Natural language processing models reveal neural dynamics of human conversation.

Human verbal communication requires a rapid interplay between speech planning, production, and comprehension. These processes are subserved by local and long-range neural dynamics across widely distributed brain areas. How linguistic information is precisely represented during natural conversation or what shared neural processes are involved, however, remain largely unknown. Here we used intracranial neural recordings in participants engaged in free dialogue and employed deep learning natural language processing models to find a striking similarity not only between neural-to-artificial network activities but also between how linguistic information is encoded in brain during production and comprehension. Collectively, neural activity patterns that encoded linguistic information were closely aligned to those reflecting speaker-listener transitions and were reduced after word utterance or when no conversation was held. They were also observed across distinct mesoscopic areas and frequency bands during production and comprehension, suggesting that these signals reflected the hierarchically structured information being conveyed during dialogue. Together, these findings suggest that linguistic information is encoded in the brain through similar neural representations during both speaking and listening, and start to reveal the distributed neural dynamics subserving human communication.

Direct Discriminative Decoder Models for Analysis of High-Dimensional Dynamical Neural Data.

With the accelerated development of neural recording technology over the past few decades, research in integrative neuroscience has become increasingly reliant on data analysis methods that are scalable to high-dimensional recordings and computationally tractable. Latent process models have shown promising results in estimating the dynamics of cognitive processes using individual models for each neuron's receptive field. However, scaling these models to work on high-dimensional neural recordings remains challenging. Not only is it impractical to build receptive field models for individual neurons of a large neural population, but most neural data analyses based on individual receptive field models discard the local history of neural activity, which has been shown to be critical in the accurate inference of the underlying cognitive processes. Here, we propose a novel, scalable latent process model that can directly estimate cognitive process dynamics without requiring precise receptive field models of individual neurons or brain nodes. We call this the direct discriminative decoder (DDD) model. The DDD model consists of (1) a discriminative process that characterizes the conditional distribution of the signal to be estimated, or state, as a function of both the current neural activity and its local history, and (2) a state transition model that characterizes the evolution of the state over a longer time period. While this modeling framework inherits advantages of existing latent process modeling methods, its computational cost is tractable. More important, the solution can incorporate any information from the history of neural activity at any timescale in computing the estimate of the state process. There are many choices in building the discriminative process, including deep neural networks or gaussian processes, which adds to the flexibility of the framework. We argue that these attributes of the proposed methodology, along with its applicability to different modalities of neural data, make it a powerful tool for high-dimensional neural data analysis. We also introduce an extension of these methods, called the discriminative-generative decoder (DGD). The DGD includes both discriminative and generative processes in characterizing observed data. As a result, we can combine physiological correlates like behavior with neural data to better estimate underlying cognitive processes. We illustrate the methods, including steps for inference and model identification, and demonstrate applications to multiple data analysis problems with high-dimensional neural recordings. The modeling results demonstrate the computational and modeling advantages of the DDD and DGD methods.

Differential Responses to High-Frequency Electrical Stimulation in Brisk-Transient and Delta Retinal Ganglion Cells.

Retinal microprostheses strive to evoke a sense of vision in individuals blinded by outer retinal degenerative diseases, by electrically stimulating the surviving retina. It is widely suspected that a stimulation strategy that can selectively activate different retinal ganglion cell types will improve the quality of evoked phosphenes. Previous efforts towards this goal demonstrated the potential for selective ON and OFF brisk-transient cell activation using high-rate (2000 pulses per second, PPS) stimulation. Here, we build upon this earlier work by testing an additional rate of stimulation and additional cell populations. We find considerable variability in responses both within and across individual cell types, but show that the sensitivity of a ganglion cell to repetitive stimulation is highly correlated to its single-pulse threshold. Consistent with this, we found thresholds for both stimuli to be correlated to soma size, and thus likely mediated by the properties of the axon initial segment. The ultimate efficacy of high-rate stimulation will likely depend on several factors, chief among which are (a) the residual ganglion types, and (b) the stimulation frequency.

Response Profiles of Retinal Ganglion Cells to Sinusoidal Electric Stimulation vary for Low vs. High Frequencies.

Microelectronic retinal prostheses electrically stimulate retinal neurons with the goal of restoring vision in patients blinded by outer retinal degeneration. Despite some success in clinical trials, the quality of vision elicited by these devices is still limited. To improve the performance of retinal prostheses, our group studied how retinal neurons respond to electric stimulation. Our previous work showed that responses of retinal ganglion cells (RGCs) are frequency-dependent and different types of RGCs can be preferentially activated with a specific frequency and current amplitude. In the present study, we systemically examined responses of RGCs to sinusoidal electric stimulation with varying frequencies and amplitudes. We found that ON sustained alpha RGCs show distinct stimulus-response relationships to low and high frequency stimulation. For example, RGCs showed monotonic response curves to 500 Hz sinusoidal stimulation, whereas they showed non-monotonic response curves to 2000 Hz stimulation. We also described how increasing stimulus frequency gradually changed the response curves of RGCs.

Visual and electric spiking responses of seven types of rabbit retinal ganglion cells.

Electric stimulation of the retina via retinal implants is currently the only commercially available method to restore vision in patients suffering from a wide range of outer retinal degenerations. To improve the quality of retinal implants, it is desirable to better understand how different retinal cell classes and types respond to electric stimuli so that more effective stimulation strategies can be developed. Here, we measured the response of seven major types of retinal ganglion cells to electric stimulation. A simple series of light stimuli were used to classify cells into known types. Electric stimulation produced unique responses in almost all ganglion cell types and the electric responses typically matched elements of the corresponding light responses.

Frequency Responses of Rat Retinal Ganglion Cells.

There are 15-20 different types of retinal ganglion cells (RGC) in the mammalian retina, each encoding different aspects of the visual scene. The mechanism by which post-synaptic signals from the retinal network generate spikes is determined by each cell's intrinsic electrical properties. Here we investigate the frequency responses of morphologically identified rat RGCs using intracellular injection of sinusoidal current waveforms, to assess their intrinsic capabilities with minimal contributions from the retinal network. Recorded cells were classified according to their morphological characteristics (A, B, C or D-type) and their stratification (inner (i), outer (o) or bistratified) in the inner plexiform layer (IPL). Most cell types had low- or band-pass frequency responses. A2, C1 and C4o cells were band-pass with peaks of 15-30 Hz and low-pass cutoffs above 56 Hz (A2 cells) and ~42 Hz (C1 and C4o cells). A1 and C2i/o cells were low-pass with peaks of 10-15 Hz (cutoffs 19-25 Hz). Bistratified D1 and D2 cells were also low-pass with peaks of 5-10 Hz (cutoffs ~16 Hz). The least responsive cells were the B2 and C3 types (peaks: 2-5 Hz, cutoffs: 8-11 Hz). We found no difference between cells stratifying in the inner and outer IPL (i.e., ON and OFF cells) or between cells with large and small somas or dendritic fields. Intrinsic physiological properties (input resistance, spike width and sag) had little impact on frequency response at low frequencies, but account for 30-40% of response variability at frequencies >30 Hz.

A Simple and Accurate Model to Predict Responses to Multi-electrode Stimulation in the Retina.

Implantable electrode arrays are widely used in therapeutic stimulation of the nervous system (e.g. cochlear, retinal, and cortical implants). Currently, most neural prostheses use serial stimulation (i.e. one electrode at a time) despite this severely limiting the repertoire of stimuli that can be applied. Methods to reliably predict the outcome of multi-electrode stimulation have not been available. Here, we demonstrate that a linear-nonlinear model accurately predicts neural responses to arbitrary patterns of stimulation using in vitro recordings from single retinal ganglion cells (RGCs) stimulated with a subretinal multi-electrode array. In the model, the stimulus is projected onto a low-dimensional subspace and then undergoes a nonlinear transformation to produce an estimate of spiking probability. The low-dimensional subspace is estimated using principal components analysis, which gives the neuron's electrical receptive field (ERF), i.e. the electrodes to which the neuron is most sensitive. Our model suggests that stimulation proportional to the ERF yields a higher efficacy given a fixed amount of power when compared to equal amplitude stimulation on up to three electrodes. We find that the model captures the responses of all the cells recorded in the study, suggesting that it will generalize to most cell types in the retina. The model is computationally efficient to evaluate and, therefore, appropriate for future real-time applications including stimulation strategies that make use of recorded neural activity to improve the stimulation strategy.

Prosthetic vision: devices, patient outcomes and retinal research.

Retinal disease and its associated retinal degeneration can lead to the loss of photoreceptors and therefore, profound blindness. While retinal degeneration destroys the photoreceptors, the neural circuits that convey information from the eye to the brain are sufficiently preserved to make it possible to restore sight using prosthetic devices. Typically, these devices consist of a digital camera and an implantable neurostimulator. The image sensor in a digital camera has the same spatiotopic arrangement as the photoreceptors of the retina. Therefore, it is possible to extract meaningful spatial information from an image and deliver it via an array of stimulating electrodes directly to the surviving retinal circuits. Here, we review the structure and function of normal and degenerate retina. The different approaches to prosthetic implant design are described in the context of human and preclinical trials. In the last section, we review studies of electrical properties of the retina and its response to electrical stimulation. These types of investigation are currently assessing a number of key challenges identified in human trials, including stimulation efficacy, spatial localisation, desensitisation to repetitive stimulation and selective activation of retinal cell populations.

Retinal ganglion cells electrophysiology: the effect of cell morphology on impulse waveform.

There are 16 morphologically defined classes of rats retinal ganglion cells (RGCs). Using computer simulation of a realistic anatomically correct A1 mouse RGC, we investigate the effect of the cell's morphology on its impulse waveform, using the first-, and second-order time derivatives as well as the phase plot features. Using whole cell patch clamp recordings, we recorded the impulse waveform for each of the rat RGCs types. While we found some clear differences in many features of the impulse waveforms for A2 and B2 cells compared to other cell classes, many cell types did not show clear differences.

Electrical stimulation of retinal ganglion cells with diamond and the development of an all diamond retinal prosthesis.

Electronic retinal implants for the blind are already a market reality. A world wide effort is underway to find the technology that offers the best combination of performance and safety for potential patients. Our approach is to construct an epi-retinally targeted device entirely encapsulated in diamond to maximise longevity and biocompatibility. The stimulating array of our device comprises a monolith of electrically insulating diamond with thousands of hermetic, microscale nitrogen doped ultra-nanocrystalline diamond (N-UNCD) feedthroughs. Here we seek to establish whether the conducting diamond feedthroughs of the array can be used as stimulating electrodes without further modification with a more traditional neural stimulation material. Efficacious stimulation of retinal ganglion cells was established using single N-UNCD microelectrodes in contact with perfused, explanted, rat retina. Evoked rat retinal ganglion cell action potentials were recorded by patch clamp recording from single ganglion cells, adjacent to the N-UNCD stimulating electrode. Separately, excellent electrochemical stability of N-UNCD was established by prolonged pulsing in phosphate buffered saline at increasing charge density up to the measured charge injection limit for the material.

Epiretinal electrical stimulation and the inner limiting membrane in rat retina.

In this paper we aim to quantify the effect of the inner limiting membrane (ILM) of the retina on the thresholds for epiretinal electrical stimulation of retinal ganglion cells by a microelectronic retinal prosthesis. A pair of bipolar stimulating electrodes was placed either above (on the epiretinal surface) or below the ILM while we made whole-cell patch-clamp recordings from retinal ganglion cells in an isolated rat retinal whole-mount preparation. Across our cell population we found no significant difference in the median threshold stimulus amplitudes when the stimulating electrodes were placed below as opposed to above the ILM (p = 0.08). However, threshold stimulus amplitudes did tend to be lower when the stimulating electrodes were placed below the ILM (30 µA vs 56 µA).