Dexmedetomidine disrupts the local and global efficiency of large-scale brain networks

Abstract

Background

A clear understanding of the neural basis of consciousness is fundamental to research in clinical and basic neuroscience disciplines, and anesthesia. Recently, decreased efficiency of information integration was suggested as a core network feature of propofol-induced unconsciousness. However, it is unclear whether this finding can be generalized to dexmedetomidine, which has a different molecular target.

Methods

Dexmedetomidine was administered as a 1 μg/kg bolus over 10 minutes, followed by a 0.7 μg/kg/hr infusion to healthy human volunteers, 18–36 years of age (n = 15). Resting state functional magnetic resonance (rsfMRI) data were acquired during baseline, dexmedetomidine-induced altered arousal, and recovery states. Zero-lag correlations between rsfMRI signals extracted from 131 brain parcellations were used to construct weighted brain networks. Network efficiency, degree distribution, and node strength were computed using graph analysis. Parcellated brain regions were also mapped to known resting state networks to study functional connectivity changes.

Results

Dexmedetomidine significantly reduced local and global efficiency of graph theory derived networks. Dexmedetomidine also reduced the average brain connectivity strength without impairing the degree distribution. Functional connectivity within and between all resting state networks was modulated by dexmedetomidine.

Conclusions

Dexmedetomidine is associated with a significant drop in the capacity for efficient information transmission at both local and global levels. These changes result from reductions in the strength of connectivity, and also manifest as reduced within and between resting state network connectivity. These findings strengthen the hypothesis that conscious processing relies on an efficient system of information transfer in the brain.

Introduction

Understanding the neural basis of consciousness is fundamental to research in clinical and basic neuroscience disciplines, and anesthesia.¹ However, we are yet to clearly decipher how the human brain mediates consciousness. This is because the brain is a complex biological system composed of components that interact dynamically to give rise to higher brain functions.^2,3 For example, the sensory experience of a painfully hot object comprises of various composite experiences⁴ (i.e. intensity, duration, quality, emotional distress) that are integrated by the brain and represented as a unified experience of pain.^2,3,5-9

Synchronization of resting state functional MRI (rsfMRI) slow signals is a proxy for the putative neural syntax that integrates various brain regions into networks mediating higher brain functions.^10-13 Numerous investigations of anesthesia-induced changes to rsfMRI networks have been conducted. For a review, see Hudetz, 2012.¹⁴ However, less studied is how anesthetics alter the capacity of information transfer in rsfMRI brain networks, and how this relates to the brain state in question. This graph-theoretic approach, which is distinct from standard functional connectivity analyses, is based on the fundamental premise that the brain is topologically organized to maximize information transfer within and between networks.^15,16 This approach parallels theories of consciousness that stress integrated information processing.^8,9

Network efficiency is a graph theoretical measure of information exchange. A recent study of the gamma amino-butyric acid receptor agonist propofol described decreased efficiency of information transfer as a key differentiating feature of conscious and unconscious brain states.¹⁷ Similarly, decreased efficiency of information transfer has also been suggested to account for decreased consciousness during non-rapid eye movement sleep.^18,19 Thus, altered states of arousal may be associated with network reconfigurations that favor decreased efficiency of communication. However, the generalizability of this finding is unclear. Thus, studies of anesthetics from different drug classes are necessary to help refine our knowledge of information processing correlates of unconsciousness.

Using a graph theoretical approach, we studied network changes associated with dexmedetomidine, an alpha-2 adrenergic agonist that activates endogenous sleep pathways.^20-24 We hypothesized that dexmedetomidine would result in decreased capacity for information integration as represented by surrogate markers of local and global network communication (local and global efficiency). Furthermore, we hypothesized that this decrease in information integration results from reduced synchronization strength between brain regions. In previous work we found that dexmedetomidine preferentially decreased blood flow and metabolism in the thalamus, Default Mode Network, and bilateral Frontoparietal Networks.¹ Therefore, we also hypothesized that between-network disruption in these and other RSNs are associated with dexmedetomidine.

To explore these hypotheses, blood oxygen level dependent (BOLD) signals obtained during baseline, dexmedetomidine-induced altered arousal, and recovery states were analyzed (n = 15, 18 to 36 years of age). Zero-lag correlations between rsfMRI signals extracted from 131 brain parcellations were used to construct weighted brain networks. Network efficiency and node strength were computed using graph analysis. Parcellated brain regions were also mapped to known RSNs (Table 1). This dataset was previously reported in a within-network analysis of the Default Mode and Frontoparietal Networks.¹

Table 1.

Regions spanning the right and left hemispheres were separated to create hemisphere specific regions. The complete parcellation scheme consisted of a total of 131 regions (Table 1). A list of these regions, their co-ordinates, as well as a description of the anatomical landmarks used for the subdivision of the Harvard-Oxford labels in smaller parcellates have previously been published.

Networks	Regions	Abbreviations	MNI Co- ordinates
			x	y	z
1	Brain stem	BrStem	0	−26	−28
1	Amygdala Right	Amyg_R	24	−4	−18
1	Caudate Right	Caud_R	12	12	10
1	Globus Pallidus Right	GP_R	16	−2	−2
1	Hippocampus Right	Hipp_R	28	−22	−16
1	Nucleus Accumbens Right	NAc_R	10	10	−8
1	Putamen Right	Put_R	20	−4	0
1	Thalamus Right	Thal_R	10	−18	8
1	Thalamus Left	Thal_L	10	−18	8
1	Putamen Left	Put_L	−20	−4	0
1	Nucleus Accumbens Left	NAc_L	−10	10	−8
1	Hippocampus Left	Hipp_L	−28	−22	−16
1	Globus Pallidus Left	GP_L	−16	−2	−2
1	Caudate Left	Caud_L	−12	14	8
1	Amygdala Left	Amyg_L	−24	−4	−18
2	Central Opercular Cortex Right	Cop_R	48	−4	8
2	Dorsal Anterior Insula Right	dINSa_R	32	20	0
2	Middle Insula Right	INSm_R	40	−2	−2
2	Posterior Insula Right	INSp_R	38	−14	8
2	Postcentral Gyrus Right	PostC_R	54	−20	46
2	Precentral Gyrus Right	PreC_R	44	−8	52
2	Supra Marginal Gyrus Right	SMGa_R	58	32	40
2	Ventral Anterior Insula Right	vINSa_R	36	10	−14
2	Temporal Occipital Fusiform Cortex Right	TOF_R	34	−54	−16
2	Temporal Fusiform Cortex, posterior division Right	TFCp_R	36	−16	−32
2	Planum Polare Right	PlP_R	48	−4	−6
2	Planum Temporale Right	PlT_R	60	−22	8
2	Heschls Gyrus (includes H1 and H2) Right	He_R	48	−18	−6
2	Cuneal Cortex Right	Cun_R	4	−82	30
2	Intracalcarine Cortex Right	IC_R	6	−68	12
2	Lateral Occipital Cortex, inferior division Right	LOcci_R	48	−78	−2
2	Lateral Occipital Cortex, superior division Right	LOccs_R	40	−78	34
2	Occipital Fusiform Gyrus Right	OccFG_R	28	−76	−14
2	Occipital Pole Right	OccP_R	8	−100	6
2	Precuneous Cortex Right	pCun_R	4	−64	38
2	Supracalcarine Cortex Right	Sc_R	2	−84	12
2	Supracalcarine Cortex Left	Sc_L	−2	−84	12
2	Precuneous Cortex Left	pCun_L	−4	−82	30
2	Occipital Pole Left	OccP_L	−8	−100	6
2	Occipital Fusiform Gyrus Left	OccFG_L	−28	−76	−14
2	Lateral Occipital Cortex, superior division Left	LOccs_L	−40	−78	34
2	Lateral Occipital Cortex, inferior division Left	LOcci_L	−48	−78	−2
2	Intracalcarine Cortex Left	IC_L	−28	−76	−14
2	Cuneal Cortex Left	Cun_L	−6	−74	12
2	Heschls Gyrus (includes H1 and H2) Left	He_L	−48	−18	6
2	Planum Temporale Left	PlT_L	−60	−22	8
2	Planum Polare Left	PlP_L	−48	−4	−6
2	Temporal Fusiform Cortex, posterior division Left	TFCp_L	−36	−16	−32
2	Temporal Occipital Fusiform Cortex Left	TOF_L	−34	−54	−16
2	Ventral Anterior Insula Left	vINSa_L	−36	10	−14
2	Supra Marginal Gyrus Left	SMGa_L	−54	−56	26
2	Precentral Gyrus Left	PreC_L	−44	−8	52
2	Postcentral Gyrus Left	PostC_L	−58	32	40
2	Posterior Insula Left	INSp_L	−38	−14	8
2	Middle Insula Left	INSm_L	−40	−2	−2
2	Dorsal Anterior Insula Left	dINSa_L	−32	20	0
2	Central Opercular Cortex Left	Cop_L	−48	−4	8
3	Caudal Anterior Cingulate Right	ACCc_R	4	40	−2
3	Rostral Anterior Cingulate mid posterior Right	ACCrm_R	−6	60	8
3	Rostral Anterior Cingulate posterior Right	ACCrp_R	−6	−2	42
3	Rostral Anterior Cingulate Right	ACCr_R	2	28	18
3	Subgenual Anterior Cingulate Right	ACCsg_R	4	16	−14
3	Dorsal Medial Prefrontal Cortex, anterior division Right	dMPFCa_R	4	50	28
3	Frontal Orbital Cortex Right	FO_R	40	20	4
3	Medial Prefrontal Cortex Right	MPFC_R	6	60	8
3	Ventral Medial Prefrontal Cortex Right	vMPFC_R	4	50	−20
3	Angular Gyrus Right	Ang_R	54	−56	26
3	Cingulate Gyrus, posterior division Right	Cingp_R	4	−38	32
3	Cingulate Gyrus, posterior division Left	Cingp_L	−4	−38	32
3	Angular Gyrus Left	Ang_L	−48	−4	8
3	Ventral Medial Prefrontal Cortex Left	vMPFC_L	−4	50	−20
3	Medial Prefrontal Cortex Left	MPFC_L	−6	60	8
3	Frontal Orbital Cortex Left	FO_L	−40	30	−14
3	Dorsal Medial Prefrontal Cortex, anterior division Left	dMPFCa_L	−4	50	28
3	Subgenual Anterior Cingulate Left	ACCsg_L	−2	28	18
3	Rostral Anterior Cingulate posterior Left	ACCrp_L	6	18	34
3	Rostral Anterior Cingulate mid posterior Left	ACCrm_L	20	28	18
3	Rostral Anterior Cingulate Left	ACCr_L	−6	18	34
3	Caudal Anterior Cingulate Left	ACCc_L	−4	40	−2
4	Frontal Pole Right	FP_R	30	54	20
4	Orbito Frontal Pole Right	OFP_R	32	58	−6
4	Superior Frontal Gyrus Right	SFG_R	20	18	62
4	Middle Frontal Gyrus Right	MFG_R	40	20	44
4	Supramarginal Gyrus, posterior division Right	SMGp_R	60	−48	32
4	Superior Parietal Lobule Right	SPL_R	32	−50	60
4	Mid Anterior Cingulate Right	ACCm_R	6	−2	42
4	Dorsal Medial Prefrontal Cortex, posterior division Right	dMPFCp_R	4	26	48
4	Frontal Operculum Cortex Right	Fop_R	−48	−32	20
4	Inferior Frontal Gyrus, pars opercularis Right	IFGpo_R	54	14	16
4	Inferior Frontal Gyrus, pars triangularis Right	IFGpt_R	−50	30	16
4	Parietal Operculum Cortex Right	Pop_R	48	−32	20
4	Supplementary Motor Area Right	SMA_R	4	−2	58
4	Supplementary Motor Area Left	SMA_L	−4	−2	58
4	Parietal Operculum Cortex Left	Pop_L	−38	4	0
4	Inferior Frontal Gyrus, pars triangularis Left	IFGpt_L	−50	30	16
4	Inferior Frontal Gyrus, pars opercularis Left	IFGpo_L	−54	−20	46
4	Inferior Frontal Gyrus, pars opercularis Left	Fop_L	40	30	−14
4	Dorsal Medial Prefrontal Cortex, posterior division Left	dMPFCp_L	−4	26	48
4	Mid Anterior Cingulate Left	ACCm_L	−4	50	−20
4	Superior Parietal Lobule Left	SPL_L	−60	−48	32
4	Supramarginal Gyrus, posterior division Left	SMGp_L	−40	20	4
4	Middle Frontal Gyrus Left	MFG_L	−40	20	44
4	Superior Frontal Gyrus Left	SFG_L	−22	22	54
4	Orbito Frontal Pole Left	OFP_L	−32	58	−6
4	Frontal Pole Left	FP_L	−30	54	20
5	Temporal Pole Right	TP_R	40	16	−30
5	Superior Temporal Gyrus, anterior division Right	STGa_R	58	−4	−6
5	Superior Temporal Gyrus, posterior division Right	STGp_R	66	−26	6
5	Temporal Fusiform Cortex, anterior division Right	TFCa_R	32	−6	−42
5	Parahippocampal Gyrus, anterior division Right	pHippa_R	34	−6	−34
5	Parahippocampal Gyrus, posterior division Right	pHippp_R	34	−32	−18
5	Inferior Temporal Gyrus, anterior division Right	ITGa_R	50	−6	−40
5	Inferior Temporal Gyrus, posterior division Right	ITGp_R	56	−32	−24
5	Inferior Temporal Gyrus, temporooccipital part Right	ITGtp_R	56	−54	−18
5	Lingual Gyrus Right	Ling_R	10	−68	−2
5	Middle Temporal Gyrus, anterior division Right	MTGa_R	58	−2	−22
5	Middle Temporal Gyrus, posterior division Right	MTGp_R	66	−22	12
5	Middle Temporal Gyrus, temporooccipital part Right	MTGto_R	60	−52	0
5	Middle Temporal Gyrus, temporooccipital part Left	MTGto_L	−60	−52	0
5	Middle Temporal Gyrus, posterior division Left	MTGp_L	−66	−22	12
5	Middle Temporal Gyrus, anterior division Left	MTGa_L	−58	−2	−22
5	Lingual Gyrus Left	Ling_L	−10	−68	−2
5	Inferior Temporal Gyrus, temporooccipital part Left	ITGtp_L	−56	−54	−18
5	Inferior Temporal Gyrus, posterior division Left	ITGp_L	−56	−32	−24
5	Inferior Temporal Gyrus, anterior division Left	ITGa_L	−50	−6	−40
5	Parahippocampal Gyrus, posterior division Left	pHippp_L	−34	−32	−18
5	Parahippocampal Gyrus, anterior division Left	pHippa_L	−34	−6	−34
5	Temporal Fusiform Cortex, anterior division Left	TFCa_L	−32	−6	−42
5	Superior Temporal Gyrus, posterior division Left	STGp_L	−66	−26	6
5	Superior Temporal Gyrus, anterior division Left	STGa_L	−58	−4	−6
5	Temporal Pole Left	TP_L	−40	16	−30

MNI, Montreal Neurological Institute

Materials and Methods

Imaging Visit

The Human Research Committee at the Massachusetts General Hospital approved this study (Protocol #: 2011P002333; NCT01485380). Written informed consent was obtained after the nature and possible study consequences of the study were explained to each healthy volunteer, 18–36 years of age (n=15). All volunteers were required to be American Society of Anesthesiologists Physical Status I. Brain imaging was performed with the Biograph mMR scanner (Siemens Healthcare, Erlangen, Germany), which allows simultaneous acquisition of whole-body PET and 3 Tesla MRI data. At the beginning of the imaging visit, structural MRI (MPRAGE volume, TR/TE = 2100/3.24ms, flip angle = 7°, voxel size = 1mm isotropic) was acquired for the purpose of anatomical localization, and spatial normalization of the imaging data. BOLD fMRI data were collected using a whole brain T2*-weighted gradient echo BOLD echo planar imaging pulse sequence was used (TR/TE =3000/35ms, flip angle=90°, voxel size=2.3×2.3×3.8mm, number of slices=35).

Dexmedetomidine was administered as a 1mcg/kg loading bolus over 10 minutes, followed by a 0.7mcg/kg/hr infusion. During the infusion period, an anesthesiologist monitored cuff blood pressure, capnography, electrocardiogram and pulse-oximetry. Volunteers were instructed to keep their eyes open during the course of the study. During start of the dexmedetomidine infusion (i.e. bolus), a 20-minute “induction” pulsed arterial spin label (pASL) scan was acquired. Altered arousal was defined as the onset of sustained eye closure/lack of response to a verbal request to open the eyes at 1-minute intervals during the induction pASL scan. Sustained eye closure and lack or response to verbal stimuli was confirmed in all subjects during the induction pASL. The dexmedetomidine-induced altered arousal BOLD rsfMRI scans were acquired after the induction pASL scan. At the conclusion of the dexmedetomidine infusion, a 20-minute “recovery” pulsed arterial spin label (pASL) scan was also acquired. During this period, arousal was also periodically assessed verbally at 1-minute intervals. Spontaneous eye opening and a positive response to give a thumbs-up signal were used to determine recovery. Arousability to the verbal stimulus was confirmed in all subjects during the 20-minute recovery pASL scan. The recovery BOLD rsfMRI scan was acquired after the pASL scan. The resting state nature of our data acquisition precluded continuous assessments of arousal during rsfMRI BOLD data acquisition.

Data Preprocessing

Both FMRIB Software Library v5.0 (FSL) and Analysis of Functional Neuro-Images software were used to preprocess data in line with procedures adapted for the 1000 Functional Connectomes project.²⁵ Data were slice time corrected for interleaved acquisitions using Fourier interpolation, motion corrected using least squares alignment of each volume to the eighth image using Fourier interpolation, despiked of extreme time series outliers using a continuous transformation function, temporal band-pass filtered between 0.009–0.3 Hz using Fourier transformation, and further filtered to remove linear and quadratic trends using AFNI. In addition, FSL was used for spatially smoothing the images (Gaussian kernel FWHM = 6 mm), and for normalizing mean-based intensity by the same factor (10,000). Next, eight nuisance signals (time courses of white matter and cerebrospinal fluid, and six motion parameters) were used as regressors of no interest. White matter and cerebrospinal fluid timeseries were extracted from masks obtained by segmenting each individual's high-resolution structural image using FMRIB's Automated Segmentation Tool, thresholded at 80% tissue type probability. The six motion parameters were generated in the FSL based motion-correction step. These six vectors included rotational movement around three axes (pitch, yaw, and roll) and movement in each of the three cardinal directions (X, Y, and Z). All of these steps were conducted in native functional space.

For registration, FMRIB's Linear and Non LINEAR Image Registration Tools were used for transformations from native functional and structural space to the Montreal Neurological Institute MNI152 template with 2×2×2 mm resolution. First, the high-resolution structural image was registered to the MNI152 2mm template with a 12-degree-of-freedom linear affine transformation. The transformation was further refined using FMRIB's Non LINEAR Image Registration Tool. Next, each participant's functional data were registered to their high-resolution structural image using a linear transformation with 6 degrees of freedom. The structural-to-standard nonlinear transformation matrix was used to register the functional volume to MNI152 standard space.

Brain parcellation and time course extraction

The brain was anatomically parcellated using a previously published parcellation scheme.²⁶ Briefly, the Harvard Oxford Atlas was refined by increasing the anatomical partitioning of the cingulate, medial and lateral prefrontal cortices. In addition, anatomical partitioning of the insular label was also performed. Thus, instead of the single ROI spanning the entire insula in the Harvard Oxford Atlas, the insula was subdivided into posterior, middle, dorsal anterior, and ventral anterior regions based on a previously published scheme.²⁷ Regions spanning the right and left hemispheres were separated to create hemisphere specific regions. The complete parcellation scheme consisted of a total of 131 regions (Table 1). A list of these regions, their co-ordinates, as well as a description of the anatomical landmarks used for the subdivision of the Harvard-Oxford labels in smaller parcellates have previously been published.²⁶ Each region of interest was designated as a node. Thus from each node, the BOLD time series were extracted and averaged to generate a 131 time series for each subject.

RSN Network construction

Each parcellated brain region or node was classified as belonging to a designated RSN based on spatial overlap with a specific RSN map. The RSN maps were identified using functional connectivity maps in the neurosynth framework (last accessed 2/15/16).²⁸ Each region was represented in only one RSN, based on maximum overlap.

Graph Analysis

Whole brain networks were constructed and network measures were assessed using the Brain Connectivity Toolbox. Formulae used for calculating network measures has been previously described.²⁹ For each patient, the BOLD time series in each region was correlated with every other region to create a 131 × 131 weighted connectivity or adjacency matrix. The adjacency matrices were thresholded at connectivity densities of 0.05, 0.1, 0.15, 0.2, 0.25 and 0.3.

The following graph functions were computed from each matrix:

Strength

This function measures the strength of connections in the graph. Node strength is the sum of weights of links connected to the node.

$$\text{str}=\text{sum}\left(\text{CIJ}\right)$$

where CIJ is the undirected weighted connection matrix. Mean connectivity strength of the graph was measured by averaging str from all nodes.

Global and local efficiencies

Efficiency is a measure of the network’s capacity for parallel information transfer between nodes through multiple series of edges. The average global efficiency of information transfer in graph $\mathcal{G}$ having n nodes can be calculated from the inverse of the edge distances d_i,j

$$E_{\text{glob}}=E\left(\mathcal{𝒢}\right)=\frac{1}{n(n-1)}\sum _{i\ne j;v_i,v_j\in \mathcal{𝒢}}\frac{1}{d_{ij}}$$

The quantity above is a measure of the global efficiency of information transfer for the whole graph $\mathcal{𝒢}$. There is also a local efficiency for each vertex v_i measuring how efficiently its neighbors can communicate when vertex v_i is removed. If the subgraph of all neighbors of v_i is denoted by ${\mathcal{𝒢}}_i$, then its local efficiency $E\left({\mathcal{𝒢}}_i\right)$ is approximately equivalent to the clustering coefficient C_i.³⁰

$$E_{\text{loc}}=\frac{1}{n}\sum _{v_i\in \mathcal{𝒢}}E\left({\mathcal{𝒢}}_i\right)$$

Statistical comparisons

Comparison between local and global efficiency and mean connectivity strength between the three states was first conducted using repeated measures ANOVA (Table 2). Post-hoc comparisons were conducted on significant ANOVA findings using the paired t-test. Significance was set at p < 0.05 and a Bonferroni correction for multiple comparisons was employed. Statistics on graph results and graphical presentations of networks were performed with custom code written in MATLAB.

Table 2.

Comparison between local and global efficiency and mean connectivity strength between the three states was first conducted using repeated measures ANOVA

Threshold		Mean	STD	F-value	P-value
Mean local efficiency
0.05	Awake	0.501	0.040	0.499	0.611
	Dexmedetomidine	0.485	0.052
	Recovery	0.491	0.032
0.1	Awake	0.604	0.056	2.960	0.064
	Dexmedetomidine	0.557	0.064
	Recovery	0.577	0.029
0.15	Awake	0.620	0.074	3.703	0.034
	Dexmedetomidine	0.554	0.071
	Recovery	0.593	0.045
0.2	Awake	0.607	0.084	3.561	0.038
	Dexmedetomidine	0.533	0.079
	Recovery	0.580	0.054
0.25	Awake	0.607	0.084	3.702	0.034
	Dexmedetomidine	0.533	0.079
	Recovery	0.580	0.054
0.3	Awake	0.577	0.107	3.766	0.032
	Dexmedetomidine	0.484	0.086
	Recovery	0.537	0.072
Global efficiency
0.05	Awake	0.176	0.032	0.870	0.428
	Dexmedetomidine	0.164	0.020
	Recovery	0.170	0.014
0.1	Awake	0.283	0.029	4.150	0.023
	Dexmedetomidine	0.259	0.021
	Recovery	0.276	0.016
0.15	Awake	0.333	0.039	5.210	0.010
	Dexmedetomidine	0.295	0.029
	Recovery	0.319	0.022
0.2	Awake	0.360	0.050	4.660	0.015
	Dexmedetomidine	0.314	0.037
	Recovery	0.342	0.031
0.25	Awake	0.377	0.058	4.440	0.018
	Dexmedetomidine	0.324	0.042
	Recovery	0.356	0.039
0.3	Awake	0.388	0.065	4.180	0.023
	Dexmedetomidine	0.330	0.047
	Recovery	0.363	0.044
Mean node strength
0.05	Awake	5.351	0.377	3.126	0.055
	Dexmedetomidine	5.033	0.393
	Recovery	5.238	0.231
0.1	Awake	9.916	1.002	3.510	0.040
	Dexmedetomidine	9.025	0.984
	Recovery	9.564	0.659
0.15	Awake	14.031	1.768	3.589	0.037
	Dexmedetomidine	12.470	1.633
	Recovery	13.380	1.185
0.2	Awake	17.792	2.632	3.602	0.037
	Dexmedetomidine	15.500	2.306
	Recovery	16.801	1.776
0.25	Awake	21.257	3.573	3.603	0.037
	Dexmedetomidine	18.192	2.982
	Recovery	19.893	2.415
0.3	Awake	24.449	4.577	3.587	0.037
	Dexmedetomidine	20.586	3.649
	Recovery	22.690	3.089

STD, standard deviation

RSN functional connectivity analyses were also conducted using custom code written in MATLAB. First we present mean networks for the three conditions (awake, dexmedetomidine and recovery). To assess statistical differences, each network was held at a threshold proportional to 50% of strongest connections and these matrices were compared using a paired t-test. Each test was corrected for multiple comparisons setting the false discovery rate to p < 0.05. The regions that showed significant changes in connectivity are shown as binarized maps where nodes are sorted based on membership to RSN groupings. To calculate the percentage of modulated connections within and between RSN networks as a discrete value, the total number of significantly modulated connections that survived correction for multiple comparison were summed for within-RSN network and between-network connections. This value was then divided by the number of total possible connections for any given combination and multiplied by hundred to calculate the percentage of modulated connections. The final values were represented as a heat map to qualitatively assess network modulations.

Results

Dexmedetomidine disrupts local and global efficiency of brain networks

The local efficiency of weighted brain networks was significantly disrupted during the dexmedetomidine-induced altered arousal state compared to the awake state (Fig 1A; 0.15: p = 0.003, 0.2: p = 0.004, 0.25: p = 0.004, 0.3: p = 0.004). During the recovery state, the local efficiency reverted to higher values compared to the dexmedetomidine-induced state. However, these findings did not meet our conservative Bonferroni-adjusted threshold for significance (Fig 1A; 0.15: p = 0.03, 0.2: p = 0.029, 0.25: p = 0.029, 0.3: p = 0.038). There were no statistically significant differences between awake and recovery state comparisons (Fig 1A; p > 0.025).

Figure 1. Local and Global efficiency of neural information transfer is disrupted by dexmedetomidine.

A,B. Local and global efficiency are significantly decreased during the dexmedetomidine-induced altered arousal state and reverted to higher values during the recovery state. This observation was consistent at a range of network densities. *, p < 0.025 for awake vs. dexmedetomidine, error bars represent ±SEM.

The global efficiency of weighted brain networks was significantly disrupted during the dexmedetomidine-induced altered arousal state compared to the awake state (Fig 1B; 0.1: p = 0.006, 0.15: p = 0.001, 0.2: p=0.002, 0.25: p=0.002, and 0.3: p=0.002). During the recovery state, the global efficiency was significantly increased for most, but not all network thresholds (Fig 1B; 0.1: p = 0.004, 0.15: p = 0.005, 0.2: p = 0.01, 0.25: p = 0.017, 0.3: p = 0.026). There were no statistically significant differences between awake and recovery state comparisons (Fig 1B; p > 0.025)

Dexmedetomidine reduces the strength of synchronizations in brain networks

Next, using a graph-theoretic approach, a significant reduction in the mean strength of nodal connectivity was found during the dexmedetomidine-induced altered arousal state compared to the awake state (Fig. 2A; 0.1: p = 0.006, 0.15: p = 0.005, 0.2: p = 0.005, 0.25: p = 0.005, 0.3: p = 0.005). The recovery state was associated with increased mean strength of nodal connectivity compared to the dexmedetomidine-induced altered arousal state. However, this change in strength was not significant (Fig 2A; 0.05: p = 0.052, 0.1: p = 0.051, 0.15: p = 0.053, 0.2: p = 0.055, 0.25: = 0.055, 0.3: p = 0.056). There were no statistically significant differences between awake and recovery state comparisons (Fig 2A; p > 0.025)

Figure 2. Dexmedetomidine modulates strength of connectivity in brain networks.

A. Mean strength of nodal connectivity in weighted networks reported for different network sparsities (*, p < 0.025 for awake vs. dexmedetomidine, error bars represent ±SEM). B. Mean strength of connection presented for each node. Nodes are ordered in ascending order of strength of connections to show that the most strongly connected nodes were the most disrupted by dexmedetomidine. C. Mean degree (total number of connections) of each node presented in ascending order. D-F. Representations of large-scale network topology illustrating how connectivity strength (weights) modulated by dexmedetomidine contribute to alterations in network architecture. The positioning of nodes is topological rather than anatomical. The algorithm positions network nodes based on strength of connections so that regions with common connections are placed in a group and the physical distance between the nodes is adjusted based on the weight of the connection. The awake state shows an ordered modular structure with a large number of within module connections. The modules are held together by global connections. In the dexmedetomidine-induced state, the strengths of connection (weights) are reduced at both local (within module) and global (between module) levels. In the recovery state, the strength of connections is increased, but does not fully revert to the awake state. All three representative networks were constructed at a threshold proportional to 0.25 connections.

To study the nodes that principally contributed to the changes in mean strength of connectivity, nodes were first sorted (lowest to highest strength of connectivity during the awake state) and then plotted by the strength of connectivity (Fig. 2B). The largest decreases in connectivity strength were most evident in strongly connected nodes. During the recovery state, the mean strength of connectivity reverted to higher values compared to the dexmedetomidine-induced altered arousal state. The degree distribution (number of connection at each node) was studied to understand whether the changes in strength of connectivity at the nodal level primarily resulted from a reduction in a number of connections versus a reduction in the strength of connection. Dexmedetomidine did not strongly modulate the number of connections at the nodal level (Fig. 2C).

To graphically illustrate our results, plots of network graphs where nodes with common connections are plotted as clusters and the distances between nodes represent strength of connectivity were made (Fig 2D-F). This network representation more clearly depicts that changes in network strength of connections is a significant driver of topological changes during the dexmedetomidine-induced altered arousal state.

Dexmedetomidine alters connection within and between resting state networks

Parcellated brain regions were sorted based on composition to RSNs. During the dexmedetomidine-induced altered arousal state, the strength of connectivity within and between networks held at the same link density appeared different compared to the baseline and recovery states (Fig. 3A-C). Contrasts between the states were analyzed to characterize connectivity changes associated with dexmedetomidine (Awake vs. Dexmedetomidine; contrast 1; Fig 3D), and recovery (Recovery vs. Dexmedetomidine; contrast 2; Fig 3E). Although volunteers were responsive to verbal commands during the recovery period, they remained mildly sedated due to residual drug effects. Therefore, contrasts between the awake and recovery states (Awake vs. Recovery; contrast 3; Fig 3F) were also analyzed. Only connectivity changes in contrast 1 were significant after correction for multiple comparisons. Therefore corrected maps for contrast 1, and uncorrected maps (p < 0.01) for contrast 2 and 3 are shown (Fig. 3D-F).

Figure 3. Dexmedetomidine modulates within-network and between-network connections when networks are organized based on meta-analytic resting state networks.

A-C. Adjacency matrices representing mean connectivity pattern in A. awake, B. dexmedetomidine and C. recovery states. The adjacency matrices consist of parcellation nodes ordered based on their membership in 5 resting state networks listed in boxed legend. Resting state networks were identified based on node overlap with connectivity maps in the neurosynth metaanalytic framework. D-F. Contrast maps showing regions that significantly changed in connectivity within and between the 5 listed networks. Results are based on t-tests, where contrast shown in D is FDR corrected at p < 0.05. E and F are uncorrected at p < 0.01. G-I Matrices showing percentage of nodes modulated in D-F. The percent values in G-I are based on sum of significantly modulated nodes observed in D-F respectively, scaled by total number of possible connections.

FDR, false discovery rate.

The percentages of modulated connections scaled by the total number of possible connections are presented for qualitative assessment (Fig. 3, G-I). The dexmedetomidine-induced altered arousal state was associated with significant reductions in functional connectivity between regions that comprise all RSNs (Fig. 3D; Awake > Dexmedetomidine, false discovery rate corrected p < 0.05). Recovery was associated with partial restoration of connectivity within the Default Mode Network, and increased functional connectivity between subcortical regions and all other networks (Fig. 3E; Recovery > Dexmedetomidine, uncorrected p < 0.01). Functional connectivity changes associated with dexmedetomidine did not fully revert to baseline values during the recovery state (Fig. 3F; Awake > Recovery, uncorrected p < 0.01). The awake state, relative to the recovery state had more within-network connectivity in Attention/Executive Networks (Fig. 3F,I). Also, the Language/Memory Networks had more between-network connectivity with subcortical regions, Sensory Network, Default Mode Network, and Attention/Executive (Fig. 3F,I).

Discussion

We previously reported that impaired thalamic information processing – loss of functional connectivity between the Default Mode Network and the thalamus and bidirectional changes between the Frontoparietal Networks and cerebellar clusters – is a neural correlate of dexmedetomidine-induced altered arousal.¹ In this investigation, we applied network metrics of information processing to study differences between baseline, dexmedetomidine-induced altered arousal and recovery states. First, we found that dexmedetomidine reversibly reduced the local and global efficiency of brain networks. Second, dexmedetomidine was associated with a reduction in the mean strength of nodal connectivity but did not alter the relative distribution of connections between nodes. Third, by using a global network approach, we show that these changes are not specific to any given RSN. Taken together, these findings parallel the decreased efficiency of information transfer within the brain that has been reported for propofol-induced unconsciousness^17,31 and non-rapid eye movement sleep,^18,19 and strengthen the hypothesis that conscious processing in the brain relies on an efficient system of information transfer.

Local and global efficiency are both measures of information integration that are derived from the characteristic path length (the average shortest path length between all possible pairs of nodes).³² Because the paths in this investigation represent statistical dependency or functional connectivity between nodes, our efficiency and mean node strength findings are consistent with our previous report of dexmedetomidine-induced functional connectivity changes.¹ Although our approach of representing each brain region in only one RSN based on maximum overlap was different from the independent component analysis method we previously employed, our finding that functional connectivity was reduced between the DMN and subcortical regions is also consistent with our previous report.¹ However, our present results extend upon those previously reported findings by showing that functional connectivity changes between subcortical regions are not specific to any RSN. Since sleep slow-delta (0.1 – 4 Hz) and spindle oscillations (13-16 Hz) reflect altered sensory information processing in the brainstem and thalamus,³³ it follows that dexmedetomidine-induced altered arousal which is also associated with slow-delta and spindle oscillations^34-36 should manifest with altered subcortical-cortical functional connectivity. We speculate that other anesthesia-induced slow-delta oscillations (propofol, sevoflurane, nitrous oxide),^35,37-41 theta oscillations (4-8 Hz; ketamine, sevoflurane),^37,38,42 frontal alpha oscillations (8-12 Hz; propofol, sevoflurane)^35,37-41 and gamma oscillations (< 40 Hz; ketamine)⁴² may also manifest as altered subcortical-cortical and cortico-cortical fMRI bold network connectivity. A likely mechanism for this speculated finding is the disruption of “normal sensory processing” gamma oscillations (>40 Hz) that have been related to fMRI BOLD signals.⁴³

Recovery from the dexmedetomidine-induced altered arousal state was associated with partially recovered connectivity between brain regions that comprise the Default Mode Network. Notably, within network alterations in Default Mode Network connectivity have been reported for dexmedetomidine,¹ sevoflurane,⁴⁴ propofol⁴⁵ and ketamine.⁴⁶ Thus, this RSN, which is associated with stimulus-independent thought and self-consciousness,^47-49 may play a significant role in recovery from altered states of arousal. Our findings also suggest that key differences between the baseline and recovery states were increased within-network connectivity in Attention/Executive and Language/Memory Networks. Thus, recovery from altered states of arousal may follow a graded pathway where within-network restoration of Default Mode functional connectivity precedes within-network connectivity in Attention/Executive and Language/Memory Networks. This finding which suggests gradations in the level of arousal is consistent with an information integration theory of consciousness⁹ may result from specific time-varying disruption (altered arousal) or reintegration (recovery) of hub nodes.

Hubs are crucial for efficient information transmission in brain networks.⁵⁰ Although a detailed analysis of network hub dynamics was beyond the scope of this investigation, our finding of reduced strength of connectivity in highly connected nodes suggests that dexmedetomidine disrupts hub nodes. Importantly, a consistent finding from graph theoretical studies of electroencephalogram⁵¹ and rsfMRI^17,31 data during propofol-induced unconsciousness is the disruption of hubs nodes. Further confirming the role of hubs in information transfer, a simulation of hub disruption using various anesthetics (propofol, sevoflurane, ketamine) resulted in disrupted surrogates of information transfer.⁵² This suggests that even though anesthetics have distinct pharmacological targets – reflected by differences in behavioral states and oscillatory dynamics – preferential disruption of information flow at network hubs is a common “macrocircuit” dynamic. An open question is whether unique patterns of hub disruption may explain the pharmacological and behavioral diversity inherent to these anesthetics.

A limitation of this study is that the level of arousal was not experimentally manipulated in a graded manner. Therefore, studies of dexmedetomidine and other anesthetics with graded manipulations of level of arousal levels are needed to more clearly delineate how the efficiency measures described in this manuscript covary with the level of arousal. Another limitation of the present study is that the present results cannot be directly translated to molecular or neurophysiological function at the neuronal level. We note that although statistically significant, our effect sizes were small in spite of clinically significant alterations to the level of arousal. However, this finding is likely a reflection of the mathematical construct underlying efficiency measures (i.e. global and local efficiency are not exponential functions).¹⁰ Finally, it is important to note that the use of graph theoretical metrics as a proxy for higher cognitive processes is not definitive.

Our findings demonstrate that dexmedetomidine is associated with a significant drop in the capacity for efficient information transfer in functional networks at both local and global levels. These findings strengthen the hypothesis that conscious processing relies on an efficient system of information transfer in the brain.