spiking patterns. Systematically varying spike patterns and measuring evoked movements and sensations enables ‘decoding’ of the single-cell spike patterns and provides insights into the readout mechanisms of sensory and motor cortical spikes.

1. Introduction

Elucidating the link between neural activity and sensation or motor control remains a major challenge in neuroscience. Classical recording techniques provide strong correlations between neural activity and sensation. However, establishing causal links between spike patterns and perception became possible only with the introduction of ‘reverse physiology’ approaches, in which behavioural responses are analysed in response to induction of cellular activity. The first attempt to manipulate cortical activity was that of Fritsch & Hitzig [1] who applied cortical electrical stimulation in dogs, demonstrating for the first time a connection between a localized brain region and motor function. These findings were later extended to humans, where electrical stimulation of different cortical regions yielded sensations and movements [2]. The later development of intracortical microstimulation (ICMS) [3–5], a technique in which trains of short (100–200 μs) constant electrical pulses of small current intensities (1–100 μA) are delivered extracellularly via a microelectrode at rates of tens to hundreds of Hertz, enabled a more reliable activation of localized populations of neurons and directly influenced sensory perception [6–8], movement [3,9–11] and cognition [12–14].

Until recently, electrical microstimulation was the only method that allowed researchers to precisely modulate neuronal activity in animals performing behavioural tasks and investigate its underlying neural code. In a set of pioneering experiments, Newsome and co-workers demonstrated a causal relationship between the physiological properties of neurons and perception [6,15,16]. They trained monkeys on a visual detection task and after characterizing the direction selective columns in the middle temporal (MT) visual area, they applied microstimulation and strongly biased the animal's perceptual judgement towards the preferred direction of the stimulated column. Furthermore, by parametrically varying the amplitude and frequency of the signal they showed that while increased amplitude resulted in reduction of the behavioural performance, an increase in the signal frequency preserved the directional specificity and altered its perceived intensity [17]. Similarly, microstimulation of disparity-tuned columns in the same area can also bias stereo-depth perceptual decisions [18,19]. Manipulation of discrimination performance by microstimulation was also achieved in the somatosensory area [7,20]. Here, investigators trained monkeys on a frequency discrimination task and then substituted the mechanical stimulation of one or both stimuli by cortical stimulation of the appropriate somatosensory area. Surprisingly, the subject could perform the task normally, indicating that stimulation of these neurons is sufficient to produce discriminable perceptions. Furthermore, by parametrically manipulating the periodicity of the microstimulation, they found no change in the psychophysical performance between periodic and aperiodic trials, and concluded that periodicity does not play a role in such frequency discrimination tasks [7,20]. In another study on the rat's barrel cortex, Butovas & Schwarz [21] trained rats to report direct microstimulation by licking. They found that stimulation frequency and intensity were positively correlated with behavioural performance on this task. Unexpectedly, double-pulse stimulation was found to be more effective in producing perceptual effects compared to longer stimulus trains.

ICMS activates populations of neurons, whereby the number of activated neurons can only be estimated and will vary on injected current intensity and frequency. Interest in the meaning of the activity of single cells has been ignited by converging evidence from several experimental approaches, which suggested that neural activity is more sparse than previously thought [22–28]. The purpose of this review is to focus on how single neuron stimulation experiments have contributed to this progress. We first describe novel techniques for achieving control of single neuron spiking in vivo. Next, we show that single-cell stimulation effects differ across brain structures and cell types. Finally, we discuss how single-cell stimulation effects vary with artificially inserted spike patterns compared to other stimulation techniques.

2. Effects of stimulating single sensory neurons

(a). Report of single afferent stimulation

The first evidence for an effect of single neurons on sensation came from the pioneering experiments of Vallbo and co-workers in 1984 as shown in figure 1a [29]. The investigators applied intranerve microstimulation [32–34] to electrically stimulate single mechanoreceptive fibres innervating the hand of human subjects, who experienced a subjective sensation as a result. In half of the stimulation experiments, the evoked sensation was specific to locations at the subject's skin surface and matched the receptive field of the stimulated fibre. Often a stunning correspondence between receptive and ‘perceptive’ field was observed (figure 1b). Usually, subjects described the stimulation as a light indentation. Importantly, the sensation remained stable with increased stimulation current amplitudes until a second threshold was crossed and a new sensation was reported, implying that a second fibre was activated, producing a separate sensation. The seminal findings of Vallbo sparked interest in sensory correlates of single cell activity. However, further progress investigating single neuron sensory correlates was delayed by the technical challenges of activating single cells in the central nervous system, as we discuss next.

Figure 1.

Stimulation of single sensory neurons. (a,b) Recording and stimulation of single afferent fibres isolated in the median nerve. (a) Schematic of the recording/stimulation arrangement, receptive fields are mapped using single-unit recordings, threshold level microstimulation is applied and percepts are verbally reported. (b) A stunning correspondence of receptive and perceptive fields, recorded and elicited by microstimulation of a slowly adapting (SA I) and fast-adapting (FA I) unit is observed. (c) Nanostimulation in the ventral posteromedial nucleus (whisker) thalamus and barrel cortex of the rat. See text for methodological details. (d,e) Single neuron stimulation in whisker thalamus does not lead to a behaviourally detectable effect. (d) Recording of a putative pyramidal neuron during 200 ms nanostimulation trials, a no-current-injection catch trial and microstimulation. Action potential (ticks) raster plots and first lick responses (red squares) during single thalamic neuron by 200 ms 16 nA current step (top), catch trials (middle) and microstimulation (40 cathodal pulses at 200 Hz, 0.3 ms pulse duration with 6–7 µA; bottom). Only 28/42 microstimulation trials are shown. No sensory effect is observed. (e) Population data. Each circle represents one single-cell stimulation experiment. Response rates in single-cell stimulation trials (hits) are plotted versus response rates during catch trials (false positives), n = 36 neurons; note several points coincide. (f,g) Single neuron stimulation in barrel cortex causes biases towards responding. (f) Stimulation of a putative cortical excitatory neuron (4 nA current step). Only 20/59 microstimulation trials are shown. No activity is shown during microstimulation (4 µA, grey area) because it could not be measured. Conventions as in panel d. (g) Population data for barrel cortex neurons (n = 270 neurons; note several points coincide). Conventions as in panel e. Adapted from Vallbo et al. [29] (a,b), Voigt et al. [30] (d,e) and Doron et al. [31] (f,g).

(b). Techniques for single neuron stimulation in vivo

Intracellular stimulation was the first technique that allowed controlled single-cell stimulation in the intact brain. The development of intracellular and whole cell recording techniques [35] made stimulation of identified single neurons possible in vivo, owing to its increased stability and the possibility of rapid loading of fluorescent indicators or dyes [36]. Importantly, intracellular stimulation provides excellent spike train control owing to its direct access to the membrane potential, allowing injection of various waveforms and reliable control of precise spike timing [37]. However, these techniques usually are difficult to apply in awake animals, and thus often result in only a limited number of successful trials. In addition, the duration of application of intracellular stimulation is limited owing to the deterioration of the preparation with time [38–41]. Another disadvantage of blind whole-cell recording, where neurons are searched using evoked square voltage steps [40], is the arbitrary and non-representative sampling of the manipulated neuron, which hinders comparison and data analysis across different brain areas. However, this drawback can be overcome by recovering neurons or by genetic labelling of neuronal subpopulations and visually targeted patch clamping recordings, where the latter is limited to the upper layers of superficial cortex [42,43].

To allow long-lasting, stable access to stimulate single cells, a new technique was needed. Thus, we developed the ‘nanostimulation’ approach as an alternative to intracellular stimulation, which involved juxtacellular current injection [44]. Nanostimulation, unlike intracellular stimulation, is easy to apply and does not facilitate run-down owing to dialysis or leakiness conferred by intracellular methods. This allows for multiple recording sessions that can last hours. Juxtacellular stimulation was originally developed by Pinault and Deschenes as a method for single neuron labelling [45,46]. Here, we use ‘juxtacellular’ to describe cell-attached stimulation used for labelling, and ‘nanostimulation’ to describe the cell-attached stimulation we use to evoke and record spikes. Juxtacellular stimulation has been used in a variety of labelling studies [47–52]. The term nanostimulation is in part a reference to the minute current injection (in the nanoamp range), but also a reference to the precise modifications in action potentials (APs) discharge of single, identifiable neurons. Because of its specificity, it can be used to stimulate a diversity of identifiable neurons in anaesthetized and awake animals. Furthermore, it permits control of AP frequency and number through positive current injections as well as inhibition of spiking in single neurons using negative currents. Sustained DC current injections result in AP responses with large temporal jitter, similar to AP responses evoked by DC current injections in the intracellular configuration [37], whereas injection of arbitrary waveforms, using bandpass limited Gaussian noise, results in increased spike timing reliability [53].

(c). Nanostimulation in the thalamus

The unexpected finding that single fibre stimulation could be sensed led to several follow up studies on the behavioural detectability of single neuron stimulation in the rat's ventral posterior medial (VPM) thalamic nucleus, the main thalamic target of whisker input [30] to the barrel cortex, which is responsible for whisker information processing [54]. In order to assess the rat's ability to behaviourally report activation of single neurons the authors trained rats to detect low intensity microstimulation. Specifically, the animals were conditioned to lick in response to 200 ms microstimulation trains (200 Hz) at the VPM and were rewarded with sweetened water with decreased intensities until a threshold of below 5 μA was reached. Importantly, the animals learned to report microstimulation within a single session and usually reached threshold levels within only two to three additional sessions. Then, the researchers intermingled microstimulation trials with catch trials, in which no stimulation was given, and nanostimulation trials to evoke neuronal activity in single neurons. This reward scheme, along with no punishment of false alarm report of catch trials, biased the rats towards a strategy of responding to stimulation under uncertainty conditions. In contrast with detectability of electrical stimulation of single tactile afferents, figure 1c–e shows that stimulation of single VPM neurons could not be behaviourally reported [30]. The rats' inability to detect this minute perturbation may be due to local neural circuitry of the thalamic nucleus, which lacks local recurrent excitatory connectivity [55,56] and thus means of signal amplification. In addition, the authors noted that the thalamic synapses might be too weak to evoke a sensation [57,58].

(d). Nanostimulation in the cortex

While stimulation of single neurons of the thalamus was not detected by rats, stimulation of single cortical neurons in the barrel cortex could be behaviourally reported by rats [31,54]. Overall, the authors found that nanostimulation of a single neuron (adding approx. 15 APs to the baseline firing activity) biased the animals towards responding 4–5% more than expected by chance (by comparing with catch trials). This effect was usually small and variable, compared with microstimulation, and became significant only by pooling data across cells. These results indicate that in certain conditions, activation of only a single barrel cortex neuron can be sensed (figure 1c,f,g). One possible explanation for the positive results might be recurrent excitation, observed in pyramidal cortical neurons but not in subcortical areas [59]. This implies that activation of a single neuron would directly affect other nearby cortical neurons. Interestingly, stimulation of single inhibitory neurons, identified by labelling or by maximal evoked firing rate, was reported more often by the rat compared with excitatory neurons [31,54]. Hence, certain types of inhibitory interneurons may play a significant role in controlling spiking in the local neural microcircuits that mediate sensation.

3. Effects of stimulating single neurons in the motor system

(a). Motor neurons

The rodent's vibrissal motor system offers several advantages for analysis of motor control, primarily owing to the ability to accurately measure speed and displacement of single whisker movements, availability of powerful methods for high-speed tracking such movements, and the ease of manipulation of single whiskers [60]. The intrinsic and extrinsic muscles surround the whiskers and produce rhythmic movements via consecutive contractions that cause whisker protraction and retraction, respectively [61–63]. The lateral facial nucleus receives direct projections from the vibrissal motor cortex (VMC), indicated by viral-based anterograde tracer labelling [64], and is the sole brain structure which sends cholinergic motor inputs onto these muscles. It contains large neurons whose numbers are orders of magnitude fewer than in the VMC [65], thus forming a potential bottleneck in vibrissa motor control. A recent study used nanostimulation to examine the effect of a single facial nucleus motor neuron on whisker movements in anaesthetized rats [66] (figure 2a–d). The investigators found that evoking spikes in about half of the neurons resulted in whisker movements of various amplitudes. Nanostimulation in most neurons caused protractions of single whiskers (figure 2c), whereas a minority of the stimulated neurons resulted in complex retraction movements in several whiskers simultaneously. Interestingly, every stimulated spike resulted in a whisker deflection (figure 2d). A closer examination of the effect of multiple evoked spikes led the authors to conclude that shorter interspike intervals caused merged, slightly larger amplitude whisker deflection than expected by a linear superposition of single spike effects [66].

Figure 2.

Stimulation of single motor neurons. (a–d) Example of whisker protractions evoked by single neurons in facial nucleus of rat. (a) Schematic of the rat's face and whiskers. (b) Reconstruction of the dendrites (red) and axon (blue) of the stimulated neuron in the rat facial nucleus. (c,d) Motor effects of single (c) and multiple spikes (d). Movement trajectories of whiskers B4, B3 and C2 (c,d, top) evoked by single or multiple spikes (c,d, bottom). (e,f) Whisker movements initiation by intracellular stimulation of a motor cortex pyramidal neuron. (e) Top, position of whisker E1 (wE1) in an intracellular stimulation trial (10 APs at 50 Hz). Bottom, membrane potential recordings and current injection steps traces. Arrows mark whisker positions. (f) Top view of rat head and foil-labelled (bright spots) whiskers (left) and superimposed images of pre and poststimulation (black and white outlines, respectively). Adapted from Herfst & Brecht [66] (a–d) and Brecht et al. [67] (e,f).

(b). Motor cortical neurons

The primary motor cortex (M1) was the first brain structure whose functional organization was explored by electrical stimulation techniques, revealing its role in motor control [1,68,69]. In contrast to sensation where variability can contribute to processing of incoming stimuli, in the motor system it is unavoidable owing to the highly variable motor output produced by muscle activation, depending on the time and complexity of the evoked motor command. This variability is reflected well in the motor cortex where movement direction cannot be predicted from activation of single neurons, which fire broadly to movement [70–72]. Owing to this variability it has been estimated that tens to thousands of neurons would be needed in order to accurately decode and control movement in disabled subjects [73–76]. However, both a quantitative analysis of extracellular recordings by a spike population vector [77,78] and by real-time analysis of AP activity in M1 [73,79,80] suggest that complex movements can be caused by activity in only tens to hundreds of neurons. If only small numbers of neurons, which spike sparsely, can affect movement generation, could a similar amount of AP discharge from a single neuron also evoke similar action? Extracellular stimulation studies in M1 indicated that very low activation thresholds of only 1–2 µA are sufficient for initiation of muscles twitches and skeletal movements [81]. Furthermore, pyramidal neurons activity can affect single muscles, as found using extracellular spike correlation in single cortical motor neurons and electromyogram (EMG) activity [82], and intracellular stimulation of a single motor neuron can also result in EMG changes [83]. However, these studies could not resolve whether stimulation of single neurons could produce overt movements, because low-current extracellular stimulation might produce a small number of spikes in many neurons, or may weakly activate a small number of neurons. Therefore, Brecht et al. [67] applied intracellular stimulation in M1 of lightly anaesthetized rats. Twenty per cent of stimulated neurons in this preparation evoked whisker movements (figure 2e,f). The whisker movements evoked by intracellular stimulation were usually of small amplitude (on the average about 0.5°) and complex, as they contained long sequences of multiple whiskers deflection lasting many seconds, suggesting that single neurons represent features of disparate motor programmes rather than specific muscle contractions.

4. Readout mechanisms of single neuron activity

(a). Postsynaptic spiking evoked by single-cell stimulation

One candidate cell type to mediate single neuron effects in the cerebral cortex is the Martinotti cell, a somatostatin-positive (SOM) interneuron. Excitatory inputs to this neuron type show robust facilitation, and repetitive activation of a single synaptic connection makes the cell fire [84–86]. In addition, SOM cells were shown to modulate surround suppression in layer 2/3 pyramidal neurons in mouse primary visual cortex (V1), thus contributing to contextual processing of sensory information in this region [87]. Evidence for the importance of these interneurons in translating single neuron stimulation comes from a study which used two-photon imaging along with intracellular stimulation of single pyramidal neurons [88]. The researchers estimated, based on readout of postsynaptic firing with two-photon calcium imaging, that a burst of approximately five spikes in a single pyramidal neuron could activate a small percentage of neighbouring pyramidal neurons (2%, approx. 14 cells) and a large percentage (30%, 3–9) of SOM interneurons (figure 3a). Interestingly, no significant activation of parvalbumin-positive (PV) fast-spiking interneurons by pyramids was reported in this study. When considered with the finding that activity in single fast-spiking interneurons can have a powerful effect on a rat's behaviour [31,54] and that microstimulation frequently results in long-lasting inhibition of stimulated neurons [91], it suggests that animals can not only sense the addition of spikes in the cortical network, but that they may also be sensitive to inhibition, i.e. putatively the subtraction of spikes in such local circuits. A recent study examined the connectivity of different interneuron subtypes in the primary visual cortex of mice and found that PV fast-spiking interneurons mostly inhibit other PV interneurons, suggesting that stimulating a fast-spiking neuron can also lead to disynaptic inhibition and increased firing in the network [92]. Altogether, these types of single-cell stimulation studies in identified interneurons subtypes may reveal unique insights into the functional micro-circuitry of the neocortex.

Figure 3.

Readout mechanisms of single cell activity. (a) Examples of spatial maps of the functional connectivity between a single stimulated pyramidal neuron (blue) and imaged neurons in mouse primary visual cortex: left, somatostatin-positive (SOM) interneurons (red); middle, parvalbumin positive (PV) interneurons (green); right, putative pyramidal (Pyr) neurons (black). (b) Single neuron stimulation induces brain switch from persistent-UP to up/down state (left) and vice versa (right). Top, sample whole-cell recording traces during periods marked by arrows. Bottom, state switch indicated by change from bimodal to unimodal membrane potential (E_m) distribution (colour coded, computed in 20 s windows). Blue bar, burst spiking period. (c) The effect of adding an extra spike on network activity in vivo. Left, schematic of the recording configuration (red, extracellular silicon probe; blue, patch electrode); middle, example of extracellular spikes (top) and intracellular membrane potential trace (bottom); right, peristimulus time histogram triggered on the stimulus and binned at 5 ms; inset: change in firing rate per neuron. Adapted from Kwan & Dan [88] (a), Li et al. [89] (b) and London et al. [90] (c).

(b). Single neuron effects on population activity

Dan and co-workers [89] also used intracellular stimulation to examine the effects of single neuron activation on brain states. They reported that repetitive high-frequency burst spiking of a single cortical neuron could trigger a switch in the global brain state between slow-wave and rapid eye movement sleep patterns (figure 3b). The effect of single neuron activation on large neuronal populations was also demonstrated in the developing CA3 hippocampus, where stimulation of distinct populations of GABAergic interneurons affected spontaneous network synchronization [93]. By combining two-photon calcium imaging with intracellular stimulation of individual interneurons in slices, the authors reported that about 40% of activated neurons, termed ‘hub neurons’, could alter the on-going network activity. Interestingly, these anatomically identified neurons showed high functional connectivity and could be separated into synchronization initiation or delaying neurons, based on their anatomical features. Similarly, it was shown that activation of individual interneurons in hippocampal slices can both suppress and enhance the local generation of sharp wave ripples [94]. Another related study used an optical probing technique to examine the connectivity patterns of mouse layer 5 corticotectal neurons and found that ‘trigger’ neurons evoked activity in ‘follower’ neurons belonging to selective anatomical classes with stereotyped physiological and synaptic responses, and surprisingly also in precise positions across animals [95]. Furthermore, a recent study in human brain cortical slices examined interactions between human pyramidal neurons and their postsynaptic targets using whole cell recordings [96]. Remarkably, they found that single evoked spikes were followed not only by monosynaptic events but also by complex and long-lasting event sequences composed of inhibitory and excitatory effects.

(c). Effects of single cortical spikes

The influence of single APs on network firing was elegantly experimentally demonstrated in vivo in rat barrel cortex (figure 3c) [90]. Combining experimental work with a theoretical model the authors calculated that a single spike in a layer 5 pyramidal neuron results in extra APs in approximately 28 postsynaptic neurons. They showed that such perturbation may lead to short-lived (less than 5 ms) detectable changes in network activity, but ruled out a possibility of long-lasting effects, owing to induced inhibition, which would cancel out the evoked spike. This work is in accord with Kwan & Dan [88]; both studies show that inducing spikes in one cortical cell produces more than 10 extra spikes in other cells in the cortical network.

5. Effects of parametric variation of spike patterns on movement and sensation

(a). Sensory and motor coding

It is still debated how spike train characteristics represent content in the cerebral cortex [97–100]. The rate-coding hypothesis emphasizes the mean firing rate in carrying information [101]. According to this view, the neuron encodes a stimulus by firing a specific number of spikes proportionally to some stimulus value parameter and the downstream neuron decodes it by simply counting spikes. In contrast, the temporal-coding hypothesis stresses the exact timing of individual spikes as the crucial factor in transmitting information [102,103]. While these distinctions are on some level simplifications to help guide theory and experimentation, nanostimulation is an ideal tool to probe these dichotomous views, find any potential overlap and explain their potential behavioural dynamics.

(b). Parametric control of single sensory neurons

In order to directly assess coding mechanisms underlying sensations, Doron et al. [31] trained rats to report activity of single excitatory and inhibitory neurons in the barrel cortex using a nanostimulation detection task (figure 4a–d). The effects of three different stimulation parameters were compared: AP frequency, number and irregularity. To parametrically control the elicited AP frequency, step currents of varying durations and intensities were combined. Because increased duration and intensity of nanostimulation pulses results in increased AP firing [44], such combinations evoked similar numbers of APs over different durations, creating different stimulation frequencies. In a similar fashion, manipulation of single step current duration without changing its intensity evoked an increased number of spikes at a constant frequency. To parametrically vary the irregularity of the evoked APs, pseudo-random sequences of short step currents of different durations and frequencies as well as periods of inhibition by negative step current pulses were combined [44]. These trials were compared with a single step current of similar duration, which resulted in a regular firing pattern.

Figure 4.

Effects of parametric single-cell stimulation. (a) Recording and nanostimulation arrangement in the rat cortex; dorsolateral view of a rat cortical hemisphere, indicating somatosensory (S1) and motor (M1) cortices; inset: flattened maps of the primary M1 and S1 cortices. (b) Recordings of a putative pyramidal neuron during trials with randomized fluctuating current injection (top, two sample traces, 18 ± 6 spikes) or regular nanostimulation trials with single current step (bottom 13 ± 6 spikes); triangles indicate stimulation onset and offset artefacts. Action potential (ticks) raster plots and first lick responses (red squares; c) and response rate (d) for the two stimulation conditions. (e–h) Effect of AP number and frequency on evoked movements in a single motor cortex pyramidal neuron: (e,f) average movements of whisker stimulation trials with initiation of 10 APs at 10, 50 or 100 Hz (e) or with initiation of 2, 5 or 10 APs at 50 Hz (f). Vertical red lines indicate the 50% amplitude time point and the amplitude measured and horizontal red lines indicate the 20-80% rise times. (g,h) Population data for the effect of AP frequency on mean deflection amplitudes (g) and AP number on movement dynamics (h). Adapted from Brecht et al. [67] (a,e–h) and Doron et al. [31] (b–d).

Surprisingly, the behavioural response was largely modulated by the spike train irregularity for both putative pyramidal neurons and interneurons, with more irregular spike patterns being more often reported compared with regular trains. The observed behavioural effect was roughly twice as large (approx. 8–10%) when compared with previous work in the same laboratory using periodic stimuli [54]. In addition, AP frequency also influenced single excitatory neuron detectability, however, in an unexpected way: the lower frequency stimulation trials were more readily detected by the animal, whereas higher frequency trials resulted in a bias away from reporting single neuron activity. Here, the observed behavioural effect was of about 4–5% as previously observed from regular spike trains [54]. However, no effect of AP number was found for either cell type, implying that the animal does not respond by counting single spikes in this task. In order to evaluate the time window used to read out the evoked spiking patterns by potential downstream neurons, an ideal observer approach was applied. It was found that a time window of 200–300 ms before the induced lick was needed, implying that the rats used information contained in the entire stimulation train to evaluate its meaning. These data indicate that at least at the single neuron level, the timing of individual spikes over several hundreds of milliseconds provides critical information for sensation. The effects of spiking irregularity in single-cell stimulation experiments results deviate from previous results obtained with other neuronal stimulation methods, such as microstimulation and optogenetics, in several ways: first, the negative correlation reported between AP frequency and behavioural performance observed by stimulating a single pyramidal neuron is opposite to earlier reports on the effect of microstimulation frequency on psychophysical performance in the same species and brain region [21]; second, no effect of AP number was observed at the single neuron level, in contrast to a recent report using optogenetics in the visual and somatosensory cortices of mice [104,105]; finally, the effect of irregularity of APs on behavioural detectability of single neurons was not observed using microstimulation [106] or photostimulation [26,104]. However, Butovas & Schwarz [21] reported that microstimulation of double pulses, which induce short bursting activity, resulted in the lowest detection threshold of electrical stimulation, in agreement with the ability of rats to respond faster to induced high-frequency motifs of the irregular spike train. Notably, Lak et al. [107] found that induction of irregular whisker deflections resulted in a larger and more temporally precise activity in rat barrel cortex, suggesting that irregularity may be more efficiently encoded compared with a periodic stimulus [102,108]. However, it is unclear whether the perceptual effect of evoking a few spikes in many neurons is similar to inducing many spikes in a single neuron. Furthermore, it remains unknown whether temporally accurate spike sequences are interpreted more efficiently by downstream neurons or merely produce an upward bias towards response, as synchronous network activity would do.

(c). Parametric effects of single motor cortical neurons

It was suggested that the motor system might produce natural movement by minimizing the noise and reducing neuronal variability using increased neuronal population activity to control muscle activity [109]. However, a shortcoming of both extracellular electrical and optogenetic stimulation techniques is that they activate many neurons simultaneously, rendering it hard to elucidate the role of AP frequency and number on single neurons in M1. Brecht et al. [67] assessed the effect of spike frequency of single neuron stimulation by introducing train pulses of increasing frequencies and reported that frequencies above 50 Hz resulted in increased backward whisking amplitudes (figure 4a,e–h). Interestingly, a low frequency of 10 Hz caused a small forward whisking movement (figure 4e,g), a result that was confirmed by microstimulation at similar frequencies. The investigators also examined the effect of AP number by evoking spike trains of different durations and found no change in whisker amplitude following stimulation. Nevertheless, increased AP number had an effect on the latency and rise time of backward whisker movement (figure 4f,h). These effects of varying spike patterns in motor cortex are remarkably more complex and different from varying spike number in facial nucleus motor neurons (figure 2c,d). Whereas single spike effects simply add up in motor neurons, effects (amplitude and direction of the evoked movement) of motor cortical spikes depend on the spike ‘context’. Thus, motor cortical spikes are read as ‘words’ or ‘sequences’. Future research will be needed to closely examine the anatomical connectivity between single motor cortical neurons and facial motor neurons as well as the connectivity of these to muscle fibres, to suggest a mechanism for the observed effects.

6. Synopsis

(a). Effects of single neuron activity

The recent development of single neuron stimulation techniques revealed single cell effects on perception [30,31,54], movement [66,67] and even brain states [89], in contrast to the earlier belief that single neurons are ‘meaningless’ [110]. Single cell effects are typically small [89] and depend on the identity of the stimulated neuron. Interestingly, the number of cells in the stimulated brain region does not determine effect size: while the effect of single cells was reported in the rat's barrel cortex during a detection task [31,54], this was not the case for stimulating the VPM, which is smaller in size and contains orders of magnitude fewer neurons [30]. Furthermore, it seems that in some cases, such as in the motor system, the complexity of the neuronal coding and resulting stimulation effects increases from the periphery to the cortex. Specifically, stimulating single neurons of the lateral facial nucleus, responsible for whisker movement, results in a spike-by-spike coding, where each spike elicits a forward single whisker deflection [66]. In contrast, stimulating single pyramidal neurons in the rat's primary motor cortex evokes long and complex backward deflections in multiple whiskers [67]. Surprisingly, in the sensory system, while stimulation of periphery nerves rely on rate coding [29], report of single cortical neuron stimulation is highly sensitive to the temporal pattern of the evoked spikes [31]. Further work is needed to explain these contrasting results [111].

(b). Comparison with microstimulation and optogenetic stimulation effects

(i). Sensory effects

While microstimulation allows the causal relationships between neuronal activity and sensation to be established, it is limited by its blindness to the subtype of stimulated neuron, resulting in a mixture of excitation and inhibition [91]. The recent development of optogenetics, a method that can make neurons photosensitive [112], allows the selective activation of neuronal subpopulations and control of them parametrically by inducing a variety of optical stimulation patterns. Using this methodology, a recent study examined the involvement of superficial pyramidal cortical neurons in the barrel cortex of mice during a psychophysical detection task [26]. The authors found that mice could detect activity in a small population of about 60 neurons and the behavioural performance was positively correlated to photostimulation frequency and resulting number of evoked APs. Similarly, Histed and Maunsell [104] applied photostimulation in order to investigate the contribution of pulse duration and intensity to the detectability of a direct optical stimulation of excitatory neurons in the mouse V1 and found that psychophysical performance was well predicted by the total number of spikes, induced by increased amounts of light. Similar results were obtained at the somatosensory cortex in a study which examined the effects of optogenetic stimulation of layer 4 (L4) excitatory neurons on report of illusory touch [105]. The researchers trained mice to report the position of a pole using their whiskers and then manipulated the sensation by photostimulating L4 activity. They concluded that mice used spike count rather than spike timing to reliably determine the position of the pole.

(ii). Motor effects

Electrical microstimulation is an outstanding approach for mapping motor circuits [3,113]. Recent studies in rodents and primates attempted to derive the optimal stimulation frequencies needed to evoke forelimb movements [114,115]. Interestingly, application of long electrical or optogenetic stimulation currents of high intensities evokes complex movements in primates, cats and rodents [10,116–121]. Moreover, recent studies compared the effects of optogenetic stimulation of motor and sensory cortices pyramidal neurons with those of ICMS and surprisingly found that both methods produced similar sensations and movements, suggesting that they might influence the neural activity of the circuit in a similar fashion [122,123]. The comparison of the motor effect of single-cell stimulation and microstimulation in the rat motor cortex led to quite unexpected results. Movement fields are larger and movement durations are longer with single-cell stimulation than with microstimulation; it has been suggested that this surprising pattern of results might be related to the recruitment of inhibition by microstimulation [60,67].

(c). Unresolved issues

Despite the progress in single-cell stimulation, basic issues remain unresolved. Thus, we still do not understand how single cortical neurons can exert behavioural effects even when they are embedded in large neuronal networks. Similarly, we do not know why some temporal patterns are more effective than others. It is also unclear how the brain can operate so robustly, when even the smallest perturbation—a single spike—leads to a cascade of activity, as shown in rodents [90] and in human brain tissue [96]. Finally, the functional implications of single neuron effects remain unknown. However, the different behavioural effects from stimulating single pyramidal cells versus interneurons highlights the need to combine both electrical and optical methods in order to stimulate specific elements of the neural circuit. Novel methods of all-optical imaging and stimulation of single or multiple neurons simultaneously enable the recording and manipulation of confined neuronal circuits with unprecedented accuracy [124,125] and promise to provide answers to some of these questions.

Competing interests

We declare we have no competing interests.

Funding

We received no funding for this study.