Multimodal pathophysiological dataset of gradual cerebral ischemia in a cohort of juvenile pigs

Ischemic brain injuries are frequent and difficult to detect reliably or early. We present the multi-modal data set containing cardiovascular (blood pressure, blood flow, electrocardiogram) and brain electrical activities to derive electroencephalogram (EEG) biomarkers of corticothalamic communication under normal, sedation, and hypoxic/ischemic conditions with ensuing recovery. We provide technical validation using EEGLAB. We also delineate the corresponding changes in the electrocardiogram (ECG)-derived heart rate variability (HRV) with the potential for future in-depth analyses of joint EEG-ECG dynamics. We review an open-source methodology to derive signatures of coupling between the ECoG and electrothalamogram (EThG) signals contained in the presented data set to better characterize the dynamics of thalamocortical communication during these clinically relevant states. The data set is presented in full band sampled at 2000 Hz, so the additional potential exists for insights from the full-band EEG and high-frequency oscillations under the bespoke experimental conditions. Future studies on the dataset may contribute to the development of new brain monitoring technologies, which will facilitate the prevention of neurological injuries. Measurement(s) EEG with persistent abnormal rhythmic activity • electrocardiogram • Electrothalamogram • cerebral blood flow measurement • positive regulation of heart rate involved in baroreceptor response to decreased systemic arterial blood pressure • cerebral metabolic rate of oxygen • cerebral perfusion pressure Technology Type(s) electrophysiology assay • extracellular electrophysiology recording assay • fluorescent colormicrospheres • Electrophysiology Factor Type(s) sedation • cerebral ischemia • recovery Sample Characteristic - Organism Sus scrofa Sample Characteristic - Environment laboratory environment Measurement(s) EEG with persistent abnormal rhythmic activity • electrocardiogram • Electrothalamogram • cerebral blood flow measurement • positive regulation of heart rate involved in baroreceptor response to decreased systemic arterial blood pressure • cerebral metabolic rate of oxygen • cerebral perfusion pressure Technology Type(s) electrophysiology assay • extracellular electrophysiology recording assay • fluorescent colormicrospheres • Electrophysiology Factor Type(s) sedation • cerebral ischemia • recovery Sample Characteristic - Organism Sus scrofa Sample Characteristic - Environment laboratory environment Machine-accessible metadata file describing the reported data: https://doi.org/10.6084/m9.figshare.13251176


Background & Summary
Surface EEG contains information about corticothalamic communication, which can be quantified even without invasive insertion of thalamic electrodes 1,2 . In this study on juvenile pigs, we introduced the basic stereotaxic approach to chronically recording electrothalamogram (EThG) and quantifying the effects of isoflurane and fentanyl sedation on the brain electrical activity from a ten-channel electrocorticogram (ECoG), EThG, and the cerebral blood flow 2 , followed by the characterization of the effects of gradual propofol sedation on these parameters 1 .
The choice of this animal model is dictated by its amenability to complex stereotactic chronic instrumentation, monitoring studies of sedation, and clinically relevant patterns of hypoxic/ischemic injury in a relatively large and gyrencephalic brain 2 .
Here, we present the unique data set underlying the above experiments and expand it to include the conditions of gradual ischemia and recovery that were part of the original experiments but were never presented before. We focus on the spontaneous ECoG/EThG activity and exclude the evoked potential responses from this dataset. We believe this makes for a focused presentation of this complex multimodal recording series and will present the evoked potentials dataset separately. By way of technical validation, we present a basic approach to derive ECoG/ EEG biomarkers of normal, sedation, and hypoxic/ischemic conditions using EEGLAB 3 ; we present an elementary approach to quantify the ECoG activity on the cohort level identifying state-specific independent component (IC) features of ECoG common to all animals studied.
Moreover, for the first time, we present the electrocardiogram (ECG) dataset accompanying the ECoG/EThG providing its basic technical validation, annotating, and providing the corresponding cardiovascular and brain regional and systemic continuous blood flow data.
We hope the present data set will lay the foundation for new brain monitoring technologies, which will facilitate the prevention of neurological disorders.
Instrumentation of the head. The animals were placed in a sphinx position with the head fixed in a stereotactic frame (Kopf Instruments, Tujunga, USA) (Fig. 1). The skull was exposed and burr holes were made for the insertion of ten electrocorticographic electrodes for bilateral leads from frontal, parietal, central, temporal, and occipital regions and guides for electrodes into the left thalamus, specifically, the reticular (RTN) and dorsolateral (LD) nuclei (unipolar recordings; reference, nasal bone).  Table 1. We induced gradual cerebral ischemia as follows 4 . First, the cisterna magna was punctured by a lumbar puncture needle that was fixed in place by dental acrylic resin for elective artificial cerebrospinal fluid infusion/withdrawal to control ICP. Then, the mean ABP was adjusted to about 90 mmHg by the appropriate curbing of the pulmonary trunk diameter with the plastic-coated cerclage. The cerebral perfusion pressure (CPP) was then decreased at 25 mmHg, which was calculated as the difference between MABP and the intracranial pressure (ICP) by appropriate elevation of the ICP. The increase in the ICP was achieved by the infusion of artificial cerebrospinal fluid (warmed to 37 °C) into the subarachnoid space via the punctured cisterna magna. The Cushing response during severe brain ischemia was prevented by the appropriate curbing of the pulmonary trunk diameter with the plastic-coated cerclage to control cardiac output. Finally, the cerclage was opened completely, and the artificial cerebrospinal fluid was withdrawn to reach an ICP < 10 mmHg.
The states of gradual ischemia were maintained for 15 min twice (States 6 and 8), interceded by a recovery state (State 7) lasting 30 minutes in all animals (P746 -P794) except P739 where it lasted 15 minutes. This was followed by 60 min of recovery (States 9-12).
At the end of the experiment, the animals' brains were perfusion-fixed 2 . Afterward, the head was removed, immersion-fixed, the brain was removed and electrode positions were visually and histologically confirmed.
Data acquisition and analyses. Unipolar ECoG and EThG were amplified (DC-EEG-AMPLIFIER, Schwind Medizintechnik, Erlangen, Germany), filtered (time constant was 0.1 second, cut off frequency was 1000 Hz), fed into a multi-channel recording device (GJB Datentechnik Bolten & Jannek GbR, Ilmenau, Germany), stored after A/D conversion continuously on hard disc at 125 Hz and parallel-connected at 2000 Hz for about 300 seconds at predefined states (see below).
Cardiovascular and metabolic parameters. ECG, heart rate, MABP, body temperature, arterial and brain venous pH, pCO 2 , pO 2 , O 2 saturation, glucose, lactate, and hemoglobin values were measured at each State as published 1,2 .
Cerebrovascular and cerebral metabolic parameters. At states 1,2,5,6 and 12, whole and regional brain blood flows were measured with colored microspheres 5 together with brain oxygen extraction (arterial -sagittal sinus blood oxygen content difference, AVDO 2 ) and cerebral metabolic rate of oxygen (the product from cortical blood flows and AVDO 2 ) 4 .
Signal analysis methodology. EEGLAB v2019.1 for Linux was used in Matlab R2013b (MathWorks, Natick, MA) to analyze ECoG/EThG data 3 . ECG-derived heart rate variability (HRV) was assessed via a series of automated algorithms that process a waveform recording into a comprehensive multivariate characterization of its degree of variability and complexity metrics, using the Continuous Individualized Multiorgan Variability Analysis (CIMVA) software tool 12 .
Briefly, individual heartbeats were identified from the ECG waveform, deploying commonly used QRS delineation algorithms, and a time series of R-peak to R-peak time intervals (RRI) was formed. Movement artifacts, noise, disconnections, and saturations were automatically identified, and a beat-by-beat signal quality index was derived, using continuity and morphology analyses. The RRI time series was then filtered to exclude abnormal beats and the signal complexity and degree of variability were assessed using the cleaned RRI time series.
www.nature.com/scientificdata www.nature.com/scientificdata/ HRV metrics were tracked over time using a moving window analysis (5-minute windows with a 50% overlap between windows). A comprehensive set of linear and nonlinear variability metrics were calculated within each window, as each technique provides a unique perspective on the data and no single method can provide a complete characterization of the biologic signals 13,14 . Variability metrics included measures characterizing the statistical properties (e.g., standard deviation, RMSSD), the informational complexity (e.g., entropy measures), the pattern of variations across timescales (e.g., fractal measures, power-law exponents) or the energy contained in the signal (e.g., spectral measures). The output of the variability analysis was a multivariate representation of variability tracked over time 15,16 . The variability metrics were then averaged over the different periods of interest.

Data Records
The data structure and annotation are as follows. 16 channels are given at a sampling rate of 2000 Hz and a recording duration of about 300 s per state each containing the following channels.  Fig. 1 and Table 1 show the entire instrumentation approach, ECoG channel 3 was of poor quality in most instances due to intermittent technical problems (random faults) and an amplifier breakdown. We hence removed this channel from the dataset presented. That means that the left parietal ECoG (channel 3) is not included in the dataset.
The sample raw recording is shown in Fig. 3. All animals underwent gradual ischemia. The two groups are defined with respect to their exposure to propofol sedation. The experimental group comprises N = 6 animals (P728, 737, 738, 743, 746, 752). They experienced propofol burst suppression followed by moderate propofol sedation prior to gradual ischemia. The propofol sham group comprises N = 5 animals (P739, 749, 753, 791, 794). They did not receive propofol. Instead, they were continued on fentanyl analgosedation prior to gradual ischemia. www.nature.com/scientificdata www.nature.com/scientificdata/ The animal P728 experienced an experimental mishap during the first gradual ischemia stage (state 6). The recordings are presented up until state 4 and can be grouped with other animals' data for the respective states 1 through 4. The animal P738 demised prematurely during the second gradual ischemia (state 8). Consequently, the data is presented up until state 6 and can also be grouped and studied together up until this point. Here, special consideration should be made for the potential incipient deterioration leading up to the early demise which represents a point of interest.
In addition to the raw data, we deposited the corresponding, time-matched arterial and cerebral-venous measurements of blood gases, electrolytes, metabolites, cardiovascular, and cerebrovascular as well as CMRO 2 data. The exact list of the types of data is outlined in Table 2.
All data have been deposited on Figshare in BIDS format, as European Data Format (EDF) files with uniquely identifiable animal ID and states numbers as well as in the structure of EEGLAB's STUDY object which contains the specific annotation of all file names and experimental states as well as the more general states (sedation, ischemia, recovery) for any future analyses. The data can be located under the https://doi.org/10.6084/ m9.figshare.7834442.v9 and https://doi.org/10.18112/openneuro.ds003380.v1.0.0 17,18 .

Technical Validation
Brain electrical activity. We present the approach and findings of signatures of global and brain-regional changes in ECoG-derived independent component activity during sedation, ischemia, and recovery states. Table 2, bold font, lists the animal IDs and experimental states we selected for technical validation. We considered states 1-5 combined as sedation, states 6 and 8 as ischemia, and states 9-12 as post-ischemic recovery. Using the STUDY design feature of EEGLAB in Matlab, we conducted statistical analyses on a subgroup of six animals (P737, P739, P752, P753, P791, P794) as follows: (1) read raw data (2,000 Hz sampling rate); (2) select 10 ECoG channels; (3) remove DC offset; (4) resample at 100 Hz; (5) a bandpass filter with FIR 1-40 Hz and save as *.set files (also shared on figshare) 17 (6) map ECoG channel locations for better visualization using Nz as reference (.loc file shared on FigShare; 17 cf. Table 1) (7) create STUDY in EEGLAB; this approach permits statistical level inferences across animals for state-specific changes common to all subjects (8) compute independent components for all animals and states in the STUDY (9) identify IC clusters (using k-means approach) that differentiate sedation, ischemia, and recovery states.
We deployed EEGLAB software suite (http://sccn.ucsd.edu/eeglab) 3,19,20 available as Matlab/Octave add-on, or a pre-compiled open-source package for Windows, Mac, and Linux operating systems to conduct technical validation of our data set and to demonstrate some initial findings of interest for future studies. The technical advantage of this approach is that this software package is open source and readily available online for the major operating systems. www.nature.com/scientificdata www.nature.com/scientificdata/ We focused on the representative experimental states 2 (baseline), 6 (acute gradual ischemia), and 12 (60 min recovery) for all subjects using the STUDY functionality of EEGLAB which allows studying all subjects at once compared the event-related potential (ERP) responses within the group. ERP represented the response to the states of the experiment (Table 3). To facilitate the computation, we downsampled the data to 100 Hz and focused on the ECoG channels for EEGLAB-based analysis which allowed us to map the channels according to 10/20 using a channel location file (see Figshare 17 ). Future studies on this data set could consider the full bandwidth ECoG/EThG information contained in the recordings and include EThG channels in the investigations.
We performed the independent component analysis (ICA) on animals identified in Table 3 followed by wavelet time-frequency and cross-coherence analyses on the identified independent components.
The results are presented in Figs. 4-9. The analysis of the raw data (Fig. 4a) revealed the activity to be concentrated in the delta -alpha band range which experienced most of the reduction in spectral power in the individual channels (Figs. 4b, 5-7) as well as in cross-coherence (Fig. 8) due to ischemia and during recovery. Following ICA, we observed an overall reduced cross-coherence in IC2 -IC1 compared to raw channel C3 -C4 analysis, but still following a similar pattern throughout states 5, 6, and 12 (Fig. 8). As expected, a portion of the observed coherence is accounted for by volume conduction effects which are removed by ICA. As evidenced by the group analysis results of which are shown in Fig. 9, 60 min recovery did not suffice to restore the global reductions spectral power activity seen during the ischemic states.
Ischemia reduces total power across the spectrum going from sedation states over to ischemia and then recovery, however, during recovery the bihemispheric coupling is re-established. A combination of spectral power characteristics and coupling dynamics can be used to distinguish ischemia-induced loss in spectral power and post-ischemic recovery. Clustered ICA also distinguishes these states.
In the example of P752, we focus on the representative states 5 (60 min into moderate propofol sedation), 6 (gradual ischemia), and 12 (60 min post-ischemic recovery). For each state, we show the following: (1) the raw ECoG tracings (Fig. 4a, normalized and DC removed), corresponding topological power spectrum maps (Fig. 4b) showing gradual suppression of activity with peaks at 4, 12, and 23 Hz, and the time-frequency representation using wavelet transform on the channel 5 (i.e., C3; Fig. 5); (2) the independent component (IC) analyses (Fig. 6a) with focus on 12 Hz activity (one of the clearest consistent peaks in the entire group) for the five top ICs in each state; representative IC2 power spectrum (Fig. 6b), again showing the overall reduction of activity and the corresponding time-frequency representation using wavelet transform on the IC2 (Fig. 7); (3) cross-coherence representations of bihemispheric coupling between C3-C4 and IC2 -IC1 (Fig. 8) (4) Group analysis (Fig. 9) showing the gradual reduction of spectral power between the experimental states representing sedation (states 1-5, blue line), ischemia (states 6 and 8, red line), and recovery (states 9-12, green line). We observe peaks around 12 and 23 Hz which triggered our focus on these frequencies in the representative examples shown in Figs. 4-8.
Overall, this finding shows that ischemia and recovery states can be distinguished globally using such an ICA approach. As a proof-of-concept, this analytical approach in EEGLAB demonstrates the potential of the presented experimental data set to yield new pathophysiological insights into global and brain-regional responses to sedation, gradual ischemia, and post-ischemic recovery. Cerebrovascular data CBF_Whole brain, wCBF (ml/100 g*min) f cerebral perfusion pressure (mm Hg) = ABP -intracranial pressure, ICP g cerebral metabolic rate of oxygen, CMRO 2 (ml/100 g*min) = wCBF × arterial-brain venous difference of Oct, (AVDO 2 ) Brain regional blood flow (ml/100 g*min)  www.nature.com/scientificdata www.nature.com/scientificdata/ Cardiovascular activity. To demonstrate an approach to use the present dataset for studying the unique features of cardiac electrical activities during sedation, ischemia, and recovery stages, we exported the ECG channel (Ch. 3) at 1000 Hz as EDF files for the same animals for which we presented the above approach on the ECoG data (see Figshare for ECG-ECoG data and extended results) 17 .
Briefly, first, we computed HRV measures from the ECG channel using the approach described in Section 3 15,16 . Next, we grouped the findings according to the three main states of the experiments into sedation, ischemia and recovery following the same approach as in Fig. 9. Finally, we visualized the changes in the HRV measures in In detail, the HRV measures used are summarized in Online-only Table 1. First, we normalized HRV measures by the mean heart rate (HR) when calculating the variability to minimize the effect of the different HR levels in the different states. Second, we standardized each metric and grouped the variability metrics in broad categories according to what they represent (Fractal, Chaotic, Entropies, Spectral, Time Domain, Poincare, Symbolic Dynamics, RQA) and harmonized the meaning across metrics so that small values represent less variability or less complexity and large values the opposite. We did so by inverting the variability metrics or the absolute value of the variability metrics, when appropriate. For each category of metrics, we then performed a PCA transformation and reconstructed a composite HRV metric for each category, based on the N components explaining 90% of the variance in the data. Finally, we created notched boxplots per group. Such boxplots give an estimate of the confidence interval around the median, whereby two medians are significantly different at the 5% level if their intervals (extremes of the notches) do not overlap. We also prepared the PCA boxplot based on ungrouped metrics, i.e., all 54 metrics in one plot (see FigShare 17 ).
Overall, very few differences in PCA-based HRV measures were significant, which is expected given the small sample size. RQA-based HRV measures did stand out (Fig. 10) and this is consistent with our findings implicating these HRV measures in reflecting the intrinsic cardiac function which was affected severely by the induced ischemia 21,22 . Limitations. To perform technical validation, we selected six animals for sedation, five for ischemia and six for recovery periods. However, within these broad experimental periods, not all individual animals were represented equally for each experimental state. This may cause a data imbalance issue affecting the statistical results, in particular given the small sample size. With our EEGLAB STUDY design, we pursued the intention of quantifying general trends characterizing differences between the periods of sedation, ischemia and recovery. We suggest that this is adequately demonstrated through presented ECoG and HRV analyses and leave a more in-depth analysis to future studies on this dataset.  www.nature.com/scientificdata www.nature.com/scientificdata/

Usage Notes
For thalamocortical and cortico-cortical communication, sedation-dependent linear and nonlinear coupling dynamics have been reported 1 , but not yet characterized under conditions of gradual ischemia as a potential biomarker of brain state and recovery. Studies in rats have shown that global brain hypoxia-ischemia affects long-term information processing in thalamic circuitry and the transfer of sensory information in thalamocortical networks 23,24 . Newborn mice exposed to ischemic insult also suffer from the increased vulnerability of thalamocortical circuitry 25 . There is less data on thalamocortical responses to ischemia from larger mammals with stronger resemblance of brain maturity, developmental profile, and injury patterns, such as sheep or pig 26 Fig. 4 Representative findings for sedation (state 5, moderate propofol sedation), ischemia (state 6, ischemia) and recovery (state 12, 60 min recovery)*. For each state, we show the raw ECoG tracings (Fig. 4a, normalized and DC removed), corresponding topological power spectrum maps (Fig. 4b) showing gradual suppression of activity with peaks at 4, 12, and 23 Hz. *Note that all axes are set identically for easy comparison except the topological and time-frequency transform heat maps. These are individually optimized in range for best viewing experience.  www.nature.com/scientificdata www.nature.com/scientificdata/ The present dataset may help to close this data gap and yield new insights into monitoring, early detection, recovery of ischemic and post-ischemic brain states, in particular, thalamocortical communication which are important to help restore long-term brain health.  Fig. 7 Representative findings for sedation (state 5, moderate propofol sedation), ischemia (state 6, ischemia) and recovery (state 12, 60 min recovery)*. For each state, we show the time-frequency representation using wavelet transform on the IC2 corresponding to the power spectrum representation in Fig. 6b. *Note that all axes are set identically for easy comparison except the topological and time-frequency transform heat maps. These are individually optimized in range for best viewing experience.  Fig. 6a the independent component (IC) analyses with focus on 12 Hz activity (one of the clearest consistent peaks in the entire group) for the five top ICs in each state and the representative IC2 power spectrum (Fig. 6b), again showing the overall reduction of activity. *Note that all axes are set identically for easy comparison except the topological and time-frequency transform heat maps. These are individually optimized in range for best viewing experience.
Scientific Data | (2021) 8:4 | https://doi.org/10.1038/s41597-020-00781-y www.nature.com/scientificdata www.nature.com/scientificdata/ Anesthesia-induced changes in brain electrical activity, in particular, due to the dedicated GABA A receptor-mediated effects of propofol on the RTN, have been used to model and study changes in consciousness and behavioral state activity [27][28][29][30] . In this experimental design, we benefit from propofol sedation with electrode placement in the RTN, which, as discussed, is particularly amenable to linking drug-induced agonism on GABA A receptors of RTN and the sedation depth 1 . The combination of propofol sedation with subsequent ischemia data from ECoG and EThG, including Nucl. reticularis thalami, yields a rich dataset to study patterns of thalamocortical communication under conditions of sedation, ischemia, and recovery.
The present dataset has been acquired at a 2,000 Hz sampling rate and is hence amenable to studies of the properties of high-frequency oscillations under conditions of various sedation regimes, gradual ischemia, and recovery periods [31][32][33] . Of great interest, thanks to EThG recordings, it may be possible to relate the ECoG patterns of spontaneous high-frequency oscillations to their thalamic contributions in EThG.
In this data-oriented manuscript, we used EEGLAB for the demonstration of the technical rigor, data quality, and reproducibility. While rich in functionality and open source, in future studies of this dataset the EEGLAB application could be well-complemented by additional analytical approaches, also available open-source, such as the JIDT software package 34 . (https://github.com/jlizier/jidt/) JIDT is Java-based, offers GUI, requires no   www.nature.com/scientificdata www.nature.com/scientificdata/ installation, and runs on all major platforms. It also can be integrated with Matlab, Python, or R, among others. Features, of interest to this dataset that are available with JIDT, include mutual information and transfer entropy. It would be of interest to compute these before and after ICA as well as after subtracting IC expecting the bi-channel / bivariate information measures to decrease after ICA and re-increase after removing ICs. Using JIDT software, one can also determine cross-entropy measures for ECoG -EThG channels prior to and after IC computation.
Finally, the simultaneous availability of the cardiovascular and brain electrical data lends itself to studying joint dynamics of, for example, ECG and ECoG or EThG. The relationships between these time series have been reported 35,36 , sporadically, but much remains to be done to establish the physiological framework for these relations and to explore the biomarker potential of such multivariate EEG -ECG analyses. We leave this exciting direction of research to future studies using this data set.

Code availability
EEGLAB has been used which is available as open-source. CIMVA documentation is available online: https:// ohridal.org/cimva/CIMVA-Core-Description.pdf. No proprietary code has been deployed in this study.

Fig. 10
Group level overview of the behavior of the ECG-derived multi-dimensional heart rate variability (HRV) measures. Principal component analysis (PCA) was used to separate the states of sedation, gradual ischemia and recovery and shows the best performing PCA over RQA (recurrence quantification analysis) measures of HRV. Note the general drop of HRV during ischemia and recovery during the period of post-ischemic recovery. The notched boxplots representation is helpful in visualizing the difference in the median; two medians are significantly different at the 5% level if their intervals (extremes of the notches) do not overlap 39 .