Scalable Gaussian process inference of neural responses to natural images.

Predicting the responses of sensory neurons is a long-standing neuroscience goal. However, while there has been much progress in modeling neural responses to simple and/or artificial stimuli, predicting responses to natural stimuli remains an ongoing challenge. On the one hand, deep neural networks perform very well on certain datasets but can fail when data are limited. On the other hand, Gaussian processes (GPs) perform well on limited data but are poor at predicting responses to high-dimensional stimuli, such as natural images. Here, we show how structured priors, e.g., for local and smooth receptive fields, can be used to scale up GPs to model neural responses to high-dimensional stimuli. With this addition, GPs largely outperform a deep neural network trained to predict retinal responses to natural images, with the largest differences observed when both models are trained on a small dataset. Further, since they allow us to quantify the uncertainty in their predictions, GPs are well suited to closed-loop experiments, where stimuli are chosen actively so as to collect "informative" neural data. We show how GPs can be used to actively select which stimuli to present, so as to i) efficiently learn a model of retinal responses to natural images, using few data, and ii) rapidly distinguish between competing models (e.g., a linear vs. a nonlinear model). In the future, our approach could be applied to other sensory areas, beyond the retina.

Context-dependent selectivity to natural images in the retina.

Retina ganglion cells extract specific features from natural scenes and send this information to the brain. In particular, they respond to local light increase (ON responses), and/or decrease (OFF). However, it is unclear if this ON-OFF selectivity, characterized with synthetic stimuli, is maintained under natural scene stimulation. Here we recorded ganglion cell responses to natural images slightly perturbed by random noise patterns to determine their selectivity during natural stimulation. The ON-OFF selectivity strongly depended on the specific image. A single ganglion cell can signal luminance increase for one image, and luminance decrease for another. Modeling and experiments showed that this resulted from the non-linear combination of different retinal pathways. Despite the versatility of the ON-OFF selectivity, a systematic analysis demonstrated that contrast was reliably encoded in these responses. Our perturbative approach uncovered the selectivity of retinal ganglion cells to more complex features than initially thought.

An elaborate sweep-stick code in rat barrel cortex.

In rat barrel cortex, feature encoding schemes uncovered during broadband whisker stimulation are hard to reconcile with the simple stick-slip code observed during natural tactile behaviors, and this has hindered the development of a generalized computational framework. By designing broadband artificial stimuli to sample the inputs encoded under natural conditions, we resolve this disparity while markedly increasing the percentage of deep layer neurons found to encode whisker movements, as well as the diversity of these encoded features. Deep layer neurons encode two main types of events, sticks and sweeps, corresponding to high angular velocity bumps and large angular displacements with high velocity, respectively. Neurons can exclusively encode sticks or sweeps, or they can encode both, with or without direction selectivity. Beyond unifying coding theories from naturalistic and artificial stimulation studies, these findings delineate a simple and generalizable set of whisker movement features that can support a range of perceptual processes.

Rich spatio-temporal stimulus dynamics unveil sensory specialization in cortical area S2.

Tactile perception in rodents depends on simultaneous, multi-whisker contacts with objects. Although it is known that neurons in secondary somatosensory cortex (wS2) respond to individual deflections of many whiskers, wS2's precise function remains unknown. The convergence of information from multiple whiskers into wS2 neurons suggests that they are good candidates for integrating multi-whisker information. Here, we apply stimulation patterns with rich dynamics simultaneously to 24 macro-vibrissae of rats while recording large populations of single neurons. Varying inter-whisker correlations without changing single whisker statistics, we observe pronounced supra-linear multi-whisker integration. Using novel analysis methods, we show that continuous multi-whisker movements contribute to the firing of wS2 neurons over long temporal windows, facilitating spatio-temporal integration. In contrast, primary cortex (wS1) neurons encode fine features of whisker movements on precise temporal scales. These results provide the first description of wS2's representation during multi-whisker stimulation and outline its specialized role in parallel to wS1 tactile processing.

Temperature manipulation of neuronal dynamics in a forebrain motor control nucleus.

Different neuronal types within brain motor areas contribute to the generation of complex motor behaviors. A widely studied songbird forebrain nucleus (HVC) has been recognized as fundamental in shaping the precise timing characteristics of birdsong. This is based, among other evidence, on the stretching and the "breaking" of song structure when HVC is cooled. However, little is known about the temperature effects that take place in its neurons. To address this, we investigated the dynamics of HVC both experimentally and computationally. We developed a technique where simultaneous electrophysiological recordings were performed during temperature manipulation of HVC. We recorded spontaneous activity and found three effects: widening of the spike shape, decrease of the firing rate and change in the interspike interval distribution. All these effects could be explained with a detailed conductance based model of all the neurons present in HVC. Temperature dependence of the ionic channel time constants explained the first effect, while the second was based in the changes of the maximal conductance using single synaptic excitatory inputs. The last phenomenon, only emerged after introducing a more realistic synaptic input to the inhibitory interneurons. Two timescales were present in the interspike distributions. The behavior of one timescale was reproduced with different input balances received form the excitatory neurons, whereas the other, which disappears with cooling, could not be found assuming poissonian synaptic inputs. Furthermore, the computational model shows that the bursting of the excitatory neurons arises naturally at normal brain temperature and that they have an intrinsic delay at low temperatures. The same effect occurs at single synapses, which may explain song stretching. These findings shed light on the temperature dependence of neuronal dynamics and present a comprehensive framework to study neuronal connectivity. This study, which is based on intrinsic neuronal characteristics, may help to understand emergent behavioral changes.

Utilization of extracellular information before ligand-receptor binding reaches equilibrium expands and shifts the input dynamic range.

Cell signaling systems sense and respond to ligands that bind cell surface receptors. These systems often respond to changes in the concentration of extracellular ligand more rapidly than the ligand equilibrates with its receptor. We demonstrate, by modeling and experiment, a general "systems level" mechanism cells use to take advantage of the information present in the early signal, before receptor binding reaches a new steady state. This mechanism, pre-equilibrium sensing and signaling (PRESS), operates in signaling systems in which the kinetics of ligand-receptor binding are slower than the downstream signaling steps, and it typically involves transient activation of a downstream step. In the systems where it operates, PRESS expands and shifts the input dynamic range, allowing cells to make different responses to ligand concentrations so high as to be otherwise indistinguishable. Specifically, we show that PRESS applies to the yeast directional polarization in response to pheromone gradients. Consideration of preexisting kinetic data for ligand-receptor interactions suggests that PRESS operates in many cell signaling systems throughout biology. The same mechanism may also operate at other levels in signaling systems in which a slow activation step couples to a faster downstream step.

Temperature induced syllable breaking unveils nonlinearly interacting timescales in birdsong motor pathway.

The nature of telencephalic control over premotor and motor circuits is debated. Hypotheses range from complete usurping of downstream circuitry to highly interactive mechanisms of control. We show theoretically and experimentally, that telencephalic song motor control in canaries is consistent with a highly interactive strategy. As predicted from a theoretical model of respiratory control, mild cooling of a forebrain nucleus (HVC) led to song stretching, but further cooling caused progressive restructuring of song, consistent with the hypothesis that respiratory gestures are subharmonic responses to a timescale present in the output of HVC. This interaction between a life-sustaining motor function (respiration) and telencephalic song motor control suggests a more general mechanism of how nonlinear integration of evolutionarily new brain structures into existing circuitry gives rise to diverse, new behavior.

Evidence and control of bifurcations in a respiratory system.

We studied the pressure patterns used by domestic canaries in the production of birdsong. Acoustically different sound elements ("syllables") were generated by qualitatively different pressure gestures. We found that some ubiquitous transitions between syllables can be interpreted as bifurcations of a low dimensional dynamical system. We interpreted these results as evidence supporting a model in which different timescales interact nonlinearly.