Recent experiments have shown that a period of intra-cortical microstimulation (ICMS) with a weak electric current causes reorganization of neuron assemblies in the auditory (A1) cortex of the white rat ( Exp Brain Res (1996) 112:420-30). We speculate that such sensory map reorganization might affect the perception of the rat in an appropriate behavioral paradigm.
Auditory frequency discrimination ability was systematically investigated in the albino rat and its sensory resolving power indexed with the aid of operant conditioning procedures and receiver operating characteristic (ROC) analysis. Classical thresholds were measured by the method of constant stimuli. Fine wire electrodes were then implanted into the A1 cortex and the receptive fields of cells defined in the anesthetized and awake subject. The characteristic or best frequency of cells surrounding a particular electrode site was noted and the frequency discrimination performance of the awake animal was then analyzed around that best frequency in several test sessions. Before some of the sessions the cells in the electrode vicinity were stimulated with repeated short trains of 5 microamp pulses to assess its effect on performance.
The albino rat is capable of making finer frequency discriminations than has been previously reported. Under appropriate conditions classical weber fractions of around 2-3% are obtained. Behavioral data obtained in the rat are consistent with the signal detection theory model and provide an index uncorrupted by response bias. In the awake animal receptive fields of A1 cells are significantly different from those obtained in the anesthetized subject. So far (N=5) our study shows that ICMS induced plasticity has no statistically significant effect on a white rat's ability to discriminate frequencies.
A rat's reach for a food pellet is a complex and highly demanding task that has been extensively studied (e.g., Whishaw & Pellis, Behav. Brain Res. 1990). We are engaged in using this behavior to examine the relationship between the activity of assemblies of neurons in sensorimotor cortex and motor performance.
We used a capacitive proximity switch to detect the approach of the rat's paw and trigger video image capture. Rats reached through an 8 mm wide opening in a wire cage to grasp and retrieve 40 mg food pellets off the top of the proximity switch.
Both skilled (n=2) and neophyte (n=2) rats extended their digits during the approach. Neophyte rats swiped at the pellet, closing the digits only on contact. Skilled rats snapped their digits closed when attempting to grasp the pellet, often with the snap starting before contact was made. No preshaping of the digits was seen for either group. The skilled rats developed a "grasp cycle" approach. Reach attempts having as many as 8 cycles were seen. A cycle consisted of the rat extending its digits while approaching the target, snapping its digits closed at or near the target, and withdrawing its paw (but not all the way back into the cage). Typical cycles had proximity times of 40-80 ms (paw within ~8 mm of the pellet), and off times of 50-200 ms. Successful grasping of the pellet resulted in withdrawal of the paw into the cage, and transport of the pellet to the mouth. Unsuccessful attempts, even those contacting the pellet, were usually followed by another cycle. Often accurately targeted and timed grasps were generated even though the pellet had been knocked off on a previous cycle.
The results suggest two key aspects of this skill: (1) linking digit closure to position, rather than target contact; and (2) using appropriate target contact to stop the grasping cycle, and start retrieval. We are investigating whether such events are reflected in sensorimotor cortex activity.
The basic operating mode of cortical neurons is currently being debated (reviewed by König et al., 1996, TINS, 19: 130-7). Whether they act as "coincidence detectors" or "temporal integrators" has implications for whether cortical neurons encode information by precise timing of action potentials or simply by average firing rate. The present theoretical study demonstrates a broad continuum of basic operating modes and corresponding potential coding-schemes.
We attempt to characterize the class of multi-synaptic input pattern that reliably causes a neuron to fire. As a template for the input patterns we use 'Gaussian events': N excitatory synaptic inputs scattered around a mean input time with a given standard deviation, SD. These events are particularly well-suited to this investigation because a zero-SD event corresponds to perfectly coincident input, a high-SD event must be integrated over several milliseconds, and inputs with intermediate values of SD lie along a spectrum of varying degrees of synchrony.
Computational simulations performed with a simple four-state-variable passive-membrane model neuron show that output spikes are generated reliably (>90%) for a large range of N's and SD's. Within this range the output spike latency with respect to the input event depends upon both N and SD as follows: high N and/or low SD will cause an output spike with a shorter latency than low N and/or high SD. The standard deviation of this latency (the "jitter" of the output spike) depends primarily on, and is proportional to SD. Also, the slope of the membrane potential for the 0.5-1 msec preceding an output spike is found to be exquisitely sensitive to SD, suggesting that it could be used as a predictor for the temporal jitter of both the input pattern and the output spike. This implies that intracellular recordings from single cortical neurons could help resolve the debate regarding the basic operating mode of cortical neurons.
One of our current modeling projects involves simulations of the effect of feedback in neural systems, particularly corticothalamic and corticocortical systems. We are especially interested in the experimental results reported in Sillito et al. (1994), Nature 369:479-82. In this experiment, pairs of projection neurons in the lateral geniculate nucleus (LGN) of the cat were simultaneously recorded; the neurons were typically separated by 1-4 visual degrees. A drifting grating was used to stimulate the cells, with the angle chosen such that the bars were aligned with the approximate centers of the neurons' receptive fields. Thus, the two neurons were co-stimulated by each grating phase. The results were startling: even though the neurons were relatively distant, their spike trains were often highly correlated. Evidently the corticogeniculate projection played a crucial role, since the correlations were absent in preparations with visual cortex removed.
Upon further analysis of the Sillito et al. (1994) data, we found that, in a
significant number of cases, the correlation was mostly due to bursts in the
spike trains of the LGN neurons. Bursts in thalamic cells can arise from
activation of the low-threshold calcium channel (LT); we hypothesized that the
mechanism producing the LGN correlations involve this channel, along with a
dual excitatory-inhibitory corticogeniculate projection. Using simulations with
about 1000 cells, we showed that our model reproduces all of the essential
features of the experimental data, and both of our model's main features --
inclusion of the LT channels and the duality of the corticogeniculate path --
are necessary. The model also provided some interesting predictions on the
state of the cortex during the process. Our simulation results will be
We have recently focused on modeling feedback in corticocortical systems. One possible role of feedback in these systems is providing some sort of "top-down" signals that reflect current states of attention. Neurons in lower-order cortical areas have been demonstrated to be influenced by attentive processes, and we believe that experiments reported in Motter (1994) Journal of Neuroscience 14: 2178-2189 and 2190-99 are particularly interesting. In these experiments, the firing rate of monkey V4 neurons is modulated by the the presence or absence of the monkey's attention (which is in turn motivated by the possibility or impossibility of a reward). The modulations require some time to develop, and we hypothesize that this time is due to the delay arising from feedback from higher order cortical areas, which we believe are responsible for integration and "global" analysis. We have constructed a model using a hierarchy of cortical networks to illustrate that feedback, along with a very simple connectivity scheme, can reproduce the experimental results quite well. We are currently testing the model on some variations in the original stimulus paradigm used in the experiment; we are also further exploring the timing relationship between a given amount of feedback and the subsequent delay in the initiation of the modulation of lower-order area firing rates.
Effective connectivity (synaptic connectivity as measured by cross-correlation) can be rapidly modulated by "background" activity. For example, Boven & Aertsen (1990, Parallel Processing in Neural Systems and Computers, Eckmiller et al. (eds.), Elsevier: 53-56) demonstrated that efficacy of connection between two modeled neurons is a function of the rate of stochastic EPSP-bombardment onto the post-synaptic cell. The objective of the present study is to replicate and then extend the findings of these authors with a conductance-based point-neuron model.
Each neuron contains channels for synaptic connections and action-potential generation. All connections are of equal strength and relatively weak. The results of the simulations (Genesis) are compared to an analytical model based on membrane potential variations as "shot-noise" due to stochastic background activity.
Effective connectivity between directly connected cells is found to vary as a function of the rate of stochastic EPSP-bombardment of both the pre- and post-synaptic cells. The observed decrease in efficacy of connection with increased rate of input onto the pre-synaptic cell is likely due to non-linearities in spike-generation and refractoriness. Simpler models, such as the probabalistic "copy" model of Aertsen & Gerstein (1985, Brain Res., 340: 341-54), often fail to account for such non-linearities.
Effective connectivity between cells not directly connected, but receiving shared input from a third cell, is predicted to be extremely weak by the analytical model, and found to be undetectable in the simulation results. However, increased correlation between the background inputs to the two cells greatly strengthens this effective connectivity. This suggests that significant shared input correlations seen in cortical recordings are due to synchronous input from many cells.