Prefrontal Neurons Encode a Solution to the Credit-Assignment Problem.

To adapt successfully to our environments, we must use the outcomes of our choices to guide future behavior. Critically, we must be able to correctly assign credit for any particular outcome to the causal features which preceded it. In some cases, the causal features may be immediately evident, whereas in others they may be separated in time or intermingled with irrelevant environmental stimuli, creating a potentially nontrivial credit-assignment problem. We examined the neuronal representation of information relevant for credit assignment in the dorsolateral prefrontal cortex (dlPFC) of two male rhesus macaques performing a task that elicited key aspects of this problem. We found that neurons conveyed the information necessary for credit assignment. Specifically, neuronal activity reflected both the relevant cues and outcomes at the time of feedback and did so in a manner that was stable over time, in contrast to prior reports of representational instability in the dlPFC. Furthermore, these representations were most stable early in learning, when credit assignment was most needed. When the same features were not needed for credit assignment, these neuronal representations were much weaker or absent. These results demonstrate that the activity of dlPFC neurons conforms to the basic requirements of a system that performs credit assignment, and that spiking activity can serve as a stable mechanism that links causes and effects.SIGNIFICANCE STATEMENT Credit assignment is the process by which we infer the causes of our successes and failures. We found that neuronal activity in the dorsolateral prefrontal cortex conveyed the necessary information for performing credit assignment. Importantly, while there are various potential mechanisms to retain a "trace" of the causal events over time, we observed that spiking activity was sufficiently stable to act as the link between causes and effects, in contrast to prior reports that suggested spiking representations were unstable over time. In addition, we observed that this stability varied as a function of learning, such that the neural code was more reliable over time during early learning, when it was most needed.

Clinical translation of a high-performance neural prosthesis.

Neural prostheses have the potential to improve the quality of life of individuals with paralysis by directly mapping neural activity to limb- and computer-control signals. We translated a neural prosthetic system previously developed in animal model studies for use by two individuals with amyotrophic lateral sclerosis who had intracortical microelectrode arrays placed in motor cortex. Measured more than 1 year after implant, the neural cursor-control system showed the highest published performance achieved by a person to date, more than double that of previous pilot clinical trial participants.

Reliability of directional information in unsorted spikes and local field potentials recorded in human motor cortex.

OBJECTIVE: Action potentials and local field potentials (LFPs) recorded in primary motor cortex contain information about the direction of movement. LFPs are assumed to be more robust to signal instabilities than action potentials, which makes LFPs, along with action potentials, a promising signal source for brain-computer interface applications. Still, relatively little research has directly compared the utility of LFPs to action potentials in decoding movement direction in human motor cortex. APPROACH: We conducted intracortical multi-electrode recordings in motor cortex of two persons (T2 and [S3]) as they performed a motor imagery task. We then compared the offline decoding performance of LFPs and spiking extracted from the same data recorded across a one-year period in each participant. MAIN RESULTS: We obtained offline prediction accuracy of movement direction and endpoint velocity in multiple LFP bands, with the best performance in the highest (200-400 Hz) LFP frequency band, presumably also containing low-pass filtered action potentials. Cross-frequency correlations of preferred directions and directional modulation index showed high similarity of directional information between action potential firing rates (spiking) and high frequency LFPs (70-400 Hz), and increasing disparity with lower frequency bands (0-7, 10-40 and 50-65 Hz). Spikes predicted the direction of intended movement more accurately than any individual LFP band, however combined decoding of all LFPs was statistically indistinguishable from spike-based performance. As the quality of spiking signals (i.e. signal amplitude) and the number of significantly modulated spiking units decreased, the offline decoding performance decreased 3.6[5.65]%/month (for T2 and [S3] respectively). The decrease in the number of significantly modulated LFP signals and their decoding accuracy followed a similar trend (2.4[2.85]%/month, ANCOVA, p = 0.27[0.03]). SIGNIFICANCE: Field potentials provided comparable offline decoding performance to unsorted spikes. Thus, LFPs may provide useful external device control using current human intracortical recording technology. ( CLINICAL TRIAL REGISTRATION NUMBER: NCT00912041.).

Adaptive offset correction for intracortical brain-computer interfaces.

Intracortical brain-computer interfaces (iBCIs) decode intended movement from neural activity for the control of external devices such as a robotic arm. Standard approaches include a calibration phase to estimate decoding parameters. During iBCI operation, the statistical properties of the neural activity can depart from those observed during calibration, sometimes hindering a user's ability to control the iBCI. To address this problem, we adaptively correct the offset terms within a Kalman filter decoder via penalized maximum likelihood estimation. The approach can handle rapid shifts in neural signal behavior (on the order of seconds) and requires no knowledge of the intended movement. The algorithm, called multiple offset correction algorithm (MOCA), was tested using simulated neural activity and evaluated retrospectively using data collected from two people with tetraplegia operating an iBCI. In 19 clinical research test cases, where a nonadaptive Kalman filter yielded relatively high decoding errors, MOCA significantly reduced these errors ( 10.6 ± 10.1% ; p < 0.05, pairwise t-test). MOCA did not significantly change the error in the remaining 23 cases where a nonadaptive Kalman filter already performed well. These results suggest that MOCA provides more robust decoding than the standard Kalman filter for iBCIs.

Intra-day signal instabilities affect decoding performance in an intracortical neural interface system.

OBJECTIVE: Motor neural interface systems (NIS) aim to convert neural signals into motor prosthetic or assistive device control, allowing people with paralysis to regain movement or control over their immediate environment. Effector or prosthetic control can degrade if the relationship between recorded neural signals and intended motor behavior changes. Therefore, characterizing both biological and technological sources of signal variability is important for a reliable NIS. APPROACH: To address the frequency and causes of neural signal variability in a spike-based NIS, we analyzed within-day fluctuations in spiking activity and action potential amplitude recorded with silicon microelectrode arrays implanted in the motor cortex of three people with tetraplegia (BrainGate pilot clinical trial, IDE). MAIN RESULTS: 84% of the recorded units showed a statistically significant change in apparent firing rate (3.8 ± 8.71 Hz or 49% of the mean rate) across several-minute epochs of tasks performed on a single session, and 74% of the units showed a significant change in spike amplitude (3.7 ± 6.5 µV or 5.5% of mean spike amplitude). 40% of the recording sessions showed a significant correlation in the occurrence of amplitude changes across electrodes, suggesting array micro-movement. Despite the relatively frequent amplitude changes, only 15% of the observed within-day rate changes originated from recording artifacts such as spike amplitude change or electrical noise, while 85% of the rate changes most likely emerged from physiological mechanisms. Computer simulations confirmed that systematic rate changes of individual neurons could produce a directional 'bias' in the decoded neural cursor movements. Instability in apparent neuronal spike rates indeed yielded a directional bias in 56% of all performance assessments in participant cursor control (n = 2 participants, 108 and 20 assessments over two years), resulting in suboptimal performance in these sessions. SIGNIFICANCE: We anticipate that signal acquisition and decoding methods that can adapt to the reported instabilities will further improve the performance of intracortically-based NISs.

Why do axons differ in caliber?

CNS axons differ in diameter (d) by nearly 100-fold (∼0.1-10 µm); therefore, they differ in cross-sectional area (d(2)) and volume by nearly 10,000-fold. If, as found for optic nerve, mitochondrial volume fraction is constant with axon diameter, energy capacity would rise with axon volume, also as d(2). We asked, given constraints on space and energy, what functional requirements set an axon's diameter? Surveying 16 fiber groups spanning nearly the full range of diameters in five species (guinea pig, rat, monkey, locust, octopus), we found the following: (1) thin axons are most numerous; (2) mean firing frequencies, estimated for nine of the identified axon classes, are low for thin fibers and high for thick ones, ranging from ∼1 to >100 Hz; (3) a tract's distribution of fiber diameters, whether narrow or broad, and whether symmetric or skewed, reflects heterogeneity of information rates conveyed by its individual fibers; and (4) mitochondrial volume/axon length rises ≥d(2). To explain the pressure toward thin diameters, we note an established law of diminishing returns: an axon, to double its information rate, must more than double its firing rate. Since diameter is apparently linear with firing rate, doubling information rate would more than quadruple an axon's volume and energy use. Thicker axons may be needed to encode features that cannot be efficiently decoded if their information is spread over several low-rate channels. Thus, information rate may be the main variable that sets axon caliber, with axons constrained to deliver information at the lowest acceptable rate.

How the optic nerve allocates space, energy capacity, and information.

Fiber tracts should use space and energy efficiently, because both resources constrain neural computation. We found for a myelinated tract (optic nerve) that astrocytes use nearly 30% of the space and >70% of the mitochondria, establishing the significance of astrocytes for the brain's space and energy budgets. Axons are mostly thin with a skewed distribution peaking at 0.7 microm, near the lower limit set by channel noise. This distribution is matched closely by the distribution of mean firing rates measured under naturalistic conditions, suggesting that firing rate increases proportionally with axon diameter. In axons thicker than 0.7 microm, mitochondria occupy a constant fraction of axonal volume--thus, mitochondrial volumes rise as the diameter squared. These results imply a law of diminishing returns: twice the information rate requires more than twice the space and energy capacity. We conclude that the optic nerve conserves space and energy by sending most information at low rates over fine axons with small terminal arbors and sending some information at higher rates over thicker axons with larger terminal arbors but only where more bits per second are needed for a specific purpose. Thicker axons seem to be needed, not for their greater conduction velocity (nor other intrinsic electrophysiological purpose), but instead to support larger terminal arbors and more active zones that transfer information synaptically at higher rates.

Dynamics of directional selectivity in MT receptive field centre and surround.

We studied receptive field organization of motion-sensitive neurons in macaque middle temporal cortical area (MT), by mapping direction selectivity in space and in time. Stimuli consisted of pseudorandom sequences of single motion steps presented simultaneously at many different receptive field locations. Spatio-temporal receptive field profiles were constructed by cross-correlating stimuli and spikes. The resulting spike-triggered averages revealed centre-surround organization. The temporal dynamics of the receptive fields were generally biphasic with increased probability for the preferred direction at short latency (50-70 ms) and decreased probability at longer latency (80-100 ms). The response latency of the receptive field surround was on average 16 ms longer than that of the centre. Our results show that surround input and biphasic behaviour reflect two different mechanisms, which make MT cells specifically sensitive to motion contrast in space and time.

Temporal dynamics of direction tuning in motion-sensitive macaque area MT.

We studied the temporal dynamics of motion direction sensitivity in macaque area MT using a motion reverse correlation paradigm. Stimuli consisted of a random sequence of motion steps in eight different directions. Cross-correlating the stimulus with the resulting neural activity reveals the temporal dynamics of direction selectivity. The temporal dynamics of direction selectivity at the preferred speed showed two phases along the time axis: one phase corresponding to an increase in probability for the preferred direction at short latencies and a second phase corresponding to a decrease in probability for the preferred direction at longer latencies. The strength of this biphasic behavior varied between neurons from weak to very strong and was uniformly distributed. Strong biphasic behavior suggests optimal responses for motion steps in the antipreferred direction followed by a motion step in the preferred direction. Correlating spikes to combinations of motion directions corroborates this distinction. The optimal combination for weakly biphasic cells consists of successive steps in the preferred direction, whereas for strongly biphasic cells, it is a reversal of directions. Comparing reverse correlograms to combinations of stimuli to predictions based on correlograms for individual directions revealed several nonlinear effects. Correlations for successive presentations of preferred directions were smaller than predicted, which could be explained by a static nonlinearity (saturation). Correlations to pairs of (nearly) opposite directions were larger than predicted. These results show that MT neurons are generally more responsive when sudden changes in motion directions occur, irrespective of the preferred direction of the neurons. The latter nonlinearities cannot be explained by a simple static nonlinearity at the output of the neuron, but most likely reflect network interactions.

The motion reverse correlation (MRC) method: a linear systems approach in the motion domain.

We introduce the motion reverse correlation method (MRC), a novel stimulus paradigm based on a random sequence of motion impulses. The method is tailored to investigate the spatio-temporal dynamics of motion selectivity in cells responding to moving random dot patterns. Effectiveness of the MRC method is illustrated with results obtained from recordings in both anesthetized cats and an awake, fixating macaque monkey. Motion tuning functions are computed by reverse correlating the response of single cells with a rapid sequence of displacements of a random pixel array (RPA). Significant correlations between the cell's responses and various aspects of stimulus motion are obtained at high temporal resolution. These correlations provide a detailed description of the temporal dynamics of, for example, direction tuning and velocity tuning. In addition, with a spatial array of independently moving RPAs, the MRC method can be used to measure spatial as well as temporal receptive field properties. We demonstrate that MRC serves as a powerful and time-efficient tool for quantifying receptive field properties of motion selective cells that yields temporal information that cannot be derived from existing methods.