Abstract
Full text Figures and data Side by side Abstract eLife assessment Introduction Results Discussion Methods Data availability References Peer review Author response Article and author information Metrics Abstract Perturbational complexity analysis predicts the presence of consciousness in volunteers and patients by stimulating the brain with brief pulses, recording EEG responses, and computing their spatiotemporal complexity. We examined the underlying neural circuits in mice by directly stimulating cortex while recording with EEG and Neuropixels probes during wakefulness and isoflurane anesthesia. When mice are awake, stimulation of deep cortical layers reliably evokes locally a brief pulse of excitation, followed by a biphasic sequence of 120 ms profound off period and a rebound excitation. A similar pattern, partially attributed to burst spiking, is seen in thalamic nuclei and is associated with a pronounced late component in the evoked EEG. We infer that cortico-thalamo-cortical interactions drive the long-lasting evoked EEG signals elicited by deep cortical stimulation during the awake state. The cortical and thalamic off period and rebound excitation, and the late component in the EEG, are reduced during running and absent during anesthesia. eLife assessment This study makes a fundamental observation about the role of activity in the mouse thalamus on scalp recorded voltage fluctuations. The novel approach and sophisticated analysis of neural signals provides compelling support for the authors’ observations. This work will likely be of broad interest to neuroscientists. https://doi.org/10.7554/eLife.84630.3.sa0 About eLife assessments Introduction A long-standing clinical challenge has been to discover sensitive and specific biomarkers of consciousness. One obvious candidate is EEG, with high amplitude delta activity usually taken as an indicator of absence of consciousness, as during deep sleep, anesthesia, in vegetative state patients (also known as behavioral unresponsive wakefulness syndrome), or in coma (Bai et al., 2017; Kobylarz and Schiff, 2005; Schiff et al., 2014). However, given the vast diversity of patients and their etiology, spontaneous EEG can show very unusual spatiotemporal patterns, with attendant high false alarm and miss rates in diagnosing individual patients with disorders of consciousness (Farisco et al., 2022; Frohlich et al., 2021; Thibaut et al., 2019). More promising is the perturbational EEG (Bai et al., 2021), in which the brain is probed by a brief pulse (generated via transcranial magnetic stimulation [TMS] applied to the skull or electrical stimulation applied intracranially), and the resulting cortical activity is recorded using a high-density EEG electrode array or stereo-EEG electrodes (Casali et al., 2013; Casarotto et al., 2016; Comolatti et al., 2019; Massimini et al., 2005; Pigorini et al., 2015; Rosanova et al., 2018). A simple algorithm then computes the perturbational complexity index (PCI) of the brain’s reverberations to this pulse from which the presence of consciousness can be inferred with unprecedented sensitivity (low false alarm rate) and specificity (low miss rate), at the level of individual patients (Bai et al., 2021; Kondziella et al., 2020). PCI has been comprehensively studied and validated in humans since its introduction by Massimini and colleagues in 2005 (Casali et al., 2013; Casarotto et al., 2016; Comolatti et al., 2019; Ferrarelli et al., 2010; Massimini et al., 2005; Rosanova et al., 2018). Pigorini et al., 2015 proposed that during non-rapid eye movement (NREM) sleep, when the brain is characterized by cortical bistability, neurons tend to fall into a down-state following activation (either endogenous or exogenous) preventing extended causal interactions that are typical during wakefulness (Hill and Tononi, 2005; Rosanova et al., 2018; Timofeev et al., 2001). Limitations in access to high resolution cellular recordings in humans have motivated recent efforts to translate the PCI technique to model systems for more detailed studies of the circuit mechanisms underlying complexity changes. In 2018, D’Andola et al. demonstrated that an approximation of PCI developed for in vitro measurements captured complexity differences in ferret cortical slices exhibiting sleep- and awake-like activity patterns. They showed that during a sleep-like state, electrical stimulation induced a down-state that disrupted the complex pattern of activation observed in the awake-like state (D’Andola et al., 2018). Arena et al., 2021 were the first to recapitulate the human PCI results in vivo in rodents; they computed the complexity of electrically evoked EEG responses in rats and showed that PCI was high during wakefulness and low during anesthesia. In their comprehensive study, they found that in the anesthetized state (via propofol or sevoflurane), stimulation was followed by a widespread suppression of high frequencies in the EEG responses, suggestive of a down-state, in agreement with previous findings (D’Andola et al., 2018; Pigorini et al., 2015). Dasilva et al., 2021 measured PCI in vivo in anesthetized mice and showed that complexity can be modulated even within the anesthetized state. They showed that PCI was highest for mice under light isoflurane anesthesia (defined as a concentration of 0.1%), decreasing systematically at medium (0.34%) and deep (1.16%) concentrations. Recently, Cavelli et al., 2022 validated the use of PCI in freely moving rodents across wakefulness, both REM and NREM sleep, and anesthesia by demonstrating that PCI calculated on cortical laminar local field potential (LFP) signals decreases in unconscious states (both natural and induced) in rats and mice. While pioneering, none of these studies simultaneously recorded micro- and macro-signals – neurons and LFP at hundreds of sites, together with EEG. To investigate how cortico-cortical and cortico-thalamic activity influences the complexity of the evoked responses, we used EEG simultaneously with Neuropixels probes (Jun et al., 2017) to record brain-wide evoked responses to cortical electrical stimulation in head-fixed mice that were awake and, subsequently, anesthetized with isoflurane. Due to their ability to record spiking and LFP signals from hundreds of sites across cortical layers and subcortical structures at 20 μm resolution, Neuropixels probes provide unprecedented access to the intra- and interareal dynamics that underlie the macro-scale EEG signals. We show that cortical stimulation elicits a widespread, complex event-related potential (ERP) in the EEG signals in the awake state, but a much simpler ERP in the isoflurane-anesthetized state, in agreement with what has been shown in humans (Casali et al., 2013) and rodents (Arena et al., 2021; Cavelli et al., 2022). We demonstrate a stereotyped pattern of activity to stimulation in deep (but not superficial) cortical layers – brief excitation, followed by a profound off period and a rebound excitation (‘rebound’ is a term often used to describe a period of enhanced spiking following a period of inhibition or silence; Grenier et al., 1998; Guido and Weyand, 1995; Roux et al., 2014). This sequence is repeated in the thalamus, supported by burst firing in excitatory thalamic neurons. Based on relative timing between cortical and thalamic evoked activity, we infer that thalamic bursting is necessary for the late, evoked EEG component seen in response to electrical stimulation, a novel result that links the ERP to activity in the cortico-thalamo-cortical (CTC) loop. Results Global evoked responses are modulated by the depth of the cortical stimulation site We recorded global EEG-like neural signals using a multi-electrode surface array on the skull, but below the scalp (Jonak et al., 2018; Land et al., 2019), in head-fixed mice. The multi-electrode array consisted of 30 electrodes situated over primary and secondary motor, somatosensory, visual, and retrosplenial areas in both hemispheres (Figure 1A). We used an average reference montage that removed signals common to all EEG electrodes (see Methods). We inserted a bipolar wire electrode intra-cortically to repeatedly deliver a single electrical current pulse into the cortex and measured the evoked potentials with the EEG array (Figure 1B and C). The current pulses were biphasic (200 μs/phase) with an amplitude between 10 and 100 μA (details in Figure 1C inset). Stimulation artifacts in the EEG signals were reduced by replacing the signal between 0 and +2 ms following each stimulus with the signal between –2 and 0 ms; this was done to all traces in all trials during an offline, signal pre-processing step. All subsequent analyses of the ERP (the trial-averaged EEG response) exclude the signal between –2 and +2 ms. During the experiment, we closely observed the animal for signs of electrically evoked motor twitches and chose lower stimulation amplitudes if we observed any. Figure 1 with 2 supplements see all Download asset Open asset Evoked EEG responses to single pulse electrical stimulation in awake, head-fixed mice. (A) Schematic of the 30-channel surface array (yellow circles) implanted on top of the skull over major brain areas: motor, somatosensory, retrosplenial, and visual areas (schematic created using brainrender; Claudi et al., 2021). The circular, platinum EEG electrodes are 500 μm in diameter. The three light blue circles correspond to the locations of the three acute craniotomies to place up to three Neuropixels probes and the bipolar stimulating electrode. The schematic also shows two skull screws over the cerebellum that serve as the reference and the ground for the EEG signals. (B) Histological image of a coronal brain slice showing the location of the bipolar stimulating electrode (with tips in secondary motor area [MOs], layer 5; red dashed line) and one of the Neuropixels probes (spanning layers of motor and anterior cingulate areas; blue dashed line) with fluorescent dyes (that appear red and green in the image). (C) Evoked responses from each of the 30 EEG electrodes from the awake, head-fixed mouse from –0.2 to +0.8 s following the electrical stimulus (vertical green line marks the onset time). Traces are arranged in the approximate orientation of the EEG array over the skull surface. Traces in black and gray represent signals that did and did not pass a quality control step, respectively. The red star and blue circles mark the approximate insertion point of the bipolar stimulating electrode and the Neuropixels probes, respectively. Inset: Single current pulses were biphasic (200 μs/phase), charge-balanced, and cathodic-first, with a current amplitude between 10 and 100 μA. (D) Event-related potential (ERP; –0.3 to +1.1 s around stimulus onset) with all EEG electrode traces superimposed (butterfly plots). Each of the four panels represents data from a different stimulated area and depth: top and bottom left – superficial and deep layer (same as in panel C) MOs stimulation in the same subject; top and bottom right – superficial and deep layer primary somatosensory area (SSp) stimulation in a different subject. The dashed vertical line indicates the duration of the evoked signal; the marker above matches with the marker representing the value in panels E and F. The ‘baseline σ’ indicates the SD (in μV) over all electrodes during the 2 s preceding the stimulus. (E) Duration of the ERPs for all subjects based on the stimulation depth: superficial (N=12) vs. deep (N=18). (F) Normalized magnitude of the ERPs for all subjects based on the stimulation depth: superficial (N=12) vs. deep (N=18). For further details, see method ‘ERP duration and magnitude’ and Figure 1—figure supplement 1. Boxplots show median (orange line), 25th, and 75th percentiles; whiskers extend from the box by 1.5× the inter-quartile range (IQR). Student’s two-tailed t-test; * weak evidence to reject null hypothesis (0.05>p>0.01), ** strong evidence to reject null hypothesis (0.01>p>0.001), and *** very strong evidence to reject null hypothesis (0.001>p). To understand how the features of the ERP depend on the location of the electrical stimulation, we varied both the area and the depth (or cortical layer) of the stimulating electrode. We stimulated in the secondary motor area (MOs) or in the primary somatosensory area (SSp) in layer 2/3 (superficial: 0.41±0.04 mm below the brain surface) or in layer 5/6 (deep: 1.06±0.05 mm below the brain surface; Figure 1D). We found that regardless of area, when we stimulated layer 5/6 during the awake period, we usually observed two prominent peaks in the ERP: an initial response around 25 ms and a secondary peak at around 180 ms post-stimulation (Figure 1C and D). The early component was preserved, whereas the second, late component was not evident when stimulating superficially in either area (Figure 1D). The total duration of the ERP was shorter when stimulating superficial layers compared to deep layers although not significant (mean ERP duration for superficial stimulation: 0.4±0.0 s; deep stimulation: 0.5±0.1 s; Student’s two-tailed t-test, p=0.0843; Figure 1E, Figure 1—figure supplement 1). When stimulating superficially, ERPs had a significantly smaller normalized magnitude compared to stimulating deep layers (mean ERP magnitude for superficial stimulation: 3.3±0.8 SD∙s; deep stimulation: 6.2±0.9 SD∙s; Student’s two-tailed t-test, p=0.0354; Figure 1F, Figure 1—figure supplement 1). The amplitude of the first peak in the ERP decays systematically when moving away from the stimulation site until it eventually flips its sign, most likely reflecting volume conduction (Figure 1—figure supplement 2). The magnitude and polarity of the second component likewise changes continuously but in a different pattern, suggesting a different mechanistic origin. The spiking response pattern of the stimulated cortex echoes the ERP In addition to recording EEG signals, we simultaneously collected data from up to three Neuropixels probes – linear silicon probes with a 10 mm long non-tapered shank with 384 simultaneously recorded electrodes capable of capturing LFP and action potentials (Jun et al., 2017). The Neuropixels probes were placed in such a manner as to record from cortex (motor MO, anterior cingulate ACA, somatosensory SS, and visual VIS) and sensorimotor-related thalamic nuclei (SM-TH, Figure 2A), see full list of thalamic nuclei in the Methods section (Guo et al., 2017; Harris et al., 2019; Hooks et al., 2013). The Neuropixels probes were inserted approximately perpendicular to the cortical surface. This allowed us to observe the LFP and the spiking activity of hundreds of individual cortical and thalamic neurons. From the LFP, we inferred the current source density (CSD) using a computational method that assumes an ohmic conductive medium, constant extracellular conductivity (0.3 S/m), and homogeneous in-plane neuronal activity, with the boundary condition of zero current outside the sampled area (Freeman and Nicholson, 1975; Mitzdorf, 1985). Figure 2 Download asset Open asset Electrical stimulation evokes strong responses in the EEG, LFP, CSD, and in some populations of neurons (locally in MO and in SM-TH) when deep layers of MOs are directly activated. (A) Sagittal schema of the mouse brain, highlighting motor (MO), anterior cingulate (ACA), somatosensory (SS), visual (VIS), and somatomotor-related thalamic (SM-TH) areas (created using brainrender; Claudi et al., 2021). Solid black lines show the approximate locations of three acutely inserted Neuropixels probes; the red line indicates the stimulating electrode in the deep layers of MOs. (B) Butterfly plots of the event-related potential (ERP; –0.2 to +0.8 s around stimulus onset) evoked during the awake state. Each column represents data from a different stimulated depth (superficial and deep MOs) in the same subject (same subject shown in Figure 1D left top and bottom). (C) Evoked responses from the Neuropixels electrodes in MO. (Top) From the measured LFP band (black traces representing 1 out of every 10 channels), the CSD response was computationally inferred (heat map, red and blue represent sources and sinks, respectively). The number of Neuropixels channels used to compute CSD is indicated along the y-axis. Bottom: Normalized firing rate, reported as a z-score of the average, pre-stimulus firing rate, of all neurons (only regular spiking [RS] neurons in cortical regions) recorded by the Neuropixels probes targeting the area of interest. The number of neurons (n) in each area is included along the y-axis. (D) Evoked responses from ACA, same as panel C. (E) Evoked responses from VIS, same as panel C. (F) Evoked responses from SM-TH, CSD was not computed for SM-TH because thalamic structures, unlike cortex, do not contain oriented neural elements; therefore, it would not be interpretable. We inserted a Neuropixels probe near the stimulating electrode (within 0.5 mm) and, often, up to two additional Neuropixels probes at distal locations. We observed direct responses (i.e. defined operationally as neurons that spike between 2 and 25 ms following the electrical pulse; this might miss a handful of very rapidly (i.e.<2 ms) responding neurons; see Sombeck et al., 2022; Stoelzel et al., 2009) as well as indirect responses to the electrical stimulation. To look for cell-type specific activation patterns, we classified regular spiking (RS) and fast spiking (FS) neurons (putative pyramidal and putative parvalbumin-positive inhibitory neurons, respectively) based on their spike waveform duration (RS duration >0.4 ms; FS duration ≤0.4 ms; Barthó et al., 2004; Bortone et al., 2014; Bruno and Simons, 2002; Jia et al., 2019; Niell and Stryker, 2008; Sirota et al., 2008). Stimulating superficial MOs evoked a local spike, LFP and CSD response, followed by a period of quiescence lasting 94.2±16.1 ms in MO neurons (Figure 2C, left). There were minimal to no evoked responses (LFP, CSD, or spikes) in other sampled areas (Figure 2D–F, left). Stimulating deep MOs evoked a robust change in LFP and CSD, accompanied by spiking activity reflective of the two components in the ERP (Figure 2B right): an initial excitation within 25 ms (peak population firing rate 38.1±4.2 Hz), followed by quiescence (duration 125.0±5.5 ms) and a longer period of strong excitation (peak population firing rate 7.9±0.7 Hz; Figure 2C, right). This cortical response pattern was quite stereotyped – an initial excitation, followed by an off period and a strong rebound excitation – and was also observed in SM-TH (Figure 2F, right). We seldom observed this pattern for superficial stimulation (Figure 2C, left). Indeed, on average, about three times more cortical neurons respond significantly to deep to superficial stimulation (Figure and D). Global ERPs are associated with widespread evoked cortical activity by the of firing in cortical we the evoked magnitude of the LFP, CSD, and spiking in response to superficial and deep stimulation that magnitude for both as well as and is calculated as the of the z-score relative to see method CSD, and population spiking Figure The magnitude of the LFP near the stimulation site is activity in and decays with from the stimulation site in Figure the LFP the of local activity as well as volume conduction from more current sources and we examined the CSD responses, which the of volume conduction and local current The magnitude of the CSD shows a similar in the stimulated cortex in and smaller at more sites in the response is in both (Figure while we see robust evoked firing of cortical neurons in to the stimulation changes in firing are often much smaller in in Figure Figure Download asset Open asset stimulation of but not cortical layers evokes widespread (A) Evoked local field potential –0.2 to s around stimulus onset) from deep MOs stimulation (same subject as Figure 1D bottom all recorded cortical sites along the Neuropixels are The two represent data from stimulated and a cortex The magnitude is the area of the response to s from stimulus relative to the area of the activity calculated by the stimulation onset (B) Evoked current source density (CSD) from deep MOs stimulation, all recorded cortical sites along the Neuropixels are (C) Evoked neuronal firing rates from deep MOs stimulation, each represents a single (D) of the evoked LFP CSD and population spiking to superficial and deep stimulation for stimulated and cortical and dashed gray line at stimulation: stimulated in mice and cortical in mice. stimulation: stimulated in mice and cortical in mice. Boxplots show median (orange line), 25th, and 75th percentiles; whiskers extend from the box by 1.5× the for using false * weak evidence to reject null hypothesis (0.05>p>0.01), ** strong evidence to reject null hypothesis (0.01>p>0.001), and *** very strong evidence to reject null hypothesis (0.001>p). Stimulating superficial layers significantly modulated local LFP, CSD, and cortical spiking with to LFP median inter-quartile for using false rate, CSD spiking Figure left also modulated LFP and CSD in cortical areas LFP CSD to a compared to the stimulated We did not evoked spiking outside of the stimulated area for superficial stimulation When we stimulated deep we observed evoked LFP, CSD, and cortical spiking LFP for using false rate, CSD spiking Figure right In to superficial stimulation, deep responses in cortical were significantly CSD spiking these data show that cortical stimulation evoked widespread changes in the LFP and CSD, but evoked cortical spiking outside of the stimulated area when stimulating deep To further investigate the evoked neural spiking across both stimulated and we computed the of modulated or spiking relative to the pre-stimulus neurons subject in three following the ms ms and ms Figure The three were to the stereotyped firing patterns, which with peaks in the ERPs and the period in between (Figure 1—figure supplement 2). We FS neurons and found they neurons (Figure supplement therefore, further analyses on neurons. Figure with 1 supplement see all Download asset Open asset The of the population cortical area in each subject that is significantly modulated on the depth of stimulation and the (A) of regular spiking (RS) neurons cortical area in each subject that a significantly or response in the first 25 ms for superficial stimulation. of neurons that a significant or (B) ms or (C) ms following the stimulus. as panels but for deep stimulation. are from neurons from stimulated cortex and neurons from cortical in mice and panels from neurons from stimulated cortex and neurons from cortical in mice. Boxplots show median (orange line), 25th, and 75th percentiles; whiskers extend from the box by 1.5× the During the initial excitation, the of spiking relative to neurons was in the stimulated cortex other cortical for both superficial and deep stimulation. The of neurons in cortical was above zero (Figure and D). Indeed, across all an average of and of the neurons was modulated in the initial excitation, compared to an average of and in stimulated we the of modulated neurons in the off period between 25 and capturing the period between the two ERP amplitude components with deep Figure 1—figure supplement 2). There were very neurons with spiking, regardless of stimulation depth and area (mean and Figure and In spiking in subjects was significantly reduced from following either superficial or deep stimulation. some areas showed spiking but in response to deep stimulation (mean and The we was the rebound excitation between and with the second component in the All subjects deep stimulation showed significant rebound excitation in the stimulated cortex, whereas superficial stimulation subjects did (Figure and subjects deep stimulation showed significant rebound excitation in the areas Figure but it was the superficial stimulation subjects Figure To deep cortical stimulation is more likely to significantly single neurons outside of the stimulated cortical This consisted of decreases in firing in the ms and and decreases in firing during the ms (Figure interactions underlie features of the ERPs The cortical spiking pattern for deep stimulation is in the associated thalamic nuclei that are to the stimulated cortical area et al., 2017; Harris et al., 2019; Hooks et al., 2013; bottom right in Figure in we observed a brief excitation (peak population firing rate Hz), a ms long period of followed by rebound excitation (peak population firing rate This is not the when stimulating superficial layers (Figure 2F, bottom left). We the of significantly modulated neurons in the SM-TH population and found that the neurons were times more likely to be modulated by deep superficial stimulation (superficial: of SM-TH Student’s two-tailed t-test, Figure Figure Download asset Open asset and thalamic neural dynamics evoked by cortical electrical stimulation. (A) of somatomotor-related thalamic (SM-TH) neurons that a significant or in firing rate compared to between 2 ms and ms following stimulus onset for superficial and deep stimulation (B) traces from the Neuropixels spike band data for motor and SM-TH to for one with deep MOs stimulation. potential in the SM-TH are with (C) of total that the SM-TH within ms from stimulus onset for superficial and deep stimulation (D) Single showing spiking of MO and SM-TH neurons in response to a single deep electrical pulse the green vertical potential are with (E) to ms) for neurons in stimulated cortex and in associated SM-TH are recorded simultaneously in each by the black lines (F) to spike in the late ms) for neurons in stimulated cortex and in SM-TH as in panel Boxplots show median (orange line), 25th, and 75th percentiles; whiskers extend from the box by