Consciousness depends on integration between parietal cortex, striatum, and thalamus

Abstract

The neural substrates of consciousness remain elusive. Competing theories that attempt to explain consciousness disagree on the contribution of frontal versus posterior cortex and omit subcortical influences. This lack of understanding impedes the ability to monitor consciousness, which can lead to adverse clinical consequences. To test substrates and measures of consciousness, we recorded simultaneously from frontal cortex, parietal cortex, and subcortical structures, the striatum and thalamus, in awake, sleeping, and anesthetized macaques. We manipulated consciousness on a finer scale using thalamic stimulation, rousing macaques from continuously administered anesthesia. Our results show that, unlike measures targeting complexity, a measure additionally capturing neural integration (Φ^∗) robustly correlated with changes in consciousness. Machine learning approaches show parietal cortex, striatum, and thalamus contributed more than frontal cortex to decoding differences in consciousness. These findings highlight the importance of integration between parietal and subcortical structures and challenge a key role for frontal cortex in consciousness.

Keywords: anesthesia; basal ganglia; central thalamus; complexity; consciousness; frontal cortex; integration; phi; posterior parietal cortex; sleep.

Figure 1.. Parietal deep layers and subcortical areas contribute most to state decoding.

(A) Normalized average delta power (Z score ± SE) for thalamus (T), caudate (C), and superficial (s), middle (m) and deep (d) layers of frontal (F) and parietal (P) cortex during wakefulness, sleep and anesthesia (Anes). (B) Population mean Φ* (± SE) during different states for the full system (All) or subsystems with one area removed. (C) Decoding accuracy (± SE) for classifiers using LFP power, Φ*, mutual information (I) or entropy (H). Dashed line shows chance. (D) Confusion matrix of LFP power classifier for wakefulness (W), sleep (S), and anesthesia (A); color scales with classification likelihood. (E) Confusion matrix of Φ* classifier. (F-H) Mean decrease in accuracy (MDA ± SE) for LFP power classifier after removing features for (F) brain areas or (G) frequency bands, or (H) Φ* classifier after removing subsystems containing specified brain area. See also Figures S1–S3, and Tables S1–S3.

Figure 2.. Parietal deep layers and subcortical areas contribute most to decoding stimulation-induced consciousness.

(A and B) Normalized average delta power (Z score ± SE) for caudate (C) and superficial (s), middle (m) and deep (d) layers of frontal (F) and parietal (P) cortex prior to, during, and after (A) effective or (B) control thalamic stimulations. (C and D) Population mean Φ* (± SE) for the full system (All) or subsystems with one area removed surrounding (C) effective or (D) control stimulations. (E) Decoding accuracy (± SE) for classifiers using LFP power, Φ*, mutual information (I) or entropy (H). Dashed line shows chance. (F) Confusion matrix of LFP power classifier for effective (Effect) or control stimulations; color scales with classification likelihood. (G) Confusion matrix of Φ* classifier. (H-J) Mean decrease in accuracy (MDA ± SE) for LFP power classifier after removing features for (H) brain areas or (I) frequency bands, or (J) Φ* classifier after removing subsystems containing specified brain area. See also Figures S1–S3, and Tables S1, S3, and S4.

Figure 3.. Only Φ* correlates with fine, stimulation-induced changes in consciousness level.

(A) Correlations (r) of normalized changes in power (Z score, during stim – pre) and changes in consciousness level (during stim – pre) for each brain area at each frequency band. (B-H) Regression estimates of delta power correlations in (A) for (B) caudate nucleus, (C) deep, (D) middle, and (E) superficial parietal layers, as well as (F) deep, (G) middle, and (H) superficial frontal layers. Circles show values for each stimulation event. Line shows regression fit (± SE) with reported slope (β) and correlation (shaded background, r, on same scale as (A)). (I) Regression estimate for stimulation-induced consciousness level changes (during stim – pre) on Φ*; * on β slope indicates p = 9.0×10⁻¹⁵. See also Figures S3 and S4, and Table S5.

Figure 4.. Integration between parietal deep layers and subcortical areas contribute most to increases in Φ* and changes in consciousness.

(A) Average contribution of each brain area to Φ* (± SE) controlling for changes in conscious state. (B) Estimated effect size (ΔR²) of each brain area in (A) and consciousness (yellow, wake vs. sleep/anesthesia). (C) Average contribution of each brain area to Φ* (± SE) controlling for stimulation-induced changes in consciousness. (D) Estimated effect size (ΔR²) of each brain area in (C) and consciousness (yellow, effective vs. pre/post/control stimulations). (E-K) Pairwise probability (gray-scale lines) of brain areas associating on the same side of the minimum information partition for (E) wakefulness, (F) anesthesia, (G) sleep, (H) pre-stimulation, (I) effective stimulation, (J) control stimulation and (K) post-stimulation. Monosynaptic (Mono, thick lines) anatomical paths (in at least one direction) and multisynaptic (Multi, thin lines) shown. Dashed red lines show predominant minimum information partition when applicable. (L) Schematic showing pathways and brain areas contributing most to integration and classification of consciousness. See also Figure S5 and Table S6.