In preparation for transgenic studies of the entorhinal cortex, the Moser lab (out of Trondheim, Norway) released a paper this month outlining the similarities between grid cells in mice and the better characterized grid cells in rats. Grid cells were discovered fairly recently. They fire when a rat is in specific spatial locations, but unlike place cells, there are many firing locations for one grid cell and the locations form a hexagonal grid. This system may provide a coordinate system for navigation and may inform the more specific spatial coding system in the hippocampus. The entorhinal cortex is, of course, directly upstream of the hippocampus in the hierarchical funnel of information processing that makes up the medial temporal lobe (if you believe in that sort of thing.. hierarchies… i mean.. in the brain..).

Items of note.

Theta power jumps abruptly as electrodes move from the postrhinal to the entorhinal cortex. Theta is either generated by intrinsic rhythmic properties of local interneurons or principal cells OR by rhythmic input from extrinsic afferents from the hippocampus or perhaps the medial septum (he said, not knowing for sure if the septum projects there). My gut goes toward the latter because it isn’t easy to imagine a large change in membrane properties of cell populations that develop right next to each other. OTOH, one little ion channel or neuromodulator receptor (or calcium-buffering protein?) could do the trick. In any case, it will be interesting to see this replicated and determine if any innervation patterns or physiological variables correlate with the topographical theta power function.

Not a finding from this study, but cited. van Groen et al. showed that the wiring diagram for the C57BL/6J strain of mice is unusual. This will be a hindrance in interpreting studies that use only this strain, but will be of interest as a possible “natural experiment” to understand the evolution and function of the particular altered projection.

Grid sizes are similar between mouse and rat. This means that the grid coding isn’t about self-motion cues or how long a rat’s leg is. I have to think about what this means more. It’s not immediately apparent why one grid size would suit a rodent better than another. It’d sure be interesting to know what human grid sizes are.. or at least monkeys. Maybe we could use a virtual environment.

The lab is clearly on the verge of some very interesting transgenic studies. Reading their discussion, it is apparent that they are excited about reversible activation or inactivation of specific neuronal populations. Every new study on the genetics of cortical development brings us closer to having control over every distinct cell-type individually. Adding in viral gene delivery allows spatial control in the absence of specific developmental knowledge. Technologies are already available for inactivation using inducible genetic systems tied to proteins involved in synaptic physiology (for instance, Tonegawa’s tetanus-toxin mouse). These are great, but the timing is still not precise. Waiting for the genetic induction can take weeks. Another generation is on the verge, using light controlled ion channels. The major technical hurdle in my mind was how to get the right kind of light inside the right part of the skull, and I think that may have just been taken care of. The Moser Lab will be interested in controlling activity in the entorhinal cortex with these techniques, but I hope they will also avail themselves of other important transgenic methods such as monosynaptic retrograde tracing and activity-dependent cell tagging. Looking forward to hearing about the studies that this report suggests are underway.

Siik.org is where I get dope remixes that my friends don’t know about. He updates in spurts, so I consume in spurts. Just noticed his mix of Brandy - I wanna be down. I’m not sure if it’s bad etiquette for me to throw it in the flash player. I’ll email him and ask.

Also, check the longform r&b mix from earlier in July. Excellent for chillin’ out in pretty much any context.

Last night, it was all Houston for Dummies from the Rub. I can’t figure how to link to individual podcasts, but just subscribe if you’re into it. Let me warn you now. Houston rap is not for people with delicate sensibilities. That ‘25 lighters’ hook sticks in my head for days tho… ‘I got to get paid.’

Ji and Wilson put out a paper a few months ago entitled “Firing Rate Dynamics in the Hippocampus Induced by Trajectory Learning.” I covered it in journal club and we had a lot of interesting discussion but couldn’t quite square the results and interpretation. Below I’ll lay out my thoughts on the paper.

Central issues.

• Does the hippocampus “represent and link individual events in a sequence”?
• How does it represent a novel sequence of locations?
• Does the hippocampus differentiate between different location sequences by shifting firing rates (the “firing rate hypothesis”) or firing locations?
• Specifically, how do HC cells respond to a shift in reward contingency from A+, B- to A-, B+? A and B in this case are two location sequences. Also, how do the cells respond to the novelty of B, since it is rarely experienced before the reward shift?

Behavior.
The animals are taught to alternate sides in a figure eight maze, tracing out an infinity symbol. Then one side loses its appeal and the rats have to learn to just go in circles on one side of the maze. The behavior change was gradual and smooth. It would appear that there is a relatively big jump in performance right after the first time the rats try the unilateral technique.

Some people believe that gradual learning is not indicative of hippocampal processing.

“The hippocampus assigns distinct representations to input patterns to avoid interference across memories, whereas neocortex uses overlapping representations that encode shared structure across many different experiences. Furthermore, neocortex uses a small learning rate to gradually integrate new information with existing knowledge, whereas hippocampus uses a large learning rate to encode episodic memories of specific events as they unfold.” - O’Reilly and Norman 200

Or. If you don’t subscribe to a particular theory of hippocampus function, at least consider what the term ‘episode’ indicates. Episodic learning creates the memory of the first time you heard that new song, not the hundreds of times afterward when you get obsessed and burn yourself out on it.

The change in associative value (particularly extinction of A’s value) should be visible between the cutoff of the the reward and the actual first lap of trajectory B. The encoding of a new trajectory will happen after the trajectory is experienced (duh). I think it would be nice to extrapolate from changes in the period between the reward switch and the actual first lap (AFL) and subtract these predicted changes from anything that happens after to study the new trajectory. OR just move a wall and forget all this associative value stuff. Are we studying trajectory (i.e. sequence) learning or associations between sequences and rewards? It is not immediately clear which categories of learning this paradigm would engage.

How is a new trajectory represented?

The trajectories are really a combination of two possible starting positions and two possible ending positions. I’ll use their system, but ignore their counterbalancing because it clutters my mind and confuses me. You can start at L or R and end at L or R. All trajectories share the central track and most analyses in this paper are confined to this segment. LR and LL share the same past but have a different future. RL and LL have have a different past but share the same future. For Cell X, the mean firing rate over all trials and all sessions on the central track for an LL trajectory is more similar to that during an LR than an RL trajectory. The similarity is not explained by behavioral variables like head direction or running speed.

This figure collapses across time and keeps cells separate. Just a note because later on we collapse across cells and keep times separate.

This seems to indicate that novel trajectory encoding is piggybacked on old trajectories. The representation of the new trajectory ends up hijacking the coding scheme for LR. As the hippocampus representation of LL becomes very similar to LR, I wonder why this doesn’t cause confusion. If the two become equivalent and they guide behavior, how does the animal know that he likes to do LL more than LR? Clearing up this confusion would be especially important right at the choice point near the end of the central arm.

How do familiar trajectory representations respond to the contingency change and the novel trajectory?

Even after the rat should know better and go for LL all the time it still makes mistakes and takes an LR or RL trajectory from time to time. Ji and Wilson looked cell firing rates during these ‘mistake’ trials on familiar trajectories compared to baseline trials when the familiar trajectory was appropriate. The rates did change, they could go up or down. We are more concerned with the magnitude of the change than the direction. The authors used a z-score for the comparison, normalized to the standard deviation during the 20 trials before the reward switch. After the reward switch, firing rates shifted a bit (Pre in the figure below). After the behavior switch, they shifted a little bit more (Post).

One thing that really confuses me: Why shouldn’t the normalized rate change during baseline laps be exactly 1? Shouldn’t the average deviation from the mean be exactly one standard deviation away from the mean? I’m willing to accept that I got lost in the description of the equations. Maybe it’s because SD is a square root and the rate change is not? Or maybe it’s because the rate changes utilize this strange “three lap window” device instead while the SD takes all measurements independently?

Figure 4A (after this paragraph) shows that place fields don’t move around on familiar trajectories. Although it might be worth inspecting what happens after the actual first lap in a little more detail. The authors argue that since there is no linear correlation between lap number and place field center of mass, there is no change. This type of analysis would be heavily influenced by the length of the baseline you use. The trendline might look a lot different if there were only five laps before the actual first lap in figure 4A(right). I can’t see any indication that they tried polynomial contrasts to check for anything other than a linear relationship. No sense in belaboring it. We can at least take away that a center of mass shift isn’t the overwhelming trend.

How do novel trajectory representations form and change as the trajectory becomes familiar?

When novel trajectories are first encountered, the representation is much like trajectories that have a different past but the same future. The future seems to matter more. Novel LL trajectories initially have firing rates similar to RL trajectories. Over the course of many trials on the new trajectory, the representation shifts. A large portion of cells (17/42) start out mimicking the firing rate of trajectories going the same place, but eventually shift over toward a firing rate representation like those trajectories with the same past. The future is important at first and then the past comes to dominate. This is visible on the individual cell level.

The shift is also a property of cell assemblies. That is, the coding shifts in a coordinated fashion among many cells. Many analyses of hippocampal firing patterns deal with cells individually. A remarkable amount of knowledge has been generated using these techniques, but we all sort of tacitly assume this isn’t really how the brain works. There isn’t one unique cell that says we are on a left-right trajectory. One cell hardly has the ability to fire another cell downstream. Hundreds or thousands of neurons at once have to signal to the downstream cell to make an impact. Ji and Wilson explicitly acknowledged this and made an attempt to look at all recorded cells together. They created a vector containing all cell firing rates and measured the correlation between multi-unit vectors representing familiar trajectories with the novel trajectory as it changed over time. They also divided the central portion of the track into three sections to see the influence of the future and past as the rat finishes off a trajectory and decides on a new one. Segment 2 of the central track is special because there is little to no difference in sensory or motor information on this segment. Any differences in segment 2 should be almost entirely cognitive.

In the paper they present one set of data that is not controlled for behavior differences, but present more strictly controlled data in the supplementary information. This is probably the most important figure in the paper and one of the most perplexing. Figure S7B is shown below.

The blue dots represent the correlation between a cell assembly’s firing rate representation on a novel trajectory with a same-past / different-future trajectory. The red is the novel trajectory correlation with a same-future / different-past trajectory. The trend for segments 1 and 3 is clear. The future matters a lot and then, over time, the past comes to prominence. In segment 2, after controlling for behavioral variables, the correlation decreases or stays steady for same-future and same-past trajectories. Perhaps here it is possible to see a novel representation forming that is unique to the novel trajectory. The shift is most obvious in segment 3, just before the rat must decide to go left or right. Just to remind us what this analysis means, here is a quote from the paper:

“Furthermore, removing dependence of rate on behavioral parameters produced a similar dynamics of different-past r for all the 3 segments, and of same-past r for segment 1 and 3. The same-past r in segment 2 showed a nonsignificant change after this correction.”

Interpretation.
This paper provides support for the firing rate hypothesis. Subtle changes in an environment don’t produce entirely new representations. Rather, the existing landscape shifts, so some cells get a little more vocal and some get a little more timid. This is a relatively new observation and idea and more reports like this one help secure it as a significant type of coding. You always worry that intriguing schemes like this are maybe a trick of the math, so independent observations of the same phenomenon under different conditions are key.

The major observation in this paper is that novel trajectories shift from an heavy future-influence to a past-influence. The authors confusingly interpret ‘future’ as ‘current-information’:
“In our experiment, the activity patterns on the new trajectories were initially more correlated with the trajectories that shared the same current information (Fig. 7). In segment 1, the initial rate correlation was high between same-past trajectories, because of similar head direction and immediate future locations, but low between different-past trajectories because of the head direction difference. In segment 3, the initial rate correlation was high between different-past trajectories because of the similarity in head direction, lateral position, and immediate future locations, but low between same-past trajectories because of the difference in these parameters. In segment 2, the initial rate correlations were relatively high both between different-past and between same- past trajectories, because of similar head direction, lateral position, and immediate future locations across all trajectories.

With more experience on new trajectories, activity patterns became more similar with the trajectories that shared the same immediate previous experience and more different from those with different immediate previous experience. Regardless of the different initial conditions as mentioned above, activity in all three segments on new trajectories became more separated from their different-past trajectories and more correlated with their same-past trajectories (Fig. 7), indicating an increase in the influence of immediate past input on place-cell activity.”

They just said two pages ago that the trend is the same when behavioral parameters are controlled. Why then explain the shift with a shift in behavioral parameters? Am i misinterpreting one of these two directly contradictory statements?

Here’s my alternative interpretation. The representational shift does not reflect an increasing influence of the immediate-past. It reflects an initial strong dependence on the immediate-future. The reason why the effect is most striking in segment 3 is that segment 3 is the choice-point. In segment 3, the rat must recall its goal and decide whether to go left or right. Early in the game, when the reward-contingency has just changed, the rats will need to separate the old contingency from the new one and maintain the new contingency on line to avoid going on alternating trajectory auto-pilot. I suggest that effect described is due to a shift toward habitual behavior away from an effortful cognitive task. In that case, perhaps the shift in influence isn’t from entorhinal cortex (current information) to CA3 (retrieved information) as suggested by the authors, but simply reflects a decreasing reliance on prefrontal cortex active maintenance of task demands. The point is to emphasize the change in cognitive strategy over the change from novel to familiar. As a sidenote, I wish the authors would present an analysis of theta phase modulation during this same task. If this shift really is due to increasing dominance of CA3 information, would you expect to see a shift in preferred firing phase? Others have shown that the apical dendrites of CA1 cells (where entorhinal information comes in) are more excitable during the fissure trough of theta, while CA3 inputs have a larger effect at phases 180 degrees away at the fissure peak of theta.

Main conclusions:

• Firing rate hypothesis is supported. Firing rates but not locations change as during trajectory learning.
• Prospective and retrospective coding hypotheses are supported. Neurons shift from prospective coding to retrospective coding as novel trajectories become familiar.

As a treat for those who made it all the way through this relatively dry discussion, let me direct your attention to this remix of Theme from Mahogany by Captain Funk off of the excellent Diana Ross + Supremes Remix album from Universal Japan. The dubious relevance will be apparent to folks who know the song already.

We covered a paper from Loren Frank’s lab in journal club a few weeks ago. They looked at the coordination of neurons in the hippocampus as animals experienced a novel location. The hippocampus rapidly forms representations of new environments in the form of location-specific cell firing and these representations stabilize over a longer period. This paper examines some forms of neural coordination that may assist in stabilizing representations. The surprising result is that the important stabilizing events may occur during moments when the rat is replaying or thinking back over his experience in the novel location rather than while the rat is actively experiencing said location.

Here are the premises you need to understand the article. First, the mechanism for memory is synaptic plasticity in a very broad sense. Some change in the impact of a neuron’s firing is responsible for new knowledge. The ’synaptic’ part means that we are assuming that the change is specifically happening at the point of connection between neurons rather than changing overall excitability or rhythmicity or signal integration etc. These other mechanisms are possible, but we have thousands of synapses on a given neuron that we can change in an near-analogue fashion, so the parameter-space seems larger for that change than for shifting a whole-cell property up or down. Now, there are myriad types of synaptic plasticity. Almost all of them depend on some temporal relationship in firing between the neuron sending information (the pre-synaptic cell) and the one receiving info (the post-synaptic cell). Machinery in the synapse detects correlations (or consistent anti-correlations) between cell firing and adjusts the volume on the synaptic connection in response. If you’re flipping through your radio stations and all you get is fuzz, you turn the radio down. If you get your favorite song and it makes you very excited then you go ahead and turn it up a little louder. In spike-timing dependent plasticity, if the pre-synaptic neuron fires within a ~20 ms window before the post-synaptic neuron then the synapse will get strengthened. One way to think about this is to imagine that the post-synaptic neuron wants to get more friendly with neurons that make it fire. If the Neuron 1 consistently fires just a little before Neuron 2 fires, then N1 is likely to have helped push N2 over the edge. All that is to say that neuronal firing has to be coordinated in the time domain to cause much of a lasting change in connection weights. (Figure below from Bi and Poo, 2001).

Another very key piece of knowledge is that a large portion of cells in the hippocampus are spatially tuned. They fire when a rat is in a particular location. This discovery from the early 70s inspires a dominant paradigm in hippocampus research referred to as Cognitive Map theory. The initial manifesto on the subject by O’Keefe and Nadel is available for free online.

There are two major brands of network level organization in the hippocampus. Both refer to coordinated network events that register as changes in the local field potential (LFP, signals in this domain are analogous to EEG signals). One level of organization is called theta phase compression. Theta rhythm in the hippocampus is observed when a rat is roaming around and changing positions. It is easier to explain the precession/compression aspect with examples and it turns out to be less important in Frank’s study, so I’ll just wait till we get to an example. The other organizer is called a sharp wave ripple (SWR). These are large amplitude jumps in the LFP signal with high frequency ripples superimposed that occur while a rat is just chillin, although recently people have shown that they can also happen when the rat just slows down for a minute. I like to think they happen when he’s moseyin instead of truckin. The interesting thing about SWRs is that some folks have reported that sequences of neural activation are replayed on a drastically compressed timescale during these things. During normal behavior place cell X will fire before cell Y because the rat enters location X before location Y. During an SWR at the end of that jaunt, the rat will replay the X-Y sequence superfast, thus creating an opportunity for the cells to fire much closer in time. The connection between X and Y should grow if X consistently fires within a 10-20 millisecond window before Y. This is well-suited for encoding a new sequence.

Cheng and Frank’s paper doesn’t speak directly to the sequential aspect of the network organizers, but it does look at the activity/co-activity of individual cells during these organizing rhythms. Additionally, they look at these dynamics at a time when the hippocampus should be encoding new information because the rat is being exposed to novel location (one arm of a radial maze). In the first half of the paper they look at SWR-like events, and in the second half they look at phase compression. As a proxy for network organization, they examined the relation of spike timing for pairs of cells. They counted how many times Cell X fired 5 ms after Cell Y vs 6 ms vs 7 ms, etc.. If there are a lot of events in which Cell X and Cell Y fire very close in time (near 0 ms), then this is referred to as excess correlation. Oh. One other thing. All of the cell-pairs in this study have overlapping place fields. You could think of this as redundant coding (though it probably isn’t). Cell X and Cell Y both code for location Z.

The excess correlation is best observed under specific conditions. It is not apparent in familiar locations, indicating that this organization is somehow induced by novelty. Surprisingly, the excess correlation is not observed if you only look at activity in location Z. In other words, the cells sometimes fire outside of their field (you might have thought that was just pesky noise). It is during these ‘erroneous’ events that the cells coordinate their firing. Also, the coordination happens during periods of high-frequency tremors in the LFP signal. These look suspiciously like SWRs, but the amplitude isn’t high enough to count. Elminating just SWRs from the analysis wasn’t enough to get rid of the excess correlation, but getting rid of the “high-frequency events” regardless of amplitude does the trick. If you think about it, you can justify this. The local field potential is the reflection of a simultaneous change in the membrane potential of many neurons with dipoles aligned. The amplitude should probably reflect the actual number of neurons involved and the tightness of the synchrony in the potential shift. Probably some small subset of cells is assigned the task of encoding this new location within an overall familiar environment, so their coactivity doesn’t have the oomph to register a 6.7 on the Richter scale, so to speak.

To be certain that the excess correlation was truly an indicator of greater network organization rather than some other factor, the authors had to account for the cells’ excitability and activity during HFEs. If the cells just fire more often altogether then of course they’ll fire together more often, and that’s not very interesting network-wise. However, further analyses indicated that novel arm cell pairs are more coactive than they should be according to their individual activation probabilities.

In figure 5, the authors look at the time scale of the coordination. I find this data difficult to reconcile with the overall story. The time differences between spikes are expressed as root mean squares. In my reading, this is, in a sense, the average difference between spike times. The biggest difference between familiar arms and novel arms shows up at around 0.5 on the cumulative probability plot, when day 2 of novelty has RMS equal to ~40 and familiar arm has RMS equal to ~60. Neither of these is in the window for spike-timing dependent plasticity as it is classically defined. In fact, there are essentially zero events that fit within the 20 ms window that I would have though was important in any condition. I don’t know what to make of this. Either I’m misinterpreting the meaning of RMS or the timing is ideal for making miniscule alterations to synaptic connectivity or this coordination has nothing to do with plasticity. I suppose another option is that the parameters discovered in slice physiology for spike-timing dependent plasticity are more like guidelines and they don’t reflect the actual requirements in vivo. Perhaps the window for summation is extended in vivo.

What else? Oh yeah. Phase compression. Individual place cells don’t code for especially precise areas of space. The cell will fire anywhere within a ~30 cm diameter circle which defines its place field. The firing rate is highest in the center of the field and if you graph firing rate versus distance through the field you will end up with a Gaussian or maybe Poissonian distribution surrounding the field center. This means that firing fields can overlap. This, in turn, means that cells coding for sequential place fields can fire together in a very small time window, perhaps allowing plasticity or coordinating some downstream plasticity. However, if overlap was all, then firing order would be arbitrary in the shared space. Cell 1 could fire before or after cell 2, and that generally means fluctuations in synaptic strength in either direction. What if you had a way to make the cells fire in the order that their place fields are experienced but on a compressed time scale? This is what phase compression achieves.

Phase compression is a property of multiple cells. It relies on the single cell property of phase precession. Phase precession was first described by O’Keefe and Recce in 1993. The area where a given place cell fires (the place field) usually takes around 0.5-1 second. This is enough time for several cycles of theta oscillation (5-12 Hz). During that first theta oscillation, when the rat has just entered the field, the cell fires very late in the cycle, say at 330 degrees. By the time the rat is exiting the firing field the cell fires early in the cycle, maybe 30 degrees. There is a linear relationship between the firing phase and distance through the place field. The example below might help. Each tick mark is an action potential, and they are shown in relation to overall theta rhythm.

As Cell 1 is leaving its field it fires at 30 degrees theta and cell 2 (entering its field) fires at 330 degrees theta, that corresponds to ordered firing on the time scale of about 80 milliseconds. This is phase compression. The experience of traveling through field 1 and field 2 took seconds, but the cells that represent those fields re-experience the journey on the compressed time scale of milliseconds.

Cheng and Frank looked for evidence of phase compression in their ‘novel arm’ cells by examining the correlation between distance between fields and distance between theta phases. Phase compression was almost always apparent, but the strength of this relation went in the opposite way from the high-frequency event coordination discussed above. Early on in exploration of a novel arm, phase compression was not very strong and then improved with familiarity. If you really take a look at figure 6, it looks like the phase compression takes a hit on day 2 of exposure to the novel arm.This is probably due to a low number of observations. The number of cell-pairs examined changes drastically across days, from 78 to 18 to 40. Why? I don’t know. The strength of the compression is read out as a sequence compression index (SCI) in the figure:

This paper is about novelty. It’s about novel locations and how the hippocampus deals with them. The results indicate that novel representations may be stabilized during high-frequency events and not during location-related firing on the theta time scale. Theta time-scale coordination looks more like a readout of an already stable representation. I can’t understand yet why coordination during HFEs isn’t a little bit closer in time. My ideas about STDP say that this isn’t tight enough to make a big difference in synaptic strength. The other important piece of information to pluck from this article is that sharp-wave ripples are currently too strictly defined. If Cheng and Frank are correct, then the high-amplitude distinction is somewhat arbitrary. These data overall make me more interested in ripple activity and more skeptical about theta-phase compression as a mechanism for learning. I suppose that’s not entirely fair, as the theta-phase compression is often posited as the mechanism for sequence learning, not necessarily for encoding a new environment. But at least this clarifies for me that I should maybe corral theta compression off into the sequence learning arena away from a spatial function.

Muscimol, but not bicuculline, elicited and maintained self-administration when injected into the MSDB. The reinforcing properties of muscimol depended on the activation of local GABAA receptors in the MSDB and, secondarily, on a release of DA from the VTA and consecutive activation of D1 and D2/D3 dopaminergic receptors. These results suggest that MSDB-mediated reinforcement depends on the modulation of local GABAergic activity and the recruitment of the mesocorticolimbic DA system. - Gavello-Baudy et al 2008

Yo. Steven Bernstein’s Millenial Territory Orchestra is playing on Thursday at a new place in LES called Drom. It’s not free and I’m not rolling in dough right now, but it isn’t smart to miss these shows. MTO is an example of a band truly communicating and interacting on stage without noodly wankery. Cliches are avoided. Drums are not coddled. Eclectic influences abound. 15 bucks a ticket plus you gotta order food or two drink pushing your total price up to ~35. They win by providing really dope jazz and then making everybody smile with covers of Darling Nikki and Love is All You Need. I don’t particularly need to hear those again (though I’m fine with it) but it shows they aren’t trying to overbrow my uncultured ears.

Below the fold is the only full-length youtube I could find.

I’m studying the medial septum / diagonal band of Broca and its relation to hippocampus rhythms and firing patterns these days. Sometime soon I will have to do a deep lit review. Might as well get started.

Studies on this pathway apparently go back to the early 20th century. I think I can get some these early articles in print form. I’ll go back there when I get a chance. For now, I’ve been taking a look at early studies from Tamas Freund, who is now the director of the Institute of Experimental Medicine in Budapest and who had a string of important tracing studies over a little more than a decade of work.

I just want to document what I learn from each paper and when old ideas get turned over or improved. We’ll start with Freund and Antal 1988 in Nature. Three techniques are on display: Anterograde tracing with Phaseolus vulgaris leucoagglutinin (PHA-L); immunohistochemical staining for parvalbumin and calbindin (two distinguishing markers for GABA-containing interneurons); and immunogold labeling for GABA itself.

I’m looking for indications of how phaseolus lectin transport works, but for now all you need to know is that when you inject this protein into the medial septum, it gets transported to the axon terminals. If you want to see how two brain areas are connected you inject PHA-L into the sending department and look for PHA-L in the receiving structure. Previous observations of PHAL-labeled axons from the medial septum indicated that projections could be broadly classified into two categories based on the appearance and targeting of the axon terminals. The important terminal type for this study appeared thicker at the end and formed ‘baskets’ around cell bodies in the hippocampus. Immunogold labeling indicated that these terminals contain GABA and are thus likely to be inhibitory to their targets. Who are they targeting though?

Co-staining with the GABAergic interneuron markers parvalbumin and calbindin showed that the major target for GABA transmission from the septum is inhibitory interneurons in the hippocampus, mainly in CA3 and dentate gyrus as opposed to CA1. There was no apparent preference for parv vs calbindin innervation and greater than 2/3 of all interneurons get some level of innervation.

It is especially nice that the authors note which layer within the hippocampal subfields are preferred for GABAergic innervation from the septum. Klausberger and Somogyi have got me thinking that I can use this as at least a heuristic for guessing which types of information are being affected. Cholinergic innervation is mainly aimed at proximal dendrites on principal (pyramidal or granule) cells in all subfields rather than distal apical tufts. In the CA fields this indicates that it should modulate information coming from within the hippocampal formation rather than projections from the entorhinal cortex. Not sure about the dentate yet. Freund and Antal don’t really break the molecular layer down like that.

GABAergic innervation prefers the interneuron somas and proximal dendrites in stratum oriens and stratum radiatum of the CA3 field and hilus and granule layers of the dentate gyrus. CA1 doesn’t receive nearly as much GABAergic innervation from the medial septum. Good to know.

This paper provides anatomical evidence that when GABAergic projection neurons in the septum are active, they should be disinhibiting to principal cells of the dentate gyrus and CA3 regions. I’m going to guess that this occurs in a rhythmic fashion and that the rhythm is in someway coordinated with the cholinergic output. More reading = more clarity.

P.S. Found more on how PHA-L works:
This lectin, Phaseolus vulgaris-leucoagglutinin (or phytohemagglutinin-L; PHA-L), is obtained from the red kidney bean and is best known for its mitogenic and leucocyte-agglutinating capacities. It has been assumed that the efficacy of PHA-L as an axonally transported marker derives from receptor-mediated endocytosis, but because lectin-glycoconjugate binding seems to require that binding sugars exist in complex, multiply branched forms, this has not yet been tested. - Sawchenko and Gerfen 1985

No word yet on if this limits the interpretation of PHA-L studies in any way.

nakahira e16, originally uploaded by kaleidoscopikr.

The remarkable finding in the present study was the transient appearance of large numbers of EGFP-labeled multipolar cells in the SVZ and IMZ during E16-18. Their morphology has been described in the hippocampal primordium of the rabbit and monkey by the Golgi method, but their significance has not been adequately discussed. The presence of the multipolar cells in the SVZ and IMZ has also been reported in neocortical development. In our study, the multipolar cells accounted for 30.1% of the EGFP-labeled cells in the neocortical primordium of the mouse at E16. Their migratory pattern within the subcortical zone is nonradial and has been described as multipolar migration, as opposed to radial locomotion or somal translocation. The multipolar cells migrate slowly, at about one-fifth the rate of the bipolar cells in the neocortex. - nakahira and yuasa 2005

Thom Yorke remixed a classical piece for a release called ‘Cortical Songs’. The title is intriguing and apparently refers to an orchestra partially controlled by a ‘tiny computer brain’. You can sample the result at imeem, courtesy of pitchfork media. Also, check out the record label’s myspace. They are called nonclassical music and they are repeat offenders in the classical remix department. I might pick up a few things from these guys.

Finally, here is an excerpt from the article, “Synfire Chains and Cortical Songs: Temporal Modules of Cortical Activity” by Ikegaya et al. 2004. Cortical songs seem to be motifs of neuronal activation patterns that can recur every few minutes or so.

the entire timespan in this picture is two minutes

Our data agree with the prediction that cortical activity flows through chains of synchronized neurons (synfire chains), which are reactivated with high temporal precision. Moreover, we describe a higher order grammar (18), by which these chains themselves can be modules of larger temporal structures (cortical songs), defined by their sequential order of activation, and which can last for minutes. These songs resemble spiking correlates of sequential behavior, like bird songs (19, 20) or spatial navigation (21), and have compressing dynamics, as if the circuit was replaying and modifying previously learned sequences (21–23). The mechanisms that generate and propagate synfire chains and cortical songs must be intrinsic to the cortical circuit, because they are preserved in slices, and might reflect the faithful reactivation of specific circuits (24), mediated by stereotypical synaptic dynamics (25, 26 ) and driven by pacemaker cells (8, 27). Because the activity drifts with time, it is also possible that short-lived patterns, perhaps reflecting ongoing circuit memory, are generated de novo (6, 28).

Hippocampal pyramidal cells can be subcategorized based on several criteria besides anatomical location. For instance, distinct types have been described with relation to firing phase of gamma, change in firing rate during theta, tonic vs phasic firing patterns during theta, apical dendrite morphology, histological markers such as calbindin, and gene expression profiles.

I came across the calbindin distinction reading this recent review from Klausberger and Somogyi. They provided a few specimens in the supplementary info. It’s difficult to tell anything about their overall morphology though.

Not so for these beautiful babies from Spruston’s group. The apical dendrites are clearly either twinned or not. Unfortunately, the group wasn’t able to detect any clear functional difference between the two dendrite morphologies using whole cell recordings. I’m not terribly upset about that though. You might’ve even expected that dendrite morphology would have a lot more to do with network connectivity than intrinsic excitability.

There has been a lot of recent interest in the diversity of GABAergic interneurons throughout the cortex and including the hippocampus. Some folks recently characterized the most abundant (but relatively quiet) type of interneuron in the hippocampus, the Ivy Cell. PING came together and formed a new nomenclature for interneuron classification.

I dig that. We need to know about all the different types of interneurons and which parts of the circuit they act on and how they contribute to / are affected by system oscillations. OTOH, I’m into getting into a little more refined classification of pyramidal neurons. There is a lot less obvious diversity in this population. When we find differences, they are subtle. I think these are the types of neuronal properties that could be controlled by a manageable number of gene regulatory differences. What’s the difference between a phasic and a tonic Theta-ON cell? Is it perhaps expression of some small collection of ion channels or a GABA-A subunit? When we discover these precise single gene types of effects we can turn them back around on the system to allow investigation of the importance of a given physiological property. With combinatorial transgenic manipulations we should already have the capability to shut down firing of, say, Calbindin positive, CA1 pyramidal neurons. Then you can test for altered network properties, stimulus coding properties, and learning/navigation changes.

Anyway. I’m starting a collection of pyramidal pics on flickr. Gonna make me some mooooooo cards.