Dense Computer Replica of Cortical Microcircuits Unravels Cellular Underpinnings of Auditory Surprise Response

  1. Oren Amsalem
  2. James King
  3. Michael Reimann
  4. Srikanth Ramaswamy
  5. Eilif Muller
  6. Henry Markram
  7. Israel Nelken
  8. Idan Segev

Reviewed by

Read the full article See related articles

Discuss this preprint

Start a discussion What are Sciety discussions?

Listed in

Log in to save this article

Abstract

The nervous system is notorious for its strong response evoked by a surprising sensory input, but the biophysical and anatomical underpinnings of this phenomenon are only partially understood. Here we utilized in-silico experiments of a biologically-detailed model of a neocortical microcircuit to study stimulus specific adaptation (SSA) in the auditory cortex, whereby the neuronal response adapts significantly for a repeated (“expected”) tone but not for a rare (“surprise”) tone. SSA experiments were mimicked by stimulating tonotopically-mapped thalamo-cortical afferents projecting to the microcircuit; the activity of these afferents was modeled based on our in-vivo recordings from individual thalamic neurons. The modeled microcircuit expressed naturally many experimentally-observed properties of SSA, suggesting that SSA is a general property of neocortical microcircuits. By systematically modulating circuit parameters, we found that key features of SSA depended on synergistic effects of synaptic depression, spike frequency adaptation and recurrent network connectivity. The relative contribution of each of these mechanisms in shaping SSA was explored, additional SSA-related experimental results were explained and new experiments for further studying SSA were suggested.

Article activity feed

  1. eLife

    ###Reviewer #3:

    This is a great paper that takes a modelled somatosensory microcircuit and, without parameter adjustment, asks whether stimulus-specific adaptation is capable of emerging. The ability to remove synaptic depression and stimulus-frequency adaptation, in both thalamo-cortical and cortico-cortical populations was a definite highlight for me. Primary negatives were minimal mention of certain aspects of connectivity, and a complete lack of any mention of interneuron processing and its known role in SSA.

    Major Comments:

    1. The NMC model is derived from somatosensory cortex. It's not really discussed at all in the paper, but is the assumption that auditory cortex is similar enough in structure that it is valid to model it with a somatosensory model? Although I'm not a somatosensory expert, there are certainly numerous connectivity differences between auditory and visual cortices (interactions between L6 CT neurons, and the local cortical column for example).

    2. It was not immediately clear to me, how exactly the MGB->ACtx was wired up, and consequently, how this wiring affected tuning bandwidth in ACtx. I don't think it was a one-to-one mapping that was used, because there is talk of multiple TC afferents innervating a single cell, but this should be described in detail. How do these connectivity choices affect bandwidth, at a layer-specific level? (i.e. one could imagine a broadly tuned neuron being so because it's integrating auditory information from heterogeneously tuned thalamic neurons).

    3. Related to points 1&2, it looks from Figure 1C, that the TC input is generating a tonotopically ordered map in ACtx? Is this the case? If so, in light of many recent papers that have shown substantial local heterogeneity in ACtx frequency tuning, this is not particularly plausible.

    4. I appreciate that this is not the focus of the paper, but it wasn't clear to me whether the NMC model consisted primarily of excitatory neurons, or whether there were inhibitory neurons that were included in the analysis. If the population is mixed, then this will affect interpretation of the depression experiments. In some sense, this is also my biggest negative about the paper - there is almost no mention of interneurons at all, even though interneurons also play an important role in SSA (given that they shape frequency-dependent responses) - this has been the focus of several publications from the Geffen Laboratory.

    5. It was mentioned in the discussion that the model was not capable of replicating layer-specific SSA values. Related to this, does the model capture layer-specific changes in frequency tuning properties (i.e. layer 5b pyramidal cells have far broader tuning than other cell-types). And if not, might this affect the SSA differences, especially given how important bandwidth in shaping SSA (TC afferents responding to both deviant and standard).

    6. Were there any layer-specific effects on removal of thalamo-cortical vs cortico-cortical, that could be linked to the fact that different excitatory cell-types in ACtx have vastly different laminar connectivity patterns (L6 CT translaminar inhibition, L5 PT vs IT, for example).

    7. How does the model connectivity map onto the distinct morphology of heterogeneous cell-types throughout the cortex, and does this morphology affect the SSA? (The large apical dendrites of L5b neurons, for example, will play a huge role on how they integrate ascending sensory input).

  2. eLife

    ###Reviewer #2:

    In this study authors aim to explain the mechanisms responsible for induction of stimulus specific adaptation (SSA). As the model system authors pick the auditory cortex, where this phenomenon has been well explored. But the mechanisms they identify (synaptic depression, spike frequency adaptation, and recurrent connectivity) are general. It is thus plausible that their conclusions generalize beyond the auditory modality. I think the study is well conceived, its message well communicated, and the specific conclusions the authors make are well supported by the (model) data. The study demonstrates how the high biological fidelity modeling, that has been gaining traction in neuroscience, can serve as a testbed for rapid evaluation of hypothesis and elucidation of mechanism behind brain computation.

    That said, I have several major comments:

    1. I am concerned about the novelty/impact of the study. The impact of the present study can be viewed through two lenses:

    (a) The novelty and added value of the modelling approach itself. While I am very enthusiastic about the merits of the high fidelity modeling used in the present study, this modeling approach has now been well established across multiple manuscripts. The cortical model itself is already published, while I do not think the MGB extension of the model itself represents a significant advancement.

    (b) The impact of the findings of the study itself. The study claims one main novel finding: contribution of the SFA in combination with recurrent cortical connectivity to the SSA. The contribution of SFA to SSA doesn't seem particularly surprising, and as authors write it indeed has already been proposed. Also impact of recurrent connectivity on SSA has already been explored by a previous model (Yarden et al. 2014). Furthermore, my understanding is that the model was for the first time able to replicate the weaker presence of SSA in thalamo-cortical layers, and the dependence of SSA on frequency preference of the neuron. It is my understanding that all other replicated phenomena have already been demonstrated in previous models.

    1. I was surprised no comment was made on (a) the potential difference between the anatomy of the auditory cortical column in comparison to the somatosensory column, which the present model has been designed around, and (b) the lack of functionally specific connectivity, that at least in other sensory cortices (e.g. V1) has been shown to play an instrumental role in shaping the computation. This is particularly surprising in the context of the inability of the model to reproduce some of the interesting findings on SSA (distribution of SAA values in different cortical layers, specific deviance sensitivity), and on the other hand the level of optimism on the future of the model expressed in the last paragraphs of the discussion. I think for the modelling approach in future to fulfill such optimistic goals, both these major problems will have to be addressed, which represent a major body of new work - this should be acknowledged.

    2. I am concerned about the lack of functional verification of the model. Do for example the cortical neurons have frequency tuning curves characteristics that match well auditory data? Unfortunately, I am not an A1 expert, but I would expect wealth of data on elementary functional properties of A1 neurons exists. This represents somewhat of a paradox, where the model is at some level extremely detailed and well matched to experimental data, which (justifiably) authors sell as a major advantage. But it is surprisingly poorly validated against the elementary computations that A1 performs, which in the context of this study, is just as if not more important as the anatomical fidelity. I feel that, at minimum, this issue warrants thorough discussion, both in the context of the SAA, and the modelling approach itself.

  3. eLife

    ###Reviewer #1:

    This study investigates whether a detailed biophysical model of a cortical column, simulating more than 30,000 fully detailed neurons, is able to reproduce a well known property of the auditory cortex: stimulus specific adaptation or SSA. SSA has been successfully reproduced in a simplistic model which shows that adaptation mechanisms explain the qualitative phenomenology of this effect (decreased responsiveness for repeated stimuli, specific to the repeated sound and to sounds whose representation overlaps within the repeated sound). Here the authors aimed at testing whether without any parameter optimization, a detailed biophysical model is able to reproduce the observed phenomenon. As the model contains two well-known adaptation mechanisms, synaptic depression and spike frequency adaptation, unsurprisingly, a qualitative match between natural SSA and modeled SSA is observed. Moreover, effects related to representation overlap are found by including a mostly data-driven representation model and without fine tuning. Finally, the biophysical model suggests that both synaptic depression and spike frequency adaptation (SFA) contributes to SSA and that SFA exclusively contributes to the asymmetry of cross frequency adaptation with respect to the preferred frequency, that is both observed in the model and in the data, and can be explained by asymmetry of cochlear representations.

    This is a nice and important exercise to test the efficiency of a so-called detailed model at reproducing basic experimental observation. Unfortunately, here the model performs very well qualitatively but not quantitatively as little quantitative match is observed with spike data from auditory cortex (Figure 5). In fact there is little comparison with actual data, and this is disappointing. One of the purposes of detailed models is to identify their limitations and thereby identify useful details that may have been missed or incorrectly measured. Unfortunately, the quantitative mismatch in Fig. 5 is not mentioned in the results and no attempt is made to fill the gap. Hence, the conclusions of the paper do not go much beyond the well known role of adaptation and representation overlap. The identification of a measure to separate the two components, depression and SFA, is a nice contribution, but it is not tested experimentally, so it remains to be done (e.g. suppressing recurrencies by tetanus toxin light chain) to validate this hypothesis.

  4. eLife

    ##Preprint Review

    This preprint was reviewed using eLife’s Preprint Review service, which provides public peer reviews of manuscripts posted on bioRxiv for the benefit of the authors, readers, potential readers, and others interested in our assessment of the work. This review applies only to version 1 of the manuscript.

    ###Summary:

    All reviewers have acknowledged the value of a detailed model of auditory cortex, and expressed their support for an integrative approach building the link between neural circuits details and observables. It was found particularly interesting that two complementary mechanisms could play a role in stimulus specific adaptation (SSA). Nevertheless, while the reviewers recognized that the simulations were technically sound and that the conclusions represent interesting hypotheses to pursue about the mechanisms of SSA in auditory cortex, they all felt that the precision to which the specificity of auditory cortex circuits were modeled or to which the SSA observables were captured was not sufficient to demonstrate the advantage of the detailed modeling approach with respect to previous simpler models which reached similar conclusions.

  5. Version published to 10.1101/2020.05.31.126466 on bioRxiv