Science Guide

Malcom Brown came by and shared some of his perspective on recognition memory and the perirhinal cortex last week. Let's see if my notes still make sense.

One thing I should note before we get started is that the perirhinal cortex is closely associated with the hippocampal formation, and some have lumped it in with other nearby regions in the medial temporal lobe as one giant processing unit. The anatomy is confusing, but one major route of connectivity goes perirhinal -> entorhinal -> dentate gyrus etc.. and back out. One question to remember when thinking about perirhinal lesions is: if hippocampal lesions have an effect but perirhinal lesions don't, how does the hippocampus get the info it needed to perform when the perirhinal relay station is broken?

What is recognition?
Brown puts the core ability as 'Judgement of Prior Occurrence'. Recognition can be achieved by two processes: 1) Familiarity is a fast and automatic process. Subjectively it feels like 'knowing' something without necessarily remembering it in detail; 2) Recollection is the slow and effortful 'remembering' of an item or event. He attributes the familiarity signal to the perirhinal cortex and associates recollection with the hippocampus, while allowing some wiggle room for hippocampal cooperation in more complex forms of familiarity processing. I was initially confused by the characterization of a hippocampal process as slow. I've been indoctrinated in another school of thought that contrasts the speed of hippocampus encoding with slow multi-trial neocortical learning. However, the familiarity signal is a retrieval rather than encoding sort of process, so I think the two explanations are unrelated. In line with the broad consensus, contextual/associative information processing is attributed to the hippocampus (the devil is in the definition of those terms).

What is the cellular basis of familiarity?
The familiarity signal is an uncommon example of a decremental coding scheme.The first data we see is from recordings in perirhinal cortex of monkeys. A neuron that fires in response to a given visual stimulus fires less the second time that visual stimulus comes around. This effect is reliable and meets requirements for familiarity memory: high capacity (many items can be encoded in this scheme), single-trial learning (evidenced by the effect on the second trial), and long-term retention (the effect is still visible at 24 hours, though not as strong as the short-term).

If you line up the firing patterns for trials of novel vs familiar stimuli, you can distinguish the two looking at just the spikes in the first 100 ms after presentation. This extremely fast processing is one indication that the information doesn't have time to run through the convoluted hippocampal circuit. When I saw this data I couldn't help but wonder if the distinction could be achieved even faster using a timing code rather than a firing code. I'm reading Spikes right now and beginning to get an idea how incorporating oscillations into your coding/decoding scheme can increase your information transmission per action potential.

Lesioning the perirhinal cortex disrupts Delayed-Nonmatch-To-Sample (DNMTS) task performance. The format for DNTS is usually presentation of the sample, some item known as sample A, followed by a choice between A and B. You have to choose B to get a reward because you've seen A before. That is the nonmatch to sample bit. The delay is just the period between Sampling and Choice. I suppose if familiarity was really really broken you may not be able to perform the task even without a delay. I wonder if that's ever been observed. It's hard to imagine a brain that incapacitated. Amygdala and hippocampus lesions break DNMTS as well, but not as badly, and prefrontal lesions don't do anything.

Focusing in on the evidence for perirhinal computation here. Twenty-five percent of neurons in the perirhinal cortex that respond to stimuli show decreased firing rate for a familiar stimuli compared to a smaller number in hippocampus unless spatial processing or association is required. Brown claims to have checked upstream of the perirhinal cortex and found little to no sign of a familiarity signal. Also, as noted above, the speed at which the signal is detectable allows little time for feedback mechanisms.

How is the reduction in firing rate achieved?
Two major options: 1) Long-term depression (LTD) of excitatory synapses onto perirhinal pyramidal cells; or 2) Enhanced inhibitory synapses on said cells. Brown provided some theoretical/computational arguments that go toward the former, but didn't dwell there and I didn't quite catch it. The point was to indicate that the LTD scheme had a greater storage capacity? I hope this is specific to the perirhinal area because I have a strong suspicion that memory areas optimize their storage capacity, and I think the amygdala and certain hippocampal engrams utilize synaptic potentiation and depression in concert.

A more concrete answer comes from mechanistic studies of LTD in the perirhinal cortex. It is widely accepted that changes in synaptic strength are effected by changing the quantity of receptors in the synapse. In the hippocampus there is an intracellular signaling route leading from NMDA receptor activation -> activation of Hippocalcin -> AP-2 activation. AP-2 is a protein that disrupts the interaction between NSF and AMPA-type glutamate receptors. Here's another recent interesting finding related to NSF. The end result in Brown's model is AP-2 mediated internalization of AMPA receptors and reduced synaptic responses to neurotransmitter release.

Brown's group utilized a peptide-based drug that can block the interaction between AP-2 and AMPA receptors. This drug eliminates perirhinal LTD. Delivery of this drug to the perirhinal cortex of rats via lentiviral vector blocked LTD and disrupted performance in an object recognition task. While rats were impaired in object recognition, it was easy for them to recognize the spatial configuration of objects and when it was disrupted (i.e. hippocampus work).

Regional interactions.
Finally, he presented some crossed-lesion studies, but I didn't encode them well and I'm not sure they're published yet so maybe I'll just meditate on the concept of a crossed-lesion study. To simplify things, I'll just think of two bilateral brain regions: A and B. The point is to ask if A and B interact to produce some behavior or if they compute on their own. Does region A depend on region B to perform its computation? First remove one side of region A. In most cases, this doesn't produce a profound deficit. If region A is capable of acting on its own then the spared side will do the job. Lesioning region B on the same side doesn't do much to the already broken circuit. Now imagine the two outcomes of a contralateral (other side of the brain) lesion of region B. If there is no effect, then the region wasn't important and region A can really do the job on its own. If there is an effect, then region A and region B have to be able to communicate to do the job. It's not enough to have one copy of region A and one copy of region B. They have to be on the same side of the brain chatting with each other to do the job. An analogous situation would crop up in genetics if you broke the promoter region on one chromosome and the coding region on the other. It doesn't do you any good to have spared copies unless they can interact.

Miscellany.
The perirhinal cortex is juxtallocortex meaning that it is near to the hippocampus and shares some of its cytoarchitecture. It lacks columnar organization and a clearly defined layer 4. Is there a column-inducing factor inhibited by a factor from the cortical hem? Are layer 4 and columnar organization dissociable?

Can you imagine not knowing something 20 minutes after it happened but knowing it 24 hours later? That's the effect reported for a kainate receptor antagonist in the perirhinal cortex.

Real recognize real.