Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2023 Jun 30;7(2):743-768.
doi: 10.1162/netn_a_00311. eCollection 2023.

Sleep spindles in primates: Modeling the effects of distinct laminar thalamocortical connectivity in core, matrix, and reticular thalamic circuits

Arash Yazdanbakhsh  1   2   3 , Helen Barbas  2   3   4   5 , Basilis Zikopoulos  2   3   5   6
Affiliations

Sleep spindles in primates: Modeling the effects of distinct laminar thalamocortical connectivity in core, matrix, and reticular thalamic circuits

Arash Yazdanbakhsh et al. Netw Neurosci. .

Abstract

Sleep spindles are associated with the beginning of deep sleep and memory consolidation and are disrupted in schizophrenia and autism. In primates, distinct core and matrix thalamocortical (TC) circuits regulate sleep spindle activity through communications that are filtered by the inhibitory thalamic reticular nucleus (TRN); however, little is known about typical TC network interactions and the mechanisms that are disrupted in brain disorders. We developed a primate-specific, circuit-based TC computational model with distinct core and matrix loops that can simulate sleep spindles. We implemented novel multilevel cortical and thalamic mixing, and included local thalamic inhibitory interneurons, and direct layer 5 projections of variable density to TRN and thalamus to investigate the functional consequences of different ratios of core and matrix node connectivity contribution to spindle dynamics. Our simulations showed that spindle power in primates can be modulated based on the level of cortical feedback, thalamic inhibition, and engagement of model core versus matrix, with the latter having a greater role in spindle dynamics. The study of the distinct spatial and temporal dynamics of core-, matrix-, and mix-generated sleep spindles establishes a framework to study disruption of TC circuit balance underlying deficits in sleep and attentional gating seen in autism and schizophrenia.

Keywords: First-order thalamic nuclei; High-order thalamic nuclei; Laminar connectivity; Schizophrenia; Spindle dynamics; Spindle propagation.

PubMed Disclaimer

Figures

<b>Figure 1.</b>
Figure 1.
The proposed basic thalamocortical (TC) core, matrix, and mix circuit. The model circuit was based on two parallel thalamocortical loops. Core: PV+ excitatory thalamic neurons project focally to the middle cortical layers and get feedback from cortical layer 6. Matrix: CB+ thalamic neurons innervate broadly the superficial cortical layers and receive projections from cortical layer 5. Layer 5 terminals in the thalamus are larger (thick arrowheads) than terminals from layer 6. Thalamic and cortical neurons project to TRN (TRNC and TRNM). Mix: The two parallel loops cross-connect at the level of the thalamus and/or cortex. The mix dynamics were implemented in three different ways, indicated by dashed lines cross-linking core and matrix loops in panels A–C: (1) corticoreticular, where L6 and L5 concurrently excited TRNM and TRNC; (2) thalamoreticular, where PV+ and CB+ concurrently excited TRNM and TRNC; and (3) cortical, where mixing of core and matrix networks occurred at the level of the cortex, through columnar laminar interactions, simplified as a cross-connection between L5 and L6 (dashed line); this way, we could modify core (L6 and PV+) versus matrix (L5 and CB+) contribution in the mix. (A) In closed loop connections, TRN neurons directly inhibit the thalamic neurons that excite them. (B) The anatomical circuits underlying open loop connections are shown, in which TRN neurons do not directly inhibit the thalamic neurons that excited them. (C) In the model we simulated the open loop architecture through implementation of the closed loop connections in which TRN neurons directly inhibit the thalamic neurons that excited them, following a symmetric Gaussian spread with the peak at the reciprocal thalamic neuron. Accordingly, the bilateral and symmetric Gaussian connectivity strength from TRN to thalamus resembles bilateral and symmetric open loops with opposite open directions (balanced). We called this the Hybrid Loop design. Inhibitory model neurons are shown in red, including TRN (TRNC and TRNM for core and matrix), PV−, and CB− local thalamic inhibitory interneurons. The local inhibitory thalamic neurons PV− and CB− send GABAergic projections to thalamic projection neurons (PV+ and CB+; shown in panels B and C) and can receive excitatory input from the periphery, subcortical nuclei, including TRN, and cortical areas. For simplification of the schematic diagram, the local inhibitory thalamic neurons PV− and CB− in B are shown to receive inhibitory input only from TRN to highlight their potential role in open loop TRN-TC architecture, whereas in C they are shown to receive input from TRN or from cortex and they send inhibitory projections to TC neurons. Blue units are core excitatory elements, that is, neurons in cortical layers 4 and 6 (L3b–5a and L6), as well as thalamic PV+ neurons. Green units are matrix excitatory units, that is, neurons in cortical layer 5 (L5), whose apical dendrites reach upper layers (L1–3a) and receive widespread input from excitatory thalamic CB+ neurons. Note that in the primate thalamus, PV and CB stain cell bodies and dendrites of core and matrix TC projection neurons, respectively. There is some neuropil staining, especially PV axons and terminals, that come mainly from TRN and basal ganglia; however, the cellular label is predominant and clearly distinguishes the two systems. In several nuclei, for example, the ventral anterior, PV, and CB neurons are intermingled and at equal numbers, so it is not possible to define pure core or matrix regions (Zikopoulos & Barbas, 2007b). (D) When the Gaussian peak of connectivity strength is to the thalamic neuron that directly excites the TRN neuron, due to spatially symmetric bilateral Gaussian spread of connectivity strength, the circuit is composed of two open loops in opposite directions (balanced), then our architecture is functionally similar to a closed loop.
<b>Figure 2.</b>
Figure 2.
Model inputs and response. (A) Model TRN input resembling spindle burst temporal pattern. Spindle generation in the model is based on inducing TRN neurons by inputs resembling the spindle spike temporal patterns, which can entrain the model thalamocortical loop depending on the tendency of rebound depolarization and oscillation. Throughout the reported simulations, we chose a 10 Hz (100-ms interburst interval) spindle-pattern input with a 500-ms duration to induce the model TRN units. To set the TRN spindle inducing pattern as similar to a physiological spindle as possible, we considered physiological properties of TRN neurons (reviewed in McCormick & Bal, 1997) and approximated the initial membrane potential raise and then an accelerated upturn (T-current) before the spikes burst by a sixth-degree polynomial. We also approximated the bursts by densely packed spike-like pattern between 0 and 1 to have the model neuron activities within the normalized values throughout the simulation. (B) Model response to 0.5-sec tonic input to PV. Each layer neuron activity is amplitude scaled to prevent overlap for clear visibility. During the input onset, each model neuron depolarizes, reaches its peak, reverses, and after a few oscillations settles. After turning off the square input at 0.5 sec, the activities drop sharply toward baseline followed by hyperpolarization, rebound depolarization, and oscillations to finally settle on the resting state. Response to a 0.5-sec square input illustrates the dynamics of the rate-based model and its tendency for rebound depolarization. (C) Model response to a depolarizing input pulse. After PV depolarization, there is hyperpolarization and a rebound depolarization. (D) Model response to a hyperpolarizing input (due to a brief input pulse to TRN). PV neurons exhibit hyperpolarization first and then they rebound depolarize. (E) Model response to 0.5-sec depolarizing tonic input, while the network is in high spindle tendency, due to increased levels of TC inhibition by TRN, shows oscillatory activity of higher amplitude and longer duration compared to the response in panel B with the network in lower spindle tendency; here too, each layer neuron activity is amplitude scaled to highlight PV oscillations. (F) Band-pass filtering typically used to filter the signal in the relevant frequency bands; in this example, 5–15 Hz produces output similar to the original model output.
<b>Figure 3.</b>
Figure 3.
The effect of model TRN input resembles spindle temporal pattern. The model is based on direct and indirect evidence for the presence of L5 neuron projections to TRNM. (A) No direct L5 to TRNM results in STI < 0.1, which does not promote spindle tendency in the network. (B) In comparison, implementation of partial, small L5 to TRNM connection (equivalent to 5% of L6 to TRNC model connection strength) yields STI = 0.57, which shows a facilitating effect of direct (L5 to TRNM) connection in promoting spindle tendency of the network. (C) the efficacy of L5-TRN connection is further increased from 5% to 10% resulting in STI > 1 that can contribute to the elongation of spindling time and can facilitate continuous or repetitive spindles instead of isolated ones.
<b>Figure 4.</b>
Figure 4.
Spindle wave propagation across model neuron arrays. (A) The activity of 200 model TC neurons (PV+) over time. Yellow, depolarization; green, resting potential; and blue, hyperpolarization. Model recording sites are indicated by 1, 2, 3, and 4 corresponding to neurons 100, 113, 128, and 147, respectively. (B) From top to bottom, four recording locations 1–4 from the model neurons are shown. Similar to the Kim et al. (1995) recording simultaneously from multiple sites in the ferret LGNd, spindle oscillations propagate across locations 1–4 in the model. Each subpanel of panel B is normalized within its range, and therefore the depolarization and hyperpolarization colors are identical, merely showing whether the membrane potential is depolarized or hyperpolarized. Therefore, the aim of panels 1–4 is to indicate the voltage change direction either toward depolarization or hyperpolarization independent of the color map magnitude, which was adjusted to enhance visibility.
<b>Figure 5.</b>
Figure 5.
Impact of local thalamic inhibition in biasing the network toward sustaining or terminating spindles. (A) shows the dynamics of core model neurons with the absence of local inhibition (no PV− inhibition of PV+), and (B) with the presence of local inhibition. The spindle-like input (shown in Figure 2) is ON for 0.5 sec starting at time 0 and turned OFF at time 0.5 sec. By tracing the behavior of the model TC neurons, we evaluate the tendency of the network to sustain or terminate spindles. Tracing the activity of PV and L4 neurons in panel A and B on and after 0.5 sec indicates longer duration and higher amplitude of hyperpolarization and rebound polarization in panel B compared to A, indicating a higher spindle probability in the model neurons of TC loop with the presence of local inhibition in thalamus.
<b>Figure 6.</b>
Figure 6.
Impact of TRN inhibition and cortical feedback to thalamus in biasing the network toward sustaining or terminating spindles. (A) The dynamics of core model neurons without cortical feedback to thalamus (STI = 0.55) compared to (B) with cortical feedback to thalamus (STI < 0.55), indicating a decreased spindling tendency. (C) Similar to panel A, but with increased TRN inhibition of PV, which increased STI to >1 compared to panel A with STI = 0.55; (D) Similar to panel C, but with cortical feedback to thalamus, which reduced STI to 0.55 from >1 in C.
<b>Figure 7.</b>
Figure 7.
Core, Matrix, and mixed TC model spatiotemporal dynamics. (A) Only model L6, TRNC, and PV (core TC loop) were involved in spindle entrainment, induced by inputting spindle-like input (Figure 2) to TRNC, while L5, TRNM, and CB (matrix) remained inactive. (B) The mix (50–50% thalamoreticular mixing) of core and matrix, engaged all model areas. (C) Only model L5, TRNM, and CB (matrix TC loop) were involved in spindle entrainment, induced by inputting spindle-like input (Figure 2) to TRNM, while L6, TRNC, and PV (core) remained inactive. Spindle-like input (Figure 2), was delivered to neuron no. 100 (y-axis) in the TRNC and TRNM array of model neurons, lasting for 500 ms and was then shut down. The presence of color map ripples in the figure panels are equivalent to sequences of hyperpolarization and depolarization events shown in previous figures and indicate spindle tendency of the network (beyond 500 ms based on the TC model dynamics). Comparison of corresponding panels in A and C revealed a broader spatial spread and temporal extent in matrix than core. Notably, the mix circuit spatiotemporal dynamics of spindles in each site was a hybrid of core and matrix spindles in terms of broadness of spatial smear and temporal extent. In panel A, the model core cortical layer 4, L4, showed a focal tendency for spindle sustaining indicated by a curved dotted marking. Compare with panel C model matrix cortical layer 5 (L5), in which the spindle tendency across model neurons after propagation appeared synchronized and therefore the depolarization iso-amplitude traces aligned vertically at each time bin after spindle propagation for a few hundred milliseconds. A vertical dotted mark as an example is shown in panel C, L5. To further compare the effects of mixing, we set 50–50% mixing level in panel B as the baseline (angle θ: ArcCos of cosine similarity = 0°). Comparison of θ values in panel B with pure core in A and pure matrix in C provides an estimate of the blending effects, expressed as rising deviation of the angle θ from 0°. Higher deviation of the angle θ from 0° indicated increased dissimilarity from B.
<b>Figure 8.</b>
Figure 8.
Spatiotemporal dynamics of cortical mixed TC model with different ratios of involved core and matrix. (A–D) The spatiotemporal dynamics of TC neuronal activity with different ratios of mixed core/matrix: 80/20, 60/40, 40/60, and 20/80 are illustrated, with mixing occurring at the level of the cortex. The higher the ratio of core/matrix, the more core-like were the spatiotemporal dynamics, that is, focal in space and less smeared over time. Conversely, the higher the ratio of matrix/core, the more matrix-like were the spatiotemporal dynamics, that is, more diffuse in space and smeared over time. This trend could be seen in all (in particular core) regions (L4, L6, TRNC, and PV) involved in the mix TC loop; the spatiotemporal dynamics of matrix specific regions (L5, TRNM, and CB), more or less, remained the same, that is, diffuse and smeared across different ratios of mixing, as if matrix kept its diffuse spatiotemporal dynamics due to its wider horizontal (within cortical laminae) connectivity and depending on the ratio of core/matrix in the mix, the activity of core regions could be more or less matrix-like. To further compare the effects of mixing, we set 50–50% mixing level in panel C as the baseline (angle θ: ArcCos of cosine similarity = 0°). Comparison of θ values between panels provides an estimate of the magnitude of blending effects, expressed as rising deviation of the angle θ from 0°. Higher deviation of the angle θ from 0° indicated increased dissimilarity from the baseline in panel C.
<b>Figure 9.</b>
Figure 9.
Spatiotemporal dynamics of TC model with different synaptic efficacy of cortical mixing of core and matrix. The spatiotemporal dynamics of the TC model plateaued after reaching increased synaptic efficacy in cortical mixing. The plateau synaptic efficacy being used was 30% (A), 60% (B), and 90% (C). Similar to Figure 8, matrix model neurons kept their spatiotemporal dynamics, and the response of core model neurons resembled the response of matrix model neurons as the efficacy of mixing gradually increased. Model core cortical layer 4 (L4), in 90% mixing level (C), compared to 30% mixing level (A), showed faster activity propagation across neurons indicated by higher slope (i.e., number of neurons per millisecond) of red dotted marks. This shows that mixing core with matrix impacts core signal propagation rate. On the other hand, matrix fast propagation rate remained relatively the same across different mixing levels of core and matrix, indicated by the same slope of red dotted marks in panels A and C L5 panels. To further compare the effects of mixing, we set cortical mixing level in panel A as the baseline (30% of the plateau of the synaptic efficacy), with an estimated angle θ (ArcCos of cosine similarity) = 0°. Comparison of θ values in panel A with B and C provides an estimate of the cumulative blending effects of gradually increasing mixing levels in B and C, expressed as rising deviation of the angle θ from 0°. Higher deviation of the angle θ from 0° indicated incremental increasing dissimilarity of B and C from A.
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

References

    1. Agmon, A., & Connors, B. W. (1992). Correlation between intrinsic firing patterns and thalamocortical synaptic responses of neurons in mouse barrel cortex. Journal of Neuroscience, 12(1), 319–329. 10.1523/JNEUROSCI.12-01-00319.1992, - DOI - PMC - PubMed
    1. Allen, G., & Courchesne, E. (2001). Attention function and dysfunction in autism. Frontiers in Bioscience, 6, D105–D119. 10.2741/allen, - DOI - PubMed
    1. Arcelli, P., Frassoni, C., Regondi, M. C., De Biasi, S., & Spreafico, R. (1997). GABAergic neurons in mammalian thalamus: A marker of thalamic complexity? Brain Research Bulletin, 42(1), 27–37. 10.1016/S0361-9230(96)00107-4, - DOI - PubMed
    1. Baran, B., Karahanoglu, F. I., Mylonas, D., Demanuele, C., Vangel, M., Stickgold, R., … Manoach, D. S. (2019). Increased thalamocortical connectivity in schizophrenia correlates with sleep spindle deficits: Evidence for a common pathophysiology. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging, 4(8), 706–714. 10.1016/j.bpsc.2019.04.012, - DOI - PMC - PubMed
    1. Barbaresi, P., Spreafico, R., Frassoni, C., & Rustioni, A. (1986). GABAergic neurons are present in the dorsal column nuclei but not in the ventroposterior complex of rats. Brain Research, 382(2), 305–326. 10.1016/0006-8993(86)91340-5, - DOI - PubMed

LinkOut - more resources