Hippocampus anatomical properties to integrate

[rcoreilly]

This discussion has various sections for hippocampal anatomy / physiology to incorporate into our new axon model, with hypothesized functional implications. Based initially on conversations with Menno Witter.

Tagging @zycyc.

[rcoreilly]

Longitudinal and Transverse Axes

• Longitudinal axis is dorsal to ventral (in rodents; posterior to anterior in primates), which is aligned with the dorsal / ventral organization of striatum and PFC:

  • ventral is "limbic", motivational (ventral striatum, amygdala, OFC, vmPFC) -- i.e., the BOA model.
  • dorsal is sensory-motor action network (dlPFC <-> parietal).

• Transverse axis is MEC (medial entorhinal -- to proximal CA1) vs. LEC (lateral entorhinal -- to distal CA1), which mirrors the dorsal -- ventral sensory organization of posterior cortex, and perhaps more specifically the medial parietal body-scale navigational spatial pathway (PCC <-> RSC), but also has aspects of the dorsal / ventral organization of the long axis:

  • MEC <-> parietal spatial / structural (medial parietal), via the presubiculum / parasubiculum, TH (parahippocampal), retrosplenial cortex (RSC). This is where grid cells are found (only 10%), with other cells showing related spatial coding (border cells, etc). Dorsal grid cells have small RFs (high spatial frequency) while ventral are large (low spatial frequency).
    • MEC dorsal 5a output neurons do not receive output from the hippocampus (CA1) -- they are "input only" spatial signals.
  • LEC <-> temporal object, via perirhinal (incl TF parahippocampal). But also gets most of the vmPFC / BOA / limbic inputs. LEC is the master integrator of hippocampal input / output and is the most enlarged area in primates. Episodic memory is mostly about LEC, while the spatial input has a special MEC pathway that helps to organize LEC content.

• MEC layer 5 SST inhibition asymmetrically drives 2a LEC -- spatial gating of LEC.

• ventral MEC gates dorsal MEC 5a output -- large spatial receptive fields are the most important for memory -- want broad generalization across space.

[rcoreilly]

Anatomically, the above figure from Canto et al (2008) shows that the dorsal aspect of both LEC and MEC (magenta) receives the primary sensory inputs, and projects to the dorsal portion of the longitudinal axis, while the ventral portion (cyan) receives from more ventral / "limbic" areas (amygdala, vmPFC for LEC, olfactory, piriform cortex for both MEC and LEC).

Meanwhile, the transverse axis is organized along the MEC <-> LEC axis.

• Canto, C. B., Wouterlood, F. G., & Witter, M. P. (2008). What Does the Anatomical Organization of the Entorhinal Cortex Tell Us? Neural Plasticity, 2008, e381243. https://doi.org/10.1155/2008/381243

[rcoreilly]

Laminar organization of EC

EC is overall very similar to neocortex in terms of laminar organization, with a few key differences.

• 2a is the stellate / fan cell / reelin positive lamina that drives the input to the trisynaptic pathway. This is mostly where grid cells are found in MEC (maybe also in 3, 5 but not clear). These are input only -- they do not project to other lamina.

  • Perform an initial pass of pattern separation in our model, improving subsequent pattern separation in DG, CA3.
  • Have an extensive inhibitory surround network, but no lateral excitation -- still consistent with the "continuous attractor network" (CAN) model which actually depends more on the inhibition than the excitation in terms of forming topographic grid cells.
  • This is analogous to layer 4 of neocortex, which has stellate cells, and may provide a good template for its function in neocortex: pattern separation and perhaps additional PCA-like clustering as a self-organizing first-pass organization of the inputs to make a higher-dimensional basis space for subsequent learned categorization in layer 2/3 of neocortex.

• 2b and 3 are pyramidal cells that have highly topographic bidirectional interconnectivity with CA1 -- this is the monosynaptic pathway that is the invertible decoder pathway in our model (along with 5b).

  • There are very strong recurrent connections among these cells, along with standard PV fast spiking FF inhibition.

• 5a is the output pathway, receiving from CA1 and sending recalled memory to the cortex / subcortical targets (EC_out in our standard hip model). These are likely neocortical-like 5IB PT (pyramidal tract) layer 5b neurons, with an apical tuft into layer 1.

• 5b sends output info back to the CA1 for training the monosynaptic invertible decoder, receiving both from 3 and 5a, and sending back to 3.

  • EC_out in our model handled both 5a and 5b, but this division may be useful for the temporal-difference error-driven learning mechanism, with 5b switching between 3 (input) vs. 5a (output) over the theta phase, in coordination with CA1 switching between receiving from CA3 vs. EC (layer 3).

[rcoreilly]

Anterior Thalamus, Pre/Parasubiculum, and Head Direction

CA3, CA1 project via fornix to the septum (which sends GABA, ACh, glutamate projections back to hippocampus and is thought to drive theta rhythm) and the _ mammillary bodies_ which then project into the anterior thalamus (AT). AT has extensive interconnectivity with pre / parasubiculum, and is critical for the head direction integration system.

Interestingly damage to AT causes profound amnesia.