Nine (09) Long Evans male rats (Rattus norvegicus), weighting 250–300 g, obtained from Charles River (Milano, Italy), were used in this study. They were housed at the vivarium of IIT’s Animal Facility for the entire duration of the experiment, being kept in a 12-h light–dark cycle (lights on at 7 a.m. and off at 7 p.m.), at an average temperature of 23 °C ± 2 °C, with food and water ad libitum. Animals were equally divided into two groups, controls (NO LESION; N = 5) and lesion (LESION; N = 4). The sample size was consistent with previous rodent studies combining focal ischemia models and chronic multichannel electrophysiology (Carè et al., 2022). All animals underwent surgical procedure for the implantation of electrodes in two cortical regions of the sensorimotor loop, the Rostral Forebrain Area (RFA) and the primary somatosensory area (S1) (Figure 1A, bottom left inset), as well as 6 h-long electrophysiological recordings (Figure 1A, bottom right inset). Animals in the LESION group were subjected to the induction of an ischemic lesion (see below) in the Caudal Forelimb Area (CFA). All experiments were previously approved by the Italian Ministry of Health (license 513/2022) and were in accordance with international guidelines for the care of animals in research.

To begin the surgical procedure, initial anesthesia was induced with the administration of gaseous isoflurane (5% at 1 lpm) using a dedicated inhalation chamber for the animal. This was followed by the administration of ketamine (intraperitoneal, 80–100 mg/kg; intramuscular, 10–20 mg/kg) and xylazine (intramuscular, 5 mg/kg). After trichotomy of the head, protection of the eyes with ophthalmic ointment, and appropriate aseptic measurements with Betadine and 70% alcohol, animals were positioned in the stereotaxic apparatus. Body temperature was monitored by a rectal probe and controlled by a heat pad using a physiological monitoring system, Physio Suite® (Kent Scientific, Torrington, USA). The sedation status of the animal was continuously monitored by the experimenters, checking for animal twitches through a simple pinch test and, when necessary, maintenance boluses of ketamine (intramuscular, 0.10–0.17 mL) were administered. The skull surface was exposed with a rostro-caudal incision using a scalpel, followed by dissection of the periosteum, and cleaning of the bone with hydrogen peroxide. Furthermore, the cisterna magna was also exposed, and a laminectomy was performed to control brain edema, by allowing excess cerebrospinal fluid drainage. In the sequence, craniectomy was performed to access the primary somatosensory cortex (S1: −1.25 mm AP and +4.25 mm ML, referenced from the Bregma) and the rostral forelimb area (RFA, a pre-motor area: +3.5 mm AP and 2.5 mm ML, referenced from the Bregma) using a drill with a burr bit. Additional six holes were predisposed with a small drill bit into the parietal and intraparietal bones to later insertion or surgical microscrews that served both as reference and grounding, as well as to secure the implant on the animal for the chronic recordings. Dura-mater was carefully removed using a bent hypodermic needle to avoid excessive brain tissue dimpling and electrode bending and damage. Electrodes (32 channel microwire arrays, Microprobes, USA), one for each area, were slowly (0.1 mm/min) inserted into the cortex to a depth of 1.4 mm referenced from the cortical surface. Implantation sites were sealed using low toxicity adhesives for live tissues (Kwik-cast; WPI, Sarasota, USA). The whole assembly, including both electrode arrays, was secured onto the animal’s head with dental auto polymerizing acrylic cement anchored by the bone microscrews. Finally, the skin opening was sutured, the animals were cleaned, received antibiotics and anti-inflammatory drugs, and remained under observation in a heated cage until they were fully awake and moving. All animals were left to rest for 7 days before beginning experiments. During this period, animals were monitored daily for signals of discomfort or insufficient healing and were treated with analgesics, antibiotics, and anti-inflammatory drugs on the second and third days.

Animals in the LESION group were submitted to the ischemic stroke model by microinjection of Endothelin-1 (ET-1), a potent vasoconstrictor (Frost et al., 2006). During the surgical procedure, before the insertion of electrode arrays, six small holes were drilled in the skull centered around the CFA, equivalent to human primary motor cortex (combining coordinates at AP: +1.5 mm, +0.5 mm, and −0.5 mm; ML: +3.5 mm and +2.5 mm, referenced from Bregma). ET-1 was then loaded into a Hamilton syringe which was lowered 1.2 mm into brain tissue. A volume of 0.33 mL per hole (thus a total of 2 mL) was injected, divided into three steps of 0.11 mL gradually administered at a 3 nL/s rate; sessions being interspaced by 1 min. This guaranteed a seamless absorption of the vasoconstrictor agent volume without damage by tissue displacement (Carè et al., 2022).

In order to lower stress levels, avoid excessive exploratory behavior, and maximize natural expression of sleep–wake cycle, all animals were handled by the same experimenter for 15 min and left to explore the controlled environment (i.e., a squared arena with a base of 60 × 60 cm, and 60 cm height) for another 15 min, for 5 days, twice a day, before surgery (Figure 1A, experimental timeline). In the day of the recording session, animals were lightly anesthetized with gaseous isoflurane (5% at 1 lpm) for cleaning of electrode array contacts and for the placement of recording headstages and cables. Animals were then placed in the arena and cables connected to the electrophysiological setup. The system was composed of headstages (equipped with RHS 2000 series chips; Intan Technologies LCC, Los Angeles, USA) for amplification (192 V/V gain), analog bandpass filtering (1 Hz–7.5 kHz), and digitalization (16 bits resolution at 25 kHz), connected by an SPI cable to the controller for PC communication via USB. Wideband recording signals were then low-pass filtered at 300 Hz and downsampled to 1 kHz using the NigeLab pipeline for MATLAB developed at our lab,¹ already adopted in previous studies (Carè et al., 2022). Furthermore, a bipolar referencing method was chosen, and 50 Hz line noise and its harmonics were excluded with a band-stop FIR filter achieving 53 dB suppression with 1 Hz band-stop widths using custom-made Python functions specifically developed for this study. Recordings were carried out at least 1 h after termination of gaseous anesthesia for the proper metabolization of the drug, from 10 a.m. to 4 p.m. (±30 min). The NO LESION animals were subjected to a single recording session, while LESION animals underwent two recording sessions at 7 and 14 days post lesion. Specifically, recordings were scheduled to allow full recovery from surgery and the acute ischemic phase (day 7) as well as to capture the early subacute stage of stroke recovery (day 14), when reorganization processes begin (Boboc et al., 2023).

Sleep stages were automatically detected by clustering points in a state map derived from spectral content ratios (Gervasoni et al., 2004) (Figure 1B). Briefly, after removing channels plagued by movement (characterized by constant drastic increases of amplitude with signal clipping), the power spectrum of 2-s windows from LFPs recordings was computed separately for each channel at 1-s intervals (with 50% overlap). Two spectral power ratios were then calculated: ratio 1 [(0.5–20 Hz)/(0.5–55 Hz)] and ratio 2 [(0.5–4.5 Hz)/(0.5–9 Hz)]. Principal Component Analysis (PCA) was applied to these ratios, and the resulting time series of the two first principal components (1st PC of ratio 1 and 1st PC of ratio 2) were smoothed using a 20-s Hanning window. The smoothed PC values were plotted along the X (ratio 2) and Y (ratio 1) axes of the state map, respectively. Clusters in this map were then identified and assigned to one of the three major sleep–wake cycle (SWC) states, wakefulness (WK), slow-wave sleep (SWS), or rapid eye movement (REM) sleep, based on their position, following the method described in the literature (Gervasoni et al., 2004). The window size of 2 s was selected to enable the calculation of the lower limit of band ratios, while the band ranges were empirically determined and repeatedly shown in the literature to effectively separate sleep states in the state map (Gervasoni et al., 2004; Dzirasa et al., 2006; Blanco et al., 2015; Réboli et al., 2022). For the analyses of power, phase synchronization, and phase–amplitude coupling (described below), we empirically selected SWS segments of 120 consecutive seconds that were artifact-free (e.g., movement, chewing, or environmental artifacts) and contained no bad channels, defined as those with baseline drift or poor signal-to-noise ratio, after visual inspection. A total of 48 segments were included for the NO LESION group (mean ± SD = 10 ± 3 segments per animal), 91 segments for the 7-day post-LESION group (23 ± 2 segments per animal), and 94 segments for the 14-day post-LESION group (23 ± 5 segments per animal). All metrics considered for this work, and described below, were computed for each segment and later averaged over them. Figure 1C shows an example of raw LFP traces with bipolar referencing from a 120-s segment of SWS, including four channels each from the S1 and RFA areas.

To estimate spectral power during slow-wave sleep in animals, we computed the Power Spectral Density (PSD) by employing Welch’s method with a sampling rate of 1,000 Hz and a Hanning window length of 4,096 samples, corresponding to a frequency resolution of approximately 0.24 Hz. A 50% overlap was applied between the windows. PSD estimates were then averaged across channels within each region of interest, i.e., S1 and RFA.

To investigate both local and inter-areal phase synchronization, we used the Phase-Locking Value (PLV). PLV is derived as the absolute value of the complex PLV (cPLV) (Palva et al., 2018; Arnulfo et al., 2020). We band-pass filtered the LFP data using 43 Morlet wavelets (width m = 7.5) with center frequencies ranging from 1.2 to 200 Hz and we then computed the cPLV between two signals x and y, defined as:

Where x’(t) and y’ represent the complex wavelet coefficients of the signals at a given frequency, K represents the total number of samples of the entire signal, and * denotes the complex conjugate. The PLV provides a scalar measure between 0 and 1, with 1 indicating complete phase synchronization and 0 indicating complete desynchronization. An example of how PLV is computed is shown in Figure 1D.

Synchronization matrices were further analyzed using graph-theoretical metrics. Specifically, eigenvector centrality (EVC) was computed to estimate each channel’s centrality within the functional connectivity network during slow-wave sleep. These metrics were aggregated at the subject level and subsequently correlated with total sleep duration to investigate potential associations between network organization and sleep stability (cf. Statistical analysis section).

Phase-amplitude coupling (PAC) provides information about the correlation between the phase of the slower oscillation and the amplitude of the faster one (Vanhatalo et al., 2004; Siebenhühner et al., 2020). PAC was computed between selected low-frequency (LF) and high-frequency (HF) pairs as follows:

Where θ x , LF represents the phase of a given low frequency signal and θ y , HF , LF env denotes the amplitude envelope of high frequency signal filtered at the same frequency of the LF. Like the PLV, the PAC provides a measure ranging from 0 to 1, with 1 denoting full phase amplitude interaction and 0 indicating no interaction. For our analysis we considered the normalized PAC (nPAC = PAC_PLV,obs/PAC_PLV,sur), representing PAC above the null hypothesis level. The following low-frequency (LF) bands were used for the computation of phase–amplitude coupling (PAC): [1.2, 2.4, 3.7, 5.9, 8.6, 13.2, 19.5, 29.5, 47.3, 68.1] Hz. For each LF component, corresponding high frequency (HF) bands were generated by multiplying the LF value by a set of predefined ratios (HF = LF*ratio). The ratio considered here are: [2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 14.06, 17.96, 22.96, 29.34, 37.49, 47.91, 61.23, 78.25, 100, 110, 149.11, 202.13]. The nPAC were discarded if the resulting HF component exceeded 200 Hz. To facilitate readability and matrix visualization, nPAC results were projected from the low-frequency-to-ratio format into a low-frequency-to-high-frequency space. Subsequently, nPAC values were averaged across all low-frequency bands for each high-frequency band, yielding a mean nPAC value for each high-frequency. We also differentiated between local and inter-areal nPAC. Intra-area nPAC measures phase-amplitude interactions within electrodes in the same brain region, while inter-areal nPAC evaluates interactions between the phase of low frequencies in one electrode and the amplitude of high frequencies in another electrode located in a different region. An example of how nPAC is computed is shown in Figure 1E.

To assess the validity of our PLV and PAC computation, we compared them to equivalent analyses performed on surrogate data. We designed surrogate data to disrupt inter-channel correlations while preserving the temporal autocorrelation structure of the original signals. Specifically, for each pair of narrow-band time series, we selected a random time point k uniformly along the signal and split each signal into two segments. A surrogate signal was then constructed by recombining mismatched segments from the two channels:

This procedure maintains the intrinsic temporal structure of each signal, while interactions between the signals are disrupted. Surrogate analysis was applied with two distinct purposes depending on the metric. For PLV, surrogate levels were included in the main graphs of the phase synchronization spectra to provide a visual reference for the noise level in the data, whereas for PAC, surrogate analysis was used to normalize the PAC values, as described in detail in the corresponding analysis section (cf. Phase amplitude coupling section).

Due to substantial data heterogeneity, we employed multiple statistical approaches to test for differences in PLV and PAC across frequency bands and regions of interest between experimental conditions (i.e., NO LESION, 7 days post-lesion, and 14 days post-lesion). Before performing all comparisons, we first assessed the normality of the data using the Shapiro–Wilk test, which confirmed that the data were normally distributed. To compare the no lesion group with animals at 7 days post-lesion, we used the Mann–Whitney U test to assess differences in central tendency, and the Fligner–Killeen test with the Coefficient of Variation (CV) to assess differences in variability. For the within-subject comparison between 7 and 14 days post-lesion in the same animals, we employed a paired t-test for paired differences in central tendency, and a permutation test on the logarithm of the variance, alongside the CV, to evaluate changes in variability. These tests were chosen considering the unpaired and paired conditions of the former and latter comparisons, respectively. In addition, effect sizes were computed to provide a measure of the magnitude of observed differences independently of sample size, with Cohen’s d being used for pairwise comparisons. Moreover, linear mixed-effects models (lme4 package in R (Bates et al., 2015) were employed to test for correlations between sleep duration and PLV, as well as for changes in these correlations across experimental time points (emmtrends function, emmeans library in R (Lenth and Piaskowski, 2025). We chose these two metrics as evident tokens of behavior and brain activity, given the well-established notion that the architecture of SWC (including its duration)—and the neuroplastic phenomena supported by it—are largely dependent on the proper coordination and synchronization of neural oscillations (PLV eigenvalue centrality). For all statistical tests, p-values were corrected for multiple comparisons across frequencies using the Benjamini–Hochberg procedure (α = 0.05). Frequency bands are referred to here by their canonical names: delta (0.5–4 Hz), theta (4–8 Hz), alpha (8–12 Hz), beta (12–30 Hz), and gamma (30–100 Hz).

To assess the impact of ischemic lesions on neural oscillatory activity during SWS, we analyzed the PSD of LFPs recorded from electrodes implanted in S1 and RFA (see Methods), Figure 2. Comparisons were made between the NO LESION group and the LESION group at 7 days post-lesion (Figure 2A), as well as between recordings at 7 and 14 days post-lesion (Figure 2B). In S1 (Figure 2A, left) the PSD at 7 days post-lesion did not differ significantly from the NO LESION group (p = 0.29), although a medium effect size was observed (Supplementary Figure 1A, left). Intersubject variability appeared elevated in the delta band [1–4 Hz], as reflected by a higher coefficient of variation (CV) in the 7-day post-lesion group compared to controls (i.e., NO LESION). However, this difference was not statistically significant (p = 0.98), and the corresponding log-ratio of variances (i.e., lnVR = 1.98; Supplementary Figure 1A, left). In contrast, RFA (Figure 2A, right) showed a reduction in PSD between 4–50 Hz frequency range at 7 days post-lesion. Although this change was not statistically significant (p = 0.80), the effect size was medium (Supplementary Figure 1A, right), This reduction was also reflected in the CV, which showed lower values in the same frequency range after lesion, even if without statistical significance (p = 0.5). The log-ratio of variances was −1.2 (Supplementary Figure 1A, right).

Between 7 and 14 days post-lesion, PSD in S1 (Figure 2B, left) remained stable (p = 0.90), with a negligible effect size (Supplementary Figure 1B, left). The CV was lower at 14-day than at 7-day in the delta band [1–4 Hz], although this difference was not statistically significant (p = 0.36) with lnVR of −2.53 (Supplementary Figure 1B, left). In RFA (Figure 2B, right), PSD across the full spectrum showed an increase at 14 days post-lesion (p = 0.77) with medium effect size (Supplementary Figure 1B, right). This increase was accompanied by a rise in CV, although no statistically significant differences were found in variability (p = 0.26) and with lnVR of 2.42 (Supplementary Figure 1B, right).

Next, we assessed whether ischemic lesions affected phase synchronization during SWS (Figure 3), by comparing the 7 day NO LESION and the 7 day LESION groups. In S1 (Figure 3A, left), within the delta band [1–4 Hz], PLV was higher post-lesion than in the NO LESION condition (p = 0.52) with a small effect size (Supplementary Figure 2A, left). In contrast, at frequencies above the delta band, PLV of LESION group began to decrease (p = 0.34), showing a large effect size (Supplementary Figure 2A, left). This reduction was accompanied by a decrease in the CV compared to the NO LESION group (p = 0.25) with a mean lnVR of −1.94 (Supplementary Figure 2A, left). At 7 days post lesion, in RFA (Figure 3A, middle), an increase in the delta band was observed (p = 0.53), with a large effect size (Supplementary Figure 2A, middle). CV values increased, though not significantly (p = 0.92), with a mean lnVR of 0.8 (Supplementary Figure 2A, middle). Inter-areal PLV between S1 and RFA (Figure 3A, right) showed increased delta synchronization post-lesion, even if not statistical (p = 0.82) with small effect size (Supplementary Figure 2A, right), and decreased theta [4–8 Hz] -band PLV (p = 0.56) with large effect size (Supplementary Figure 2A, right). This was accompanied by a significant reduction in variability, as indicated by decreased CV and a significant Fligner’s test result (p = 0.035), with a mean lnVR of −2.1 (Supplementary Figure 2A, right). PLV values above the theta band dropped to near-surrogate levels in both LESION and LESION conditions (Figure 3A, right).

For PAC, the spectrum in S1 (Figure 3B, left), RFA (Figure 3B, middle), and directional coupling from S1 → RFA (Figure 3B, right) showed similar patterns in NO LESION and 7 days post-lesion up to 20 Hz (p = 0.27, 0.20, and 0.81, respectively), with strong effect sizes in S1 and RFA (Supplementary Figure 2B, left and middle). In both S1 (Figure 3B, left) and RFA (Figure 3B, middle), PAC above 20 Hz was higher post-lesion (p = 0.15 and 0.17, respectively) with large effect (Supplementary Figure 2B, left and middle). The CV in the LESION group decreased in RFA and in S1 → RFA (Figure 3B, middle and right). However, no frequency bands showed statistically significant differences in variance after correction for multiple comparisons, despite trends approaching significance (p = 0.10) with a lnVR of 3.36 and lnVR of 3, respectively, (Supplementary Figure 2B, middle and right). We also evaluated the modulation from RFA to S1 (see Supplementary Figure 4A); however, no notable differences were observed between the two groups.

To assess how phase synchronization evolved over time, we compared PLV and PAC between 7 and 14 days post-lesion during SWS within the LESION group (Figure 4). In S1 (Figure 4A, left), PLV in both the delta and theta bands resulted reduced at 14 days compared to the 7 days post-lesion (p = 0.77), with a large effect size (Supplementary Figure 3A, left). At 14 days, the CV in S1 increased (p = 0.4), with a mean lnVR of 1.2 (Supplementary Figure 3A, left). A similar pattern was observed in in RFA (Figure 4A, middle), with reductions of the PLV in the delta and theta-band at 14 days (p = 0.33), and a large effect size (Supplementary Figure 3A, middle). The CV above 10 Hz also increased at 14 days (p = 0.82), with a mean lnVR of 1.1 (Supplementary Figure 3A, middle). Inter-areal PLV between S1 and RFA (Figure 4A, right) showed a decrease in delta-band synchronization (p = 0.73) with a large effect size (Supplementary Figure 3A, right), and an increase in theta-band PLV at 14 days post-lesion (p = 0.9), with a medium effect size (Supplementary Figure 3A, right). In line with these spectral changes, the CV in inter-areal synchronization decreased in the delta band (p = 0.77) with a mean lnVR of −1.2 (Supplementary Figure 3A) and increased in the theta band (p = 0.8) with a mean lnVR of 0.86 (Supplementary Figure 3A). PLV values at frequencies above the theta band dropped to near-surrogate levels at both experimental groups.

As regards the PAC analysis, in S1 (Figure 4B, left), within the delta band and for frequencies above 10 Hz, PAC values were significantly higher at 7 days post-lesion compared to 14 days post-lesion (p = 0.04 and p = 0.018, respectively), with a large effect size (Supplementary Figure 3B). The CV in S1 above 10 Hz increased at 14 days, although the changes did not reach statistical significance in any frequency band (p = 0.08) and a mean lnVR of −1.58 (Supplementary Figure 3B, left). PAC spectral profiles in RFA(Figure 4B, middle) and in the S1 → RFA direction (Figure 4B, right) remained unchanged between 7 and 14 days post-lesion for frequencies up to 20 Hz (p = 0.23 and p = 0.75, respectively), despite large effect sizes (Supplementary Figure 3B, middle and right). Above 20 Hz, however, a reduction in PAC was observed at 14 days in both regions. PAC showed p-values approaching the significance threshold (p = 0.075), along with large effect sizes (Supplementary Figure 3B, middle and right).). CV in the RFA and S1 → RFA direction remained almost stable over time for both groups (p = 0.08) and a lnVR of −1.5 (Supplementary Figure 3B, middle and right). Finally, modulation from RFA to S1 was also evaluated, but no notable differences were observed between the groups (see Supplementary Figure 4B).

To determine how SWS duration shapes functional network organization after a focal CFA lesion, we fitted linear mixed-effects models testing the slope of normalized SWS duration against square-root–transformed eigenvector centrality of PLV networks, Figure 5. Trends were estimated with emtrends and significance was corrected for multiple comparisons by the Benjamini–Hochberg method. Full statistics are provided in Supplementary Table 1.

The association between sleep duration and network centrality displayed robust frequency- and region-specific changes following the lesion. Across both regions, centrality estimates in the gamma band during SWS were positively associated with sleep duration in healthy animals (NO LESION). However, in lower frequency bands, this relationship was markedly negative in S1 and close to neutral in RFA, indicating a functional divergence between the two areas even in absence of the injury.

Focusing first on S1, where effects were strongest and most consistent, a pronounced inverse relationship between sleep duration and centrality was observed at baseline across delta, theta, alpha, and beta bands (e.g., delta: t = −6.06, p < 0.001; gamma: t = −8.81, p < 1e−15). After the lesion, this association weakened significantly, as shown by positive and statistically significant day-to-day contrasts. For instance, from the NO LESION to the LESION group at day 7, the delta band slope changed by +0.10 (t = −4.34, p = 0.00012), and the gamma band by +0.30 (t = −8.52, p = 9.1e−15), clearly indicating a reduction in the negative association.

In RFA, the lesion also disrupted sleep–PLV associations, though effects were more subtle and frequency-dependent. In NO LESION condition, PLV centrality in RFA exhibited a mild positive association with sleep duration, increasing with frequency (e.g., beta: t = 4.70, p < 0.001; gamma: t = 5.95, p = 7.5e−8). In presence of the lesion, however, these associations largely vanished, especially at lower frequencies, and only partially recovered by Day 14. The day-to-day contrasts show this flattening clearly. From the NO LESION to the LESION at Day 7 condition, the slope in the alpha band decreased significantly (t = 2.80, p = 0.015), as did theta (t = 2.34, p = 0.034) and beta (t = 3.88, p = 0.002). Gamma again showed the strongest disruption (t = 5.02, p = 4.1e−6).

In S1, the association between normalized sleep duration and PLV eigenvector centrality remained altered between Day 7 and Day 14 post-lesion, showing no signs of recovery across any frequency band. At lower frequencies, including delta, theta, and alpha, the relationship remained negative, with significant slope differences between the two time points (delta: t = −4.34, p = 1.2 × 10⁻⁴; theta: t = −5.10, p = 2.1 × 10⁻⁵; alpha: t = −5.99, p = 4.0 × 10⁻⁶). Similarly, in the beta band the lesion-induced disruption persisted over time (t = −7.98, p = 1.7 × 10⁻⁷), while in the gamma range the centrality remaining negatively associated with sleep duration even 14 days post lesion (t = −8.52, p = 9.1 × 10⁻¹⁵).

In contrast, RFA exhibited frequency-dependent effects. At lower frequencies, in the delta band, the slope differences between Day 7 and Day 14 did not reach statistical significance. However, at higher frequencies the lesion effects remained evident: in the alpha band, the association between sleep duration and centrality was significant (t = 2.34, p = 0.034), while in the beta band the positive baseline relationship was reduced (t = 3.88, p = 0.002). The gamma band showed a significant disruption persisting up to Day 14 (t = 5.02, p = 4.1 × 10⁻⁶).