LPS increases serum IL-1β
To assess the effect of LPS on serum inflammatory cytokines, IL-1β levels were measured using ELISA. IL-1β concentrations in LPS-injected rats were at least 3 times higher than in saline-injected controls. (398 ± 32.24 pg/ml vs 122 ± 63.64 pg/ml, n = 6, independent sample t-test t(10) = 3.88, p < 0.01, Fig. S1).
LPS injection causes sleep state fragmentation
Under urethane anaesthesia rats showed two distinct patterns of brain activity: a NREM-like state that was dominated by high amplitude slow waves, and a REM-like state with low-amplitude, higher frequency activity (Fig. 1B-E). The states were long and stable at baseline and in saline-injected controls (Fig. 1B-D), but much shorter after LPS injection (Fig. 1E).
The total number of episodes per recording period was significantly affected by LPS injection (rmANOVA; group effect: F(1, 10) = 5.14, p = 0.04; time effect F(1, 10) = 8.64, p = 0.01; group*time interaction: F(1, 10) = 8.55, p = 0.01, Fig. 1F-G). Post-hoc analysis showed a significantly higher number of episodes after LPS injection (87 ± 20 episodes) compared to before injection (24 ± 16 episodes). As a result, episode duration was also significantly altered (rmANOVA; group effect F(1, 10) = 0.03, p = 0.85; time effect: F(1, 10) = 10.46, p < 0.01; group*time interaction: F(1, 10) = 5.58, p = 0.04). Post-hoc analysis showed that episodes in the LPS group were significantly shorter after LPS injection (167 ± 36 s) than at baseline (694 ± 153 s).
Extensive sleep fragmentation after LPS injection was further apparent in episode length distribution (Fig. 2A). At baseline and after saline injection rats had few short episodes (20s -120s) and a relatively high number of long episodes (≥ 600 s). However, after LPS injection, brain activity patterns consisted of many short episodes and only a few long ones. These effects were present in both NREM and REM.
The relative number of short episodes in the recording period showed a significant time effect (rmANOVA; F(1, 10) = 4.75, p = 0.05) and group*time interaction (F(1, 10) = 8.63, p = 0.01), but no significant group effect (F(1, 10) = 1.08, p = 0.32). Post-hoc analysis showed an increase in the short episodes in the LPS group post injection (75 ± 4.6% of episodes) compared to pre-injection (44.1 ± 10.9% of episodes).
When quantified separately, relative amounts of short episodes in NREM were significantly higher in NREM after LPS injection, but not in REM. Short NREM episodes showed a significant time effect (rmANOVA; F(1, 10) = 4.67, p = 0.05) and a non-significant group effect (F(1, 10) = 0.25, p = 0.62). There was a significant effect of group*time interaction (F(1, 10) = 19.78, p < 0.01). Post-hoc analysis showed a significant increase in the percentage of short NREM episodes in the LPS group post injection (81.44 ± 4.58% of NREM episodes vs 46.59 ± 10.37% pre-injection, Fig. 2D). In REM, there was a significant group effect (rmANOVA; F(1, 10) = 6.58, p = 0.02) but a non-significant time effect (F(1, 10) = 4.01, p = 0.07) and group*time interaction (F(1, 10) = 2.56, p = 0.14). Post-hoc analysis showed a significant baseline difference between controls (70.12 ± 1.94%) and the LPS group (41.52 ± 11.88%) (Fig. 2C).
The percentage of long episodes in the recording period was also affected by LPS injection, but showed opposite effects to those observed for short episodes. The total percentage of all long episodes showed a significant effect of time (F(1, 10) = 9.39, p = 0.01) and group*time interaction (F(1, 10) = 9.18, p = 0.01), but no significant group effect (F(1, 10) = 1.00, p = 0.34). Post-hoc analysis showed a significant decrease in the percentage of long episodes in the LPS group post-injection (6.07 ± 1.55% of episodes) compared to pre-injection (42.20 ± 10.43% of episodes), although absolute numbers of long episodes are quite low.
Effects of LPS injection on NREM and REM states separately paralleled those found in the overall episode length distribution. The percentage of long NREM episodes showed a significant effect of time (rmANOVA; F(1, 10) = 8.28, p = 0.01) and group*time interaction (F(1, 10) = 9.40, p = 0.01). There was no significant group effect (F(1, 10) = 0.79, p = 0.39). Post-hoc analysis showed a decrease in the percentage of long NREM episodes after LPS injection (5.90 ± 1.55% of NREM episodes) compared to baseline (39.97 ± 10.48% of NREM episodes, Fig. 2F). Long REM episodes also showed a significant time effect (rmANOVA; F(1, 10) = 9.78, p = 0.01) and group*time interaction (F(1, 10) = 8.39, p = 0.01), but no significant group effect (F(1, 10) = 0.89, p = 0.36). Post-hoc analysis showed a significantly lower percentage of long REM episodes in the LPS group post-injection (6.51 ± 2.12% of REM episodes) compared to pre-injection (44.16 ± 10.42% of REM episodes, Fig. 2E). These observed effects on episode number and duration point to fragmentation in both the REM- and NREM- state.
However, despite overall changes in episode length, the time spent in NREM (rmANOVA; time effect: F(1, 10) = 0.002, p = 0.96; group effect: F(1, 10) = 0.77, p = 0.39; group*time interaction : F(1, 10) = 0.199, p = 0.66) and in REM (rmANOVA; time effect: F(1, 10) = 0.002, p = 0.96, group effect: F(1, 10) = 0.77, p = 0.39; group*time interaction: F(1, 10) = 0.199, p = 0.66) was not significantly changed by LPS injection (Fig. 1H-I).
State fragmentation was also seen on the epoch-to-epoch level, where the probability of a state transition from one epoch to the next was similarly increased for NREM to REM transitions (rmANOVA; time effect: F(1, 10) = 16.72, p < 0.01; group effect: F(1, 10) = 4.61, p = 0.06; group*time interaction: F(1, 10) = 10.53, p < 0.01) and REM to NREM transitions (rmANOVA; time effect: F(1, 10) = 16.12, p < 0.01; group effect: F(1, 10) = 4.77, p = 0.05; group*time interaction: F(1, 10) = 10.68, p < 0.01). Post-hoc analysis showed a significantly higher probability of state transition from NREM to REM (0.17 ± 0.05% pre vs. 0.62 ± 0.10% post) and REM to NREM (0.17 ± 0.05% pre vs. 0.62 ± 0.09% post) in the LPS group post injection (Fig. S2). State transition probabilities remained at baseline levels in the CTRL group (NREM to REM: 0.22 ± 0.05% vs 0.27 ± 0.05%; REM to NREM: 0.27 ± 0.05% vs 0.27 ± 0.05%)
Although a higher number of state transitions was found after LPS injection, the characteristics of these transitions were not significantly different from those in controls or at baseline (Fig. S3). Thus, LPS injection leads to fragmentation of NREM and REM to a similar degree, without affecting time spent in either state, or the characteristics of transitions between the states.
LPS leads to increased spectral similarity between REM and NREM
State-space analysis of REM and NREM based on spectral power Ratio1 (R1, 1–9 Hz/1–2 Hz) and Ratio 2 (R2, 1–45 Hz/1–15 Hz) resulted in two distinct state clusters of epochs in the control group and LPS group at baseline and after injection (Fig. 3A-D).
After LPS injection (Fig. 3D), the location of the NREM and REM clusters within state-space shifted, resulting in a decreased inter-cluster distance. The magnitude and direction of the changes in cluster location after LPS injection and in controls is shown in Fig. 3E.
The centroid of REM cluster shifted significantly in LPS group (0.23 ± 0.05 a.u.) compared to control group (0.10 ± 0.03 a.u., t(10) = 2.33, p = 0.04, Fig. 3F). The direction of mean resultant vector is towards the origin for REM cluster (Fig. 3E). In case of NREM cluster, there was no significant difference between the cluster centroids of control (0.16 ± 0.04 a.u.) and LPS group (0.35 ± 0.06 a.u., t(10) = 1.45, p = 0.17). The mean resultant vector however is directed away from origin, towards the REM cluster (Fig. 3E).
As a result of the observed shifts in cluster medians, the total distance between the two clusters decreased (Fig. 3G). We calculated the total distance between REM and NREM clusters before and after injection in control and LPS groups and observed a significant time effect (F(1,10) = 26.47, p < 0.001) and group*time interaction (F(1,10) = 17.09, p < 0.01). There was no group effect (F(1, 10) = 0.3, p = 0.59). Post-hoc analysis showed a significantly smaller distance between REM and NREM clusters in the LPS group post-injection (0.87 ± 0.19 a.u.) compared to baseline (1.30 ± 0.15 a.u.). The decreased inter-state cluster distance indicates increased spectral similarity between the states.
Spectral similarity in low frequency range
The decrease in distance between REM and NREM clusters could be the result of changes in R1, R2, or in both. By decomposing the total distance inter-cluster distance into R1 and R2 components, we determined which spectral range most affected post-LPS brain activity. Distance between the REM and NREM clusters was significantly decreased along R1 (rmANOVA; time effect F(1, 10) = 39.26, p < 0.001, group effect F(1, 10) = 0.03, p = 0.84, group*time interaction, F(1, 10) = 27.55, p < 0.001). Post-hoc analysis showed that the distance between clusters was significantly smaller in the LPS group post injection (0.69 ± 0.20 a.u.) compared to baseline (1.13 ± 0.14 a.u., Fig. S4A). By contrast, there was no significant change in the distance between the clusters along R2 (rmANOVA; time effect, F(1, 10) = 2.85, p = 0.12, group*time interaction, F(1, 10) = 1.59, p = 0.23, group effect, F(1, 10) = 1.64, p = 0.22, Fig. S4B).
Hence, LPS-mediated spectral similarity between REM and NREM is mostly the result of changes in the 1 to 9 Hz frequency range.
NREM contributes more to spectral similarity in lower frequencies than REM
After observing a reduced distance between REM and NREM clusters along R1, we investigated if this shift was state-specific or if both states contributed. In REM, we observed no significant changes in median R1 values (rmANOVA; time effect: F(1, 10) = 3.16, p = 0.10, group effect: F(1, 10) = 0.0002, p = 0.98, time*group interaction: F(1, 10) = 4.31, p = 0.06, Fig. S4C).
In NREM however, we found a significant time effect (F(1, 10) = 15.02, p < 0.01) and time*group interaction (F(1, 10) = 7.27, p = 0.02). There was no significant group effect (F(1, 10) = 0.13, p = 0.72). Post-hoc analysis showed a significant increase in R1 medians in the LPS group post-injection (0.61 ± 0.09 a.u.) compared to baseline (0.34 ± 0.09 a.u., Fig. S4D).
LPS causes instability within REM and NREM states
The observed state fragmentation and altered REM-NREM dynamics could be caused by to inflammation-related state instability. Here, we used within-state velocity, or the distance between subsequent epochs in state-space, as a measure of stability.
Velocities along R1 were not significantly affected by LPS injection in either REM (rmANOVA; time effect: F(1,10) = 1.20, p = 0.29; group effect: F(1,10) = 0.80, p = 0.39; time*group interaction: F(1,10) = 0.01, p = 0.92, Fig. S5A), or NREM (time effect: F(1,10) = 0.92, p = 0.36; group effect: F(1,10) = 0.86, p = 0.37; time*group interaction: F(1,10) = 0.0005, p = 0.98, Fig. S5B).
However, velocities along R2 were significantly increased in both REM and NREM after LPS injection. In REM, we observed a significant time*group interaction (F(1,10) = 7.93, p = 0.01) but no significant effect of group (F(1,10) = 0.96, p = 0.34) or time (F(1,10) = 1.17, p = 0.30) (Fig. S5C). Post-hoc analysis showed significantly higher velocities in the LPS group after injection compared to baseline (0.087 ± 0.03 a.u. vs. 0.074 ± 0.03 a.u.). The effects on R2 velocities in NREM were similar: there was a significant time*group interaction (F(1,10) = 7.15, p = 0.02), but no significant effect of group (F(1,10) = 0.96, p = 0.34) or time (F(1,10) = 1.17, p = 0.30, Fig. S5D). Post-hoc analysis showed no significant differences.
Effects of LPS on periodic and aperiodic power spectrum components
To investigate possible sources of the observed changes in R1, power spectra were analysed for representative channels. NREM spectra showed a marked reduction power below 3 Hz in LPS-injected rats, but not in controls (Fig. 4A). REM spectra showed the opposite effect: increased power in the < 3 Hz range, as well as a smaller increase in the 7–9 Hz range (Fig. 4B). Particularly in REM, these changes were quite variable. Power spectra of rats in the control group remained largely stable and at baseline levels.
The observed changes in spectral power ratios and spectral power distribution could be the result of alterations in EEG oscillations and in the background (aperiodic) components of the power spectrum. To better understand the observed changes in R1 in REM and NREM after LPS injection, aperiodic and periodic components of the power spectrum for each state were modelled using FOOOF in the LPS group. Overall model fits were good for NREM spectra (R2 = 0.995 ± 0.002, fit error = 0.031 ± 0.006) and REM spectra (R2 = 0.987 ± 0.008, fit error = 0.035 ± 0.007). Model fitting was not negatively affected by LPS injection (post-LPS fits in NREM: R2 = 0.996 ± 0.002, fit error = 0.029 ± 0.008, and REM: R2 = 0.985 ± 0.011, fit error = 0.035 ± 0.006).
Similar to non-anaesthetized recordings [34], REM spectra had overall slightly shallower slopes than the NREM spectra (pre-LPS REM vs. NREM exponents: 1.42 ± 0.28 vs. 1.95 ± 0.18; Fig. S6A-B). This was also the case after LPS injection (post-LPS REM vs. NREM exponents: 1.67 ± 0.22 vs. 2.09 ± 0.19; Fig. S6A-B)
The aperiodic components of the NREM spectra were not significantly affected by LPS injection (Fig. S6). Spectrum slopes remained stable at 107.5 ± 3.3% of pre-LPS slope values (Wilcoxon signed rank test, W = 21, p = 0.09 after Bonferroni correction, Fig. S6C), as did model offsets (102.15 ± 0.49%, W = 21, p = 0.09 after Bonferroni correction). Knee frequencies were more variable than the other parameters (pre-LPS: 1.83 ± 0.38 Hz, post-LPS: 2.68 ± 0.61 Hz) and showed a slight increase after LPS injection, but this effect was not significant (144.39 ± 11.36%, W = 21, p = 0.09 after Bonferroni correction, Fig. S6C).
Effects of LPS on aperiodic components in REM were similar to those observed in NREM (Fig. S6B). Like in the NREM spectra, REM slopes showed no significant changes after LPS injection (127.39 ± 15.09% of pre-LPS, W = 16, p = 0.94 after Bonferroni correction, Fig. S6D). NREM knee frequencies (pre-LPS: 5.37 ± 2.68 Hz, post-LPS: 5.49 ± 2.00 Hz, W = 15, p = 1.31 after Bonferroni correction, Fig. S6D) and spectrum offsets were likewise not significantly affected (pre-LPS: 4.13 ± 3.44 a.u., post-LPS: 4.42 ± 0.35 a.u., W = 16, p = 0.94 after Bonferroni correction, Fig. S6D). They remained at 170.83 ± 77.38% and 107.63 ± 5.45% of baseline, respectively.
In pre-LPS recordings, NREM spectra showed one major periodic component: an oscillation in the delta frequency range with a mean center frequency of 1.56 ± 0.06 Hz and peak widths between 0.2 and 0.8 Hz (Fig. 4C-D). After LPS injection, center frequencies of this oscillation increased and peak widths became more variable (Fig. 4E-F). Overall, the main oscillatory component in NREM became faster and more variable after LPS injection. As such, the amount of spectral power in frequencies higher than 2 Hz increased, resulting in the observed shift in Ratio 1 in NREM.
REM spectra showed multiple oscillations: a delta-like oscillation with a peak frequency in the 1–2 Hz range, similar to the one found in NREM, and a second theta-like oscillation with a peak frequency around 5–6 Hz (Fig. 4G-H). These oscillations remained present after LPS injection, but more oscillations were found to have slower peak frequencies, and more oscillations had peak frequencies lower than 2 Hz. Additionally, REM peak widths became more variable, and overall peak heights tended to be slightly lower (Fig. 4I-J). This slowing of the oscillatory components in REM resulted in relatively more power in the 1–2 Hz range, while the power in frequencies over 2 Hz was decreased. This effect, opposite in direction to that found in the NREM spectrum, resulted in the decreased spectral distance between NREM and REM clusters observed in state-space.