Vascular and Neuronal Effects of General Anesthesia on Functional Magnetic Resonance Imaging

A number of functional magnetic resonance imaging (fMRI) studies rely on application of anesthetic agents that can modulate and complicate interpretation of the measured hemodynamic blood oxygenation level-dependent (BOLD) response. The purpose of this study was to investigate the effect of general anesthesia on two main components of BOLD signal including neuronal activity and vascular response. Breath-holding (BH) fMRI was conducted in wakefulness and under anesthesia states in patients with drug resistant epilepsy (DRE) who needed to get scanned under anesthesia during laser interstitial thermal therapy (LITT). BOLD and BOLD-cerebrovascular reactivity (BOLD-CVR) maps were compared between two states to assess the effect of anesthesia on neuronal activity and vascular factors. Overall, our ndings revealed an increase in BOLD-CVR and decrease in BOLD response under anesthesia. The results proposed that the modulatory mechanism of anesthetics on neuronal and vascular components of BOLD signal may work in different ways. To our knowledge, this is the rst human study to examine the effect of general anesthesia using BH fMRI imaging, enhancing our understanding of the effect of anesthesia on neuronal and vascular aspects of BOLD response and assisting the implication of general anesthesia and interpretation of outcomes in clinical setting.


Introduction
Blood oxygenation level-dependent (BOLD) functional magnetic resonance imaging (fMRI) has been widely used in detection and mapping of brain function. However, it remains important to point out that BOLD response does not directly re ect neuronal activity but rather re ects cerebral hemodynamics alterations caused by neuronal activity. In fact, BOLD MRI is representative of the complex relationship between cerebral blood ow (CBF), oxygen metabolism, and neurovascular function 1,2 . Even though the empirical coupling relationships between BOLD and CBF in healthy human subjects under conventional scanning scenarios is assumed to be intact, understanding of this relationship is essential because of the confounding factors that potentially alter neurovascular coupling 3,4 . Recently, several studies have estimated the relationship between hemodynamic responses and neuronal activity in animal models. One potential confounding factor involved in these studies is the application of anesthesia. Anesthetic agents have been used in fMRI studies to limit potential artifacts induced by subject movement and to control the physiological factors. At the same time, anesthetics can depress spontaneous and evoked neuronal activity, alter both neuronal activity and vascular response, and modulate neurovascular coupling.
However, the degree and dynamics of these effects depend on the type and dose of the anesthetics [5][6][7] .
To date, animal studies have demonstrated that BOLD response decreases in anesthetized animals compared to an awake state, presumably re ecting decreased neuronal excitation under anesthesia 6,8 . However, sensitivity of brain regions to the anesthetic agents is divergent that may result in temporal variability of BOLD response in terms of its magnitude and spatial extent over time 6,8,9 . Furthermore, it has been shown that volatile anesthetics such as sevo urane have intrinsic dose-dependent vasodilatory effect that can complicate the interpretation of BOLD responses 4,10,11 . In fact, the vasodilatory effect of volatile anesthetics causes pial dilation, increases partial pressure, and elevates CBF. However, the degree of alteration in CBF varies according to the concentration and type of the agent. In fMRI studies including anesthetic agents, the brain vasculature responds to two factors including fMRI task stimulation and anesthesia that complicates the interpretation of BOLD response. As such, in these studies it is crucial to discriminate between vascular effects of the anesthetic agents and neuronal activity before one can properly interpret the associated ndings 9,12,13 . Cerebrovascular reactivity (CVR) mapping has been used to detect the ability of cerebral vessels dilation or constriction and to estimate hemodynamic response to a vasodilatory challenge such as hypercapnia [14][15][16] . As such, breath-hold (BH) fMRI task has been proposed as a non-invasive, convenient approach that can be employed to map vascular reactivity response [17][18][19][20][21][22][23][24] . CVR measurements have been performed under anesthesia in MRI-based animal studies; and using non-MRI-based approaches in humans. The present study is the rst to conduct BH fMRI acquisition to evaluate the effect of general anesthesia on BOLD and BOLD-CVR responses. We aimed to quantify the effect of a volatile anesthetic agent on BOLD and BOL-CVR brain maps in order to better understanding of the neuronal and vascular effects of anesthetic agents in human subjects.

-Cerebrovascular reactivity (CVR)
There were signi cant differences in magnitude of BOLD-CVR obtained across gray matter and thalamus between two states of awake and under anesthesia. BOLD-CVR was generally higher under anesthesia than in awake state, particularly in the right middle and superior temporal gyrus, left middle and superior frontal gyrus, right caudate, left and right anterior cingulate, and cerebellum. In a few clusters, mainly located in occipital lobe, BOLD-CVR was lower under anesthesia (p < 0.05) ( Figure 3). The list of signi cant clusters is shown in Table 1 where positive and negative regression coe cients (beta-weights) correspond with higher and lower group differences in BOLD-CVR between the two states, respectively.
The contents of the table are reported based on TT-Daemon atlas provided in AFNI package. -BH BOLD group analysis BOLD response group differences between awake and anesthetized states showed greater BOLD in awake compared to anesthetized state located in the frontal, temporal, parietal, and insular lobes; as well as the postcentral, and precentral gyrus, and thalamus (p < 0.05). The main clusters showing signi cant differences between awake and under anesthesia are shown in Figure 4. The list of signi cant clusters is represented in Table 2. -Breath holding BOLD vs. Breath holding BOLD-CVR The average of BOLD response was extracted from the three distinct masks as mentioned above. 3dmaskave in AFNI was used to measure the mean and standard deviation (SD) of BOLD from each mask. The average and SD of BOLD response were lower under anesthesia compared to awake state using all three masks. The statistical t-test revealed non-signi cant p-values comparing two groups (p > 0.05) ( Figure 5). Additionally, Levene's test showed higher between-subject variabilities in awake compared to anesthetized state for both BOLD-CVR and BOLD responses (p < 0.05).

Discussion
This study employed BH fMRI challenge in both awake and under general anesthesia states to help parse out the contributions of neuronal and vascular effects of general anesthesia. Propofol and sevo urane, the most common intravenous and inhaled anesthetics agents have been used in neurosurgical anesthesia in human studies. However, the effect of these agents on cerebrovascular hemodynamic response and during BH fMRI has not been clearly determined. It has been shown that anesthetic agents cause physiological alteration that can lead to hemodynamic BOLD response changes including alteration in neurovascular coupling, neuronal activity, and vascular reactivity. Hence, for interpretation of the results, this is requisite to contemplate pharmacology and physiological impact of the anesthetics that used in the study 4,6 .
-Effects of Anesthesia on Cerebral Vasculature Our ndings demonstrated that BOLD-CVR increased under anesthesia compared to awake state. We speculate that the increase in BOLD-CVR is linked with the physiological characteristic of the anesthetics. Anesthetic agents have been shown to affect the dynamics of the cerebral vasculature via direct action on the receptors of vascular smooth muscles that results in cerebral vasculature. In fact, volatile anesthetics such as sevo urane have been known to possess intrinsic vasodilatory activity due to their direct effect on vascular smooth muscles in a dose-dependent manner 10,25−27 .
In a recent study Sakata et al. 2019 showed that in anesthetized rats, sevo urane induced greater vasodilatory effects under hypercapnia. They concluded that cerebrovascular response to hypercapnia is higher in patients anesthetized with volatile agents 12 . Additionally, a study on patients undergoing spine surgery showed that sevo urane has vasodilatory effect on cerebral arteries 10 . Similarly, the previous study by Wang et al. 2017 on patients undergoing laparoscopic surgery found that while propofol has a vasoconstrictive effect, CVR increased due to the dominant vasodilatory impact of sevo urane. The authors concluded that in patients with impaired cerebrovascular reserve capacity, application of inhaled anesthetics can be taken into account in order to compensate vasodilation ability of cerebral vessels that leads to attenuation of the cerebrovascular resistance to partial pressure of carbon dioxide (PaCO2) 11 .
However, the effect of the general anesthetics on cerebral vessels remains controversial, as they may cause both vasodilation or vasoconstriction depending on the type and dose of the agent [25][26][27] . Several human studies reported no signi cant change in BOLD-CVR under anesthetic with sevo urane 28 . Juhász et al. 2019 used a combination of propofol and sevo urane during induction and maintenance phases in ventilated patients and found an initial increase of mean arterial pressure and vasodilation that was followed by a decrease in diastolic pressure later at the steady state phase. They concluded that vasodilation observed after induction could be associated with the effect of propofol that disappeared with no signi cant change in hemodynamic parameters over time 29 . Another study within clinical parameters of anesthetic use in neurosurgical procedures reported that CVR remains constant due to the balance between vasoconstrictive and vasodilatory properties of intravenous and inhaled anesthetics 13 . Additionally, among other volatile anesthetic agents in humans, sevo urane has been shown to have the least vasodilatory effect; while its vasodilatory effect is more prominent on non-humans 1,6,30−32 .
In the present study, BOLD-CVR mapping showed increased BOLD-CVR values under anesthetic in the cerebellum, anterior cingulate, right temporal gyrus, and left frontal gyrus. The increase in BOLD-CVR under anesthesia could be explained by the confounding effect of anesthetics over hypercapnia challenge. Given the timing of administration of the anesthetic agents, we interpret that the volatile anesthetic agent played the dominant role in cerebral vasodilation rather than propofol that resulted in higher cerebrovascular reactivity response compared to awake state. In addition, we conclude that even though volatile anesthetics such as sevo urane have a partially intrinsic vasodilatory effect, their impact on cerebral vasculature is strengthened during BH challenge and hypercapnia. Further studies are needed to explicate the nature of this pattern.
Meanwhile, it is notable that a few clusters located in the occipital lobe demonstrated decreased BOLD-CVR in anesthetized state. The signi cant group difference in BOLD-CVR involving visual-related brain regions of the occipital lobe and lingual gyrus could be driven by higher visual system activity in awake state during breath holding. Activation of the visual system has been reported in other studies because of visual cues during performing breath-holding task 33,34 . Also, our observation is in accordance with a previous study reporting that the occipital lobe may function in a opposite direction from the responses in the frontal, parietal, and temporal lobes in anesthetized state with sevo urane 35 .
-Effects of Anesthesia on Neuronal Activity It has been reported that intravenous and inhaled anesthetics have similar impact on neuronal activity in a dose-dependent manner 4,31 . From physiological perspective, they act on the ion-channel proteins that are anesthetic-sensitive and involved in neuronal excitability. The spread of neuronal excitation mostly in supra and infra granular layers of cortex has been shown to be dependent on γ-aminobutyric acid A (GABA A ), which is the main fast inhibitory neurotransmitter receptor in the central nervous system.
Volatile anesthetics modulate GABA A receptor function enhancing the inhibitory mechanism in synaptic and extrasynaptic receptors. As such, GABA A -mediated inhibition may manifest as a global reduction of BOLD response driven by a reduction of neuronal excitation through thalamic input and intra-cortical processing 36−41 . Additionally, volatile anesthetics such as sevo urane enhance the inhibitory effect of glycine receptors involved in inhibitory neurotransmission in the central neuronal system 42 . Such effects have been demonstrated in an animal study, in which several anesthetics including inhaled iso urane induced a decreased BOLD response associated with the suppression of neuronal activation 6 .
In the current study our results revealed that anesthetic agents caused a reduction in the activation area and magnitude of BOLD signal. Taken together, through the comparison between wakefulness and anesthesia states we speculate that inhaled anesthetics may complicate neurovascular coupling in which neuronal and vascular contributions of the hemodynamic response work in different ways. Therefore, it appears that these agents have a greater impact on neuronal activity (BOLD response) than on vascular effect (BOLD-CVR response). Hence, in all the studies including general anesthetics the key point to interpret the outcomes is to distinguish between neuronal excitability and vascular effects associated with using drugs.
-Effect of Anesthesia on Signal Variability In addition to the above-mentioned results, our study highlighted the effect of general anesthesia on variability of BOLD and BOLD-CVR responses. From voxel-wise Levene's tests we have found that spatial variability of BOLD and BOLD-CVR responses decreased in anesthetized compared to awake state. Our observations agree well with the previous literature comparing the variability of BOLD response in awake and anesthetized states 43 . Speci cally, it has been proposed that anesthetic agents reduce the overall dynamic complexity or neuronal variability that is primarily driven by physiological and motion noises 44 .
With regards to the quanti cation of CVR, Bright et al. 2013 showed that implementation of BH fMRI task in clinical studies could be challenging due to the highly variable performance of BH challenge in awake state 18 . However, our study proposed lower variability of BOLD-CVR responses due to the control over the physiological factors and motion artifacts that were minimized under anesthesia. We suggest that implemention of BH fMRI studies under general anesthesia improves the robustness with less variability of BOLD-CVR response, which can be utilized as a biomarker in study of patients with cerebrovascular diseases 37 .
There are several experimental limitations involved in this study which need to be considered. In interpretation of the results, the underlying pathophysiology of epilepsy in our cohort must be taken into account since it could interact with effect of the anesthetics and bias the results. Epilepsy is a cerebrovascular disease that can disrupt the balance between excitatory and inhibitory drive at the synaptic level, leading to seizures. Alteration in neuronal circuits, including anterior nucleus of the thalamus (ANT), fronto-motor cortex, and fronto-temporal cortex is involved in the pathophysiology of epilepsy [45][46][47][48][49][50] . At the same time, since each subject served as their own control, their comorbid epilepsy is less likely to have been a major factor. Also, given the small sample size, this study should be considered as an explorative work. Further research should be conducted with a greater number of participants. Future studies are needed to examine the effect of general anesthetics on BOLD responses in other group of patients with cerebrovascular diseases, as well as on healthy controls. However, the latter would be challenging to carry out since it is unethical to do such clinical practice among healthy volunteer participants.
Finally, another factor is the type and dose of the anesthetic agents. In this study we used the standardized combination of propofol and sevo urane in induction and maintenance levels, and as such, the generalizability of our ndings is limited to these agents. Different types and dosage of anesthetics need to be examined in future studies to obtain a better understanding of the effect of anesthesia in cerebrovascular and hemodynamic responses in human subjects.
In sum, our results highlight the independent effects of general anesthesia on cerebrovascular reactivity and neuronal activity. We revealed that sevo urane, an inhaled volatile anesthetic, may affect neuronal factors more signi cantly than cerebrovascular factors. While general anesthesia may help reducing variabilities arisen by physiological and motion artifacts, future studies need to consider the signi cant effect of anesthetic agents on neuronal activity that may impede a fair comparison between awake and anesthesia states, if longitudinal studies are to be done.

Methods -Participants
A total of 9 patients (4 males and 5 females, age 28-60 years old) diagnosed with drug-resistant epilepsy (DRE) were recruited from the Comprehensive Epilepsy Center at Thomas Jefferson University Hospitals. All patients had a diagnosis of mesial temporal lobe epilepsy (mTLE) with unilateral mesial temporal sclerosis (MTS) according to standard clinical criteria. As part of clinical care for mTLE patients with MTS, laser ablation of amygdala-hippocampal complex was administered. In order to deal with a typically complex patient cohort and often heterogeneous in seizure characteristics and clinical history, we had restricted the cohort through stringent inclusion/exclusion criteria: history of drug-resistant mTLE; on stable AEDs and compliant with medication use; an average of at least 1 complex partial or secondarily generalized seizure compatible with mTLE per month; seizure symptoms and/or auras compatible with mTLE; video EEG showed evidence of seizures from one temporal lobe consistent with mTLE; MRI had evidence consistent with mesial temporal lobe sclerosis. Participants were selected for this study if they were appropriate candidates for laser interstitial thermal therapy (LITT) for the treatment of their DRE. This clinical scenario required patients to be scanned in preoperative scanning in an awake state and under anesthesia during the LITT procedure 51 . There was a two-week interval between the two scanning sessions; and each session consisted of two different sets of BH fMRI acquisitions. Informed consent was obtained from all the patients and their legal guardian(s). The study was approved by the institutional review board (IRB) of Thomas Jefferson University Hospital. All methods were performed in accordance with the relevant IRB-approved guidelines and regulations.
-Anesthesia administration The second MRI scan was acquired during the LITT procedure and under general anesthesia. All patients were evaluated by an anesthesiologist and underwent institutional standard of pre-anesthetic preparation. Every patient received a standard induction of intravenous propofol (130-300 mg) 15-20 minutes before scanning. After endotracheal intubation, sevo urane anesthesia was administered through the endotracheal tube and maintained with 0.6-1.2 mean alveolar concentration (MAC) and 100% Fractional Inspiratory Oxygen (FIO2) during MRI acquisition. Mean arterial blood pressure was also maintained at 65-75 mmHg and end tidal carbone dioxide (ETCO 2 ) at 30-35 mmHg throughout the procedure. After MRI acquisition and the LITT procedure were completed, patients were reversed with Sugamadex 2-4 mg/kg, extubated as per routine, and were observed in a post-anesthetic care unit during their recovery.
-BH paradigms The breath holding (BH) protocol consisted of two separate BOLD fMRI acquisitions with a 15-minute interval between scanning sessions. The BH task consisted of 40 seconds self-paced normal breathing followed by 20 seconds breath holding. This cycle repeated 5 times, and additional 40 seconds normal breathing incorporated at the end of each task for a total duration of 5 minutes and 40 seconds ( Figure  1). This BH fMRI task was determined according to the designs suggested by the previous studies. For instance, Dlamini et al. 2018 reported that the minimum of 3 sec is enough to produce a detectable fMRI BOLD response. However, a longer period of BH (19.6-20.3 seconds) is required to produce a robust and reproductible BOLD response that could assure su cient magnitude and number of voxels manifesting the response 20 . In addition, J.J. Pillai et al. 2015 showed that 20-30 seconds of BH could provide consistent results with less interscan variability response that is feasible for healthy participants. As such, 20 seconds BH was selected in the cycles since this could be easier for the subjects to follow the task than longer durations. In addition, BH was performed with an end-inspiration to increase subject's comfort 52 .
-Clinical measurements Hemoglobin concentration (Hgb) was obtained via a blood draw immediately prior to scan acquisition.
Other clinical parameters were recorded during the scanning, including non-invasive blood pressure, peripheral capillary oxygen saturation (SpO2), and end-tidal carbon dioxide (etCO2). During the awake session, a nasal cannula was used to collect expired gases after exhalation to measure etCO2; while during the session under general anesthesia, this measurement was performed by the ventilator through the endotracheal tube.
-Data analysis Functional MRI data were analyzed using AFNI (Analysis of Functional NeuroImages, (https://afni.nimh.nih.gov/afni)) software 53 . The modules afni_proc.py and 3dDeconvolve were used for preprocessing and tting the respiration response function (RRF) to the BH BOLD signal. Following standard initial preprocessing, 3dtcat in AFNI was used to remove the rst 2 volumes of fMRI data in order to ensure the MR signal had reached a steady state. Slice timing correction was then applied. The next steps included motion correction by 6 rigid-body-movement parameters as covariates, removal of linear drift, and smoothing with a 4mm full-width half-maximum (FWHM) Gaussian kernel. All EPI volumes aligned with the low-motion volume in the dataset (min_outlier) using 3dvolreg in AFNI. They were then concatenated and wraped to standard Talairach space. The functional data were registered to the structural MPRAGE image using an A ne transform with 12 degrees of freedom, that was then normalized to the Talairach standard space. In addition, regress_censor_motion_0.3 and regress_motion_per_run were employed to censoring of large motion (≥ 0.3 ~mm between successive time points, based on the motion parameters) as well as to regress out of outlier time points and motion parameters to be modeled separately per run 54 . The two trials of BH fMRI data within each session were concatenated, while keeping the awake and anesthesia scan sessions separate.
Further data processing was carried out across the region of interest (ROI) including gray matter and thalamus areas. Statistical Parametric Mapping 12 (SPM12) and recon-all in FreeSurfer were used to create gray matter and thalamus masks respectively. The threshold of 0.1 was selected to binarize the probabilistic maps. For each patient under LITT treatment, a mask of the implanted laser probe was made manually for each subject. Figure 2 illustrates one sample of the probe mask generated using structural data for one of the participants. With the aim of excluding the potential artifacts that arise from the probe, the area under the masks were removed from both structural data and statistical maps. This scenario was replicated on preoperative data in the awake state as well to make a fair comparison between the two states. All the 3D voxel-to-voxel arithmetic calculation were executed using 3dcalc tools in AFNI.
-BOLD cerebrovascular reactivity (BOLD-CVR) estimation BOLD BH fMRI analysis In order to evaluate both vascular and neuronal effect of anesthesia, BOLD activation maps were also obtained by employing the typical gamma variate function to model BOLD response [55][56][57] . Magnitude of BOLD-CVR and BOLD responses were extracted over the ROI mask (gray matter and thalamus) from regression coe cients (beta weight values) maps that correspond with the approximate signal change associated with hypercapnia challenges.

-Statistical analysis
Second-level group analysis was conducted to estimate the contrast between two states of awake and anesthesia. The maps of regression coe cients (beta values) over ROI mask between the two states were compared using the voxel-wise student t-test with 3dttest++ (the program of the AFNI package). Twosided signi cant thresholding was chosen at p-value < 0.05 (family-wise error correction). Also, the cluster-size with minimum number of voxels in each cluster de ned at 150 voxels. The estimation was conducted by Montecarlo simulation performed with 3dClustStim from the AFNI package, and nearest neighbor (NN) level set at 1 (face-wise bordering; n=6 voxels) 58-60 . The group analysis was performed for both BOLD and BOLD-CVR maps between groups toward investigating the effect of anesthesia on both neuronal and vascular factors. Then, three separate masks were generated from the clusters showed signi cant positive and negative BOLD-CVR differences between awake and under anesthesia states, as well as the brain regions within the ROI mask by subtracting the rst two masks. These masks were used further to assess the overall mean BOLD signal in two states of awake and under anesthesia.
-Homoscedastic test To investigate the effect of anesthesia on the variability of BOLD-CVR and BOLD responses, the Homoscedastic test (Levene's test) was applied among regression coe cient (beta value) maps of the two states of awake and under anesthesia. This statistical analysis assesses the homogeneity of variances for a variable in two or more groups. The test assumes the variances are equal unless the pvalue of Levene's test is less than the prede ned signi cant level (typically 0.05) 61 .

Declarations
Authors' contributions F.A. designed the study, processed the acquired data and wrote the paper, M.A. helped with statistics and data analysis, M.T. helped with data processing and analysis, V.R. supervised and managed clinical and surgical aspects of the project and edited the paper, F. B. M. advised on data analysis and manuscript writing and editing, C. W. was the principal study supervisor, supervised clinical and surgical aspects of the project and helped with manuscript writing and editing. All authors reviewed the manuscript.  Sample mask of a probe created for one of the patients overlaid on the structural data. Red corresponds with increased BOLD signal, and blue corresponds with decreased BOLD signal in anesthesia state (p<0.05).

Figure 5
Plots and histograms of mean BOLD signal from the masks where the BOLD-CVR difference between awake and under anesthesia states was positive (a), negative (b) and the rest of the gray matter and thalamus mask (c) (p<0.05). Each plot shows the mean and SD of the BOLD response extracted from the corresponding masks combined with related density histogram. The statistical t-test between awake and anesthesia states show non-signi cant differences between groups (t-test, p-value > 0.05).