2.4 Discussion
4.4.5 Cortical Activity and Computation
I have presented evidence that superficial layers of cortex are significantly more active than deeper layers. While somewhat surprising, it is actually consistent with a widely held view of cortical processing: Layer 4 receives the direct input to cortex, and it would be very surprising if it were not significantly active during visual stimulus. Layer 2/3, or the “associational” layers are believed to perform the first processing on the input, and my results suggest that the processing in layers 2/3 involves the majority of the neurons there. Layers 5 and 6 form the output from each cortical region: layer 5 projects to other parts of cortex and subcortical structures, while layer 6 projects feedback to the LGN. My finding that only a small minority of neurons in the output layers are activated seems logical if I assume that individual neurons in layer 5 and 6 encode different specific
results about the input and only become activated when they are transmitting information about the present stimulus. For example, if an output neuron encodes the presence of an oriented edge at a particular position in the visual field it would only become active when such a stimulus is present. At the same time, my conclusions regarding sampling bias call into question the most widely held assumption about processing in V1: that what neurons in primary visual cortex “do” is detect oriented edges at different locations in the receptive field [Hubel and Wiesel, 1962]. As most studies do not detail the minimum thresholds for recording and the amplitude of their clustered units I can only guess that they were, in all probability, highly biased towards deeper layers. (Indeed, inspection of the scaleless recording illustrations in [Hubel and Wiesel, 1962] shows spikes that are many times greater than the background noise, suggesting they are very high amplitude.) Thus the standard “story” about the function of V1 neurons is most likely based primarily on deeper layer neurons, and may not apply to neurons in the superficial layers. What exactly the superficial layers do is therefore unknown, while my own results suggest that these layers are in fact the most active during visual processing.
My results bear directly on the question of whether there is a “dark matter” problem in neu- roscience, as suggested by [Shoham et al., 2006]. While [Shoham et al., 2006] used relatively simple calculations to suggest that 90% or more of the neurons in cortex are slient, my own more detailed analysis suggests that the overall activity level under my protocol is around 50%. The main reason for this discrepancy is that [Shoham et al., 2006] assume a uniform detection range of 50µm for all layers. In contrast, I have found that even the large cells in layer 5 are somewhat less detectable than this, and small neurons in superficial layers are significantly less so. Another reason for the discrepancy is that [Shoham et al., 2006] use an estimate for the packing density of neurons in cortex of 60,000-80,000 neurons/mm3, taken from [Scholl, 1956], and I use the more recent and somewhat
lower estimate taken from [Bealieau and Colonier, 1983]. In any event, while my own results still leave me to wonder what the other 50% of the neurons were doing, it does not support the view that there is a “dark matter” problem in neuroscience – rather the problem seems to be that the instruments and practices commonly employed are biased towards the “brightest” objects in the “universe” of the brain.
On the other hand, it is difficult to say whether my results are directly relevant to theories of sparse coding, such as those described in [Olshausen and Field, 2004]. In part this is due to some ambiguity in the sparse coding literature about whether “sparsity” means that neurons fire only a few spikes to transmit information, or whether only a small proportion of neurons respond in processing a stimulus. My results seem to contradict the latter interpretation, or at least show that this type of sparsity pertains only to the output of cortex and not to the input and processing.
My results suggest strongly that the function of the different layers of cortex are significantly different, as reflected by significantly different spiking rates. Further, it is most likely that less
information has been gathered in the past about the function of superficial layers of cortex than about their counterparts in deep layers. In order to improve this situation future studies should do their best to explicitly collect data about what layer of cortex their recordings are made in.
Chapter 5
High-Amplitude Positive Spikes
5.1
Introduction
Using a protocol similar to Hubel and Wiesel [Hubel and Wiesel, 1962] as described in chapter 4, I methodically sampled the average spike waveforms of neurons in V1 of the anesthetized cat. In this process I observed a subset of spike waveforms that had unusual features in comparison to spike waveforms recorded in CA1: In CA1 spikes usually have negative polarity (“negative spikes”), meaning that the leading peak voltage amplitude is negative, while spikes whose leading peak is positive (“positive spikes”) typically have only a fraction of the amplitude of negative spikes. In cortex I recorded a significant number of positive spikes whose amplitude is as large or larger than the negative spikes as illustrated in Chapter 4, Figure 4.4. When I tried to recreate high amplitude positive spikes (HAPS) in my biophysical model, I found that the positive polarity is probably due to a process of action potential (AP) initiation in the distal dendrites of the pyramidal cell. The most plausible explanation for the high amplitude is the near simultaneous discharge of a small cluster of neurons.