Microglial TNFα controls synaptic GABAARs, sleep slow waves and memory consolidation

  1. Maria Joana Pinto
  2. Lucy Bizien
  3. Julie M.J. Fabre
  4. Nina Ðukanović
  5. Valentin Lepetz
  6. Fiona Henderson
  7. Marine Pujol
  8. Romain W. Sala
  9. Thibault Tarpin
  10. Daniela Popa
  11. Antoine Triller
  12. Clément Léna
  13. Véronique Fabre
  14. Alain Bessis

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

Microglia sense the changes in their environment. How microglia actively translate these changes into suitable cues to adapt brain physiology is unknown. We reveal an activity-dependent regulation of cortical inhibitory synapses by microglia, driven by purinergic signaling acting on P2RX7 and mediated by microglia-derived TNFα. We demonstrate that sleep induces microglia-dependent synaptic enrichment of GABA A Rs in a manner dependent on microglial TNFα and P2RX7. We further show that microglia-specific depletion of TNFα alters slow waves during NREM sleep and blunt memory consolidation in sleep-dependent learning tasks. Together, our results reveal that microglia orchestrate sleep-intrinsic plasticity of synaptic GABA A Rs, sculpt sleep slow waves and support memory consolidation.

Article activity feed

  5. Version published to 10.1101/2022.02.21.481254v3 on bioRxiv
  6. Review Commons

    Note: This rebuttal was posted by the corresponding author to Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Reply to the reviewers

    To simplify the reading of our answers, we have numbered the questions of the reviewers. Similarly, to make the identification of the changes easier, we have written the changes in the manuscript in blue, orange and green when they were asked by reviewer 1, 2 or 3 respectively.

    __Reviewer #1 __

    Major comments:

    1. One issue concerns the conclusion that microglial TNFa signaling shapes slow waves during NREM sleep (e.g., title; lines 148, 175-176; 180; 222-223; 288) on the basis of the data shown in Fig. 4b-d. Slow waves normally consist of two components, The amount of changes reported in the study might indeed seem small if the role of microglia was to gate slow-wave-sleep itself. In contrast, the effects we report are in the upper range of the modulations reported for NREM sleep slow-waves: depletion of microglial TNFα yielded a ~17 % decrease (relative to controls) in maximal slope (fig 4d); while sleep deprivation or physiological sleep-wake cycle have been associated with similar changes in the slope of slow-waves:
    • 6% increase after sleep deprivation and 17% increase after SOM-IN optogenetic stimulation (Funk et al. 2017, ref 17);
    • 15% above or below average in early and late sleep (Vyazovskiy et al. 2007, ref 79);
    • Daily fluctuation of the slow-wave slope is smaller that 15% (Hubbard et al. 2020, ref 50). Noteworthy, as noticed by this reviewer, slow saves consist of slow oscillations and delta waves. We also measured the slope and duration of slow oscillations and delta waves and found significant differences in both between control and micTNFα-KO mice. (see response to Reviewer 2 point 8).

    The change in delta peak reported here following microglial TNFα depletion reveals a pre-eminence of slower waves (“d1” waves 0.75-1.75Hz), over the faster waves (“d2” waves 2.5-3.5Hz) the latter being the main form of oscillations potentiated by sleep need (Hubbard et al. 2020; ref 50). Similarly, enhanced SW-slope and short SW-period is associated with high sleep need (Hubbard et al. 2020; ref 50) and we found that microglial TNFα depletion reduces SW slope and enhances SW duration. Together, these results suggest a role of microglial TNFα in modulating delta oscillations in response to sleep need, which gives further support to our recent publication on the role of microglia in the expression of sleep need and sleep homeostasis (Pinto et al 2023; ref 80). Accordingly, we have now replaced in the figure 4 the panel 4b with the value of the peak of delta (whose significance is less clear) with the ratio of power in the two delta bands which has a clearer interpretation in light of the work cited above. Figure 4b and the text were changed accordingly (line 161 of the results and line 277 in the discussion). Similarly, microglia may contribute to changes in SWA related to memory consolidation following intense recruitment of cortical circuits (Huber et al. 2004, ref 83). This is now more clearly stated in the text (line 292).

    We do not claim to have demonstrated that the change in slow oscillations fully explains the loss of memory consolidation, but instead we report the convergent findings that TNFa depletion in microglia produces alterations in sleep slow oscillations of the order of magnitude of sleep-need induced effects, and disrupts memory consolidation known to be sleep-dependent.

    VGAT was used to identify GABAergic synapses in conjunction with GABAA receptors. Of various GABAergic interneurons, somatostatin (SOM)-containing GABAergic interneurons are known to be crucial for generating slow waves during NREM sleep through their axon terminals that target and concentrate in L1 (e.g., Funk et al., 2017, ref. 17). However, not all GABAA receptors in L1 would be associated with the inputs from SOM-containing GABA interneurons. For example, there are parvalbumin-containing GABA interneurons and their activation has been reported to DECREASE slow waves (Funk et al. 2017). This is relevant and should be discussed in relation to the results.

    Answer: As pointed out by the reviewer, we do acknowledge that not all GABAergic synapses in L1 are associated to inputs of SOM+ interneurons. Other than the axonal projections of SOM+ interneurons, axons of the inhibitory neuronal types neurogliaform cells and canopy cells can be found in L1. This does not seem to be the case for PV+ interneurons whose somata is not located in L1 and lack projections to L1 (Schuman et al. 2021 - ref 19). We have now acknowledged and discussed this in the discussion:

    Line 285: “In this study we show that L1 inhibitory synapses are modulated by microglia in a sleep-dependent manner. Our data favor the possibility that the microglia-targeted GABAergic synapses arise from SOM-IN. However, inhibition on L1 also originates local axonal arbors of neurogliaform and canopy cells and the identity of the microglia-regulated presynaptic terminals remains to be established.”

    To follow up on the above, (1) it is unclear why NeuN was used to delineate cell bodies (Fig. 1e). In fact, SOM-containing GABA neurons (see above) have been shown to inhibit pyramidal neurons through presynaptic inhibition of excitatory inputs as well as postsynaptic inhibition of dendrites, but not cell bodies, of pyramidal neurons (see Funk et al. 2017 for references). Some discussion along this line would be useful and potentially important. (2) In addition, it would have been interesting to add an immunolabel for SOM to identify SOM-containing axon terminals associated with VGAT (Figs 1, 2), and this could be done for parvalbumin (see above) terminals as well; however, this analysis is optional and not required.

    Answer:

    • The rationale for analyzing GABAergic synapses in L5 was to assess whether the daily change in synaptic GABAAR observed in L1 had some degree of regional specificity or was rather a widespread event throughout the cortical layers. In this regard, NeuN was used to delineate cell bodies in order to assess changes in synaptic GABAARs at somatic synapses, known to be targeted by parvalbumin inhibitory neurons. In addition to the quantification provided in figure 1e, we also analyzed synaptic GABAARs at non-somatic VGAT clusters in L5 and observed no ZT6/ZT18 difference (data not shown).
    • We do agree with this reviewer that looking selectively at GABAAR located at somatostatin pre-terminals would be extremely interesting. Accordingly, we have now tried to quantify the synaptic GABAR facing SOM positive presynaptic boutons identified in mice expressing tdTomato in SOM interneurons (figure S3). However, GABAAR immunostaining requires using an antigen retrieval method (see below) that, in our hands, bleaches tdTomato fluorescence. Our attempt to retrieve the lost SOM-tdTomato signal by IHC with RFP antibodies was unsuccessful (see below) suggesting that the structure of tdTomato is altered, thereby preventing such quantification.

    Quantification of changes in synaptic GABAARs at synapses targeted by SOM interneurons (SOM-IN) is not technically possible. Top, Confocal images of adult brain cortical layer 1 show signal of GABAARγ2, gephyrin and VGAT obtained with different protocols for tissue processing (fresh-frozen, perfusion with paraformaldehyde and perfusion with paraformaldehyde followed by decloaking chamber-based heat-induced epitope retrieval). Visualization of bona-fide gephyrin+VGAT+ synapses and synaptic GABAAR clusters was only possible by epitope retrieval. Bottom, Confocal images of mice expressing tdTomato in SOM-IN (SOMCre/+R26tdTom/+) show expected signal revealing SOM-tdTomato+ presynaptic boutons after perfusion. When the protocol for epitope retrieval is used, SOM-tdTomato+ signal is lost, which could not be retrieved by IHC with 3 different antibodies (source and catalog number are indicated).

    Minor comments:

    It appears that n's are not consistently reported. Please check.

    Answer: We have now:

    -added n’s in the legend of figure 4.

    -corrected a typo in the legend of figure 5 (line 664: “(f-i) n= 11” instead of “(f, g) n= 11”).

    -in the legend of figure S2, we have added “g. Mean intensity…b-g, n= 53-60 FOVs from 5 mice per group”.

    The Y-axis does not start from zero in some graphs. Although this might be a matter of preference, it can be misleading.

    Answer: The Y-axis that did not start from zero have now been explicitly signaled (figure 4b, d; figure 5b, d, e)

    In the supplementary information PDF, under Immunohistochemistry (IHC): "In direct IHC" in the first line of the paragraph should be "Indirect IHC".

    Answer: this is corrected. The sentence now starts by “Cryostat sections…”. Also, “Table 2 – Antibodies and IHC method used” has been corrected accordingly.

    __Reviewer #2 __

    Major Comments:

    1. (1) There are several instances where the authors state the experiments occurred "across the 24 h light/dark cycle" (Lines 42, 139), "during the sleep/wake cycle" (Lines 87, 242, 248), or "during sleep" (Lines 155, 220, 254). These statements are imprecise and can lead to erroneous interpretations of the data. For molecular studies, data were collected at a light period timepoint (Zeitgeber Time (ZT) 6) and a dark period timepoint (ZT18). While I appreciate the comparisons of the light and dark phases, 2 timepoints are not sufficient to claim that phenomena were tested across the light-dark cycle. More importantly, though, it is not accurate to claim outcomes from data collected during ZT6 occurred "during sleep" (or ZT18 outcomes occurred during wake). Although mice sleep more in the light period vs. the dark period, they are polyphasic sleepers and thus can be awake at ZT6 and asleep at ZT18. Therefore, statements should be edited for accuracy to instead state that phenomena were observed at ZT6/ZT18 or light/dark periods. (2) In addition, any figures (e.g., Figure S1) using x-axis labels of "W" and "S" should be relabeled as "ZT18" and "ZT6," respectively.

    Answer : (1) We agree with this reviewer and we have consistently corrected these imprecise wordings.

    (2) We have relabeled the figure S1 and replaced “W” and “S” by “ZT18” and “ZT6” respectively.

    1. The authors claim that microglial TNFα plays a role in sleep-dependent memory consolidation (Title and Lines 20, 22, 178, 198, 224, 276, 288) based on a series of experiments using tests previously shown to have a sleep-dependent consolidation component. However, the authors did not assess sleep-dependent consolidation in the micTNFα-KO and the tCtl mice, and thus this conclusion cannot be drawn. This is because the experimental paradigms did not include sleep deprivation. Claims that outcomes are sleep-dependent need to be shown as absent/impaired after sleep deprivation especially in mutant (and control) lines that have not been previously tested in this context. As such, claims of sleep-dependent memory consolidation (including in the title) should be removed OR new experiments including sleep deprivation should be included.

    Answer: Previous studies have shown that in the learning tasks that we have used, memory consolidation is lost in sleep-deprived animals (refs 53 to 55). Our current study indicates that memory consolidation is also lost in micTNFα-KO mice. Thus, we believe it would not be informative to perform sleep-deprivation in micTNFα-KO as requested by this reviewer since the outcomes (loss of memory consolidation) cannot be additive. However, we acknowledge that even though microglial TNFα depletion impacts slow waves that are known to play a causal role in consolidating the memory during sleep (refs 51 and 52), it cannot be excluded that TNFα depletion impairs memory consolidation by a non-sleep dependent pathway. Therefore, we have modified the title of the study to prevent misleading interpretations:

    New title: “Microglial TNFα controls synaptic GABAARs, sleep slow waves and memory consolidation”

    We have changed wording (lines 225 and 643). We have further added a cautionary note in the discussion:

    Line 299 : “We now show that mice lacking microglial TNFα display impaired memory consolidation when tested in a complex rotator motor learning task or in the floor-texture recognition (FTR) task. In these two tasks, memory consolidation is sleep-dependent. This suggests a possible involvement of microglial TNFα in the sleep processes that promote memory consolidation, while leaving open the possibility of a non-sleep-dependent mechanism.”.

    We are now more cautious in our conclusion and write (line 313) “This work demonstrates that microglia tune slow waves and support memory consolidation probably by acting during sleep”

    1. (1) "This shows that P2RX7 and microglial TNFα drive daily fluctuations in CaMKII Thr286-phosphorylation and are required for sleep-dependent GABAAR synaptic upregulation in L1 during the light phase" (Lines 144 - 146). Similar to the above comment, it cannot be definitively concluded the P2X7R or microglial TNFα are required for sleep-dependent GABAAR synaptic upregulation because sleep deprivation studies were not conducted in the P2rx7-KO or micTNFα-KO mice. (2) Furthermore, there is no analysis (or citation) of P2rx7-KO mice sleep-wake expression nor has the micTNFα-KO sleep data been presented at this point to make any determinations on how (possibly perturbed) sleep-wake expression in these mice could affect the stated outcomes.

    Answer:

    (1) As discussed in our previous answer, both sleep deprivation (fig 1d) and inactivation of TNFα or P2RX7 (fig 3) completely prevent the synaptic GABAAR accumulation and CaMKII phosphorylation at ZT6. Performing sleep deprivation on micTNFα-KO and P2RX7 KO is thus not expected to exert an effect since the outcomes (loss of synaptic accumulation and phosphorylation of CaMKII at ZT6) cannot be additive. We actually conducted sleep deprivation studies in micTNFα-KO as suggested by this reviewer (see below). As expected however, sleep deprivation (SD6) has no further effect on GABAAR accumulation and CaMKII phosphorylation on micTNFα-KOs when compared to ZT6.

    Impact of sleep deprivation on synaptic GABAAR and CaMKII phosphorylation in micTNFα-KO mouse brain. In complement to figure 3, SD prevents that increased accumulation of synaptic GABAARγ2 (left) and CaMKII phosphorylation (right) in tCTR, but has no effect in micTNFα-KO (green).

    left: Mean intensity of GABAARγ2 clusters at gephyrin+VGAT+ synapses normalized to ZT18. n= 48 to 65 FOVs from 4-5 mice per group.

    Right: Mean intensity of Thr286-phosphorylated CaMKII signal in L1 normalized to ZT18. n= 37 to 50 FOVs from 4-5 mice per group.

    (2) Please note that analysis of sleep structure of micTNF-KO mice is shown in figure 4, which reveals “that microglial TNFα has limited effects on sleep-wake patterns as shown by the lack of major alterations in the amounts of wake, NREM and REM sleep between micTNFα-KO and tCTL mice along a light/dark cycle (fig. 4a).” (line 158). Similarly, analysis of baseline sleep-wake structure in P2RX7-KO mice revealed no abnormalities (Krueger et al 2010, ref 45). This has now been discussed in the text (lines 143).

    1. There are some details regarding data analysis that are lacking:
    1. How were bouts defined for each arousal state? Answer: We have now defined the bouts in the Materials and Methods (section : “Sleep recording and analysis”) : “Bouts were defined as consecutive 10-s epochs of similar vigilance state and could be as short as one epoch.”
    1. (1) It seems more details are needed for EEG spectra analysis. From what values was the median derived and over what time period? How was each spectral bin normalized and over what time period? (2) What data (i.e., from what time period and duration) are shown in Figure 4b? Same question for Figure 4c-d? Were these time periods the same for controls and mutants given that NREM SWA changes across the light-dark cycle? Answer:

    (1) The spectrum of bouts of 2.56s (512 points at 200Hz) was computed by FFT and the spectrum corresponds to the median of the FFT of all bouts. This information is now in the material and methods section “Spectral analysis”.

    (2) The graphs reported in the text correspond to the 24h time period for both tCTL and micTNFα-KO mice. This has now been clearly indicated in figure 4 legend and in the material and methods section “Slow-wave analysis”.

    1. How was NREM delta power normalized and analyzed and over what time period? Answer: Normalization corresponds to the division for each animal of the spectrum by the total power of the spectrum. Data reported correspond to the 24h time period.
    1. Claims that GABAAR enrichment at synapses is sleep-dependent is based primarily on the data presented in Figure 1d reporting no increase in cortical GABAAR after sleep deprivation. A previous study (not cited) showed sleep deprivation increased GABAAR expression in CaMKIIα+ neurons in barrel cortex (Del Cid-Pellitero et al., Front Syst Neurosci, 2017). It would be helpful if the authors cited and discussed this study.

    Answer: Indeed, Del Cid-Pellitero et al. show that GABAARs located around the soma of layer 5 neurons are increased upon sleep deprivation. Importantly, they used brain tissue collected after perfusion with paraformaldehyde without antigen retrieval. This protocol was shown to result in uniform surface labeling of GABAARs without staining the pool of receptors clustered at synapses (Gasser et al. 2006 Nature Protocols, PMID: 17487173). In this study we performed antigen retrieval that allows visualization of synaptic GABAARs (see figure 1). We and them are thus labelling different pools of GABAARs: synaptic vs. membrane at the soma level (comprising both synaptic and extrasynaptic). On the other hand, as we did not observe a difference in synaptic GABAARs at the soma level in L5 between ZT6 and ZT18, we did not assess sleep-dependency by performing sleep deprivation in this cortical layer. Together, we do not interpret their results as conflicting to our findings, but rather that different sleep- and wake-dependent mechanism exist to regulate the abundance of GABAARs at the subcellular level. We have now cited this work and include a brief discussion:

    Line 58: “Sleep deprivation has previously been shown to increase GABAARs located around excitatory somas60. This suggests that the expression of GABAARs are differentially regulated depending on their subcellular localization”.

    1. Some sentences/conclusions are overstatements:
    1. "...discarding the possibility that lack of synaptic GABAARs enrichment upon PLX3397 treatment results from perturbed sleep during the light phase" (Lines 68 - 69). Only sleep time is reported to make this claim, but bout frequency, bout duration, and EEG spectra could be perturbed with this manipulation. This claim should be edited for accuracy or additional data (e.g., bout and spectral analysis) should be presented. In addition, Line 68 should be edited to state that "...microglia depletion does not alter sleep time during the light phase..." unless additional analyses are provided. Answer: We have now added the bout analysis in microglia depleted mice in extended table 2.

    "TNFα, which is mostly if not exclusively produced by microglia in the brain..." (Lines 93 -94). Although microglia are a major source of TNFα, there is evidence other brain cell types also release TNFα. In addition, the citation provided does not support this exclusivity claim.

    Answer: The reference 29 (Zeisel et al., reference 28 in the previous version) is a single cell transcriptomic study. The data are available online: http://mousebrain.org/adolescent/genesearch.html and show that TNFα mRNA is only detected in microglia. We are not aware of evidence showing that other brain cell types release TNFα. To our knowledge there is no brain RNA seq repository that shows TNFα expression in other brain cell-types e.g :

    1. "We thus anticipate that microglial TNFα may control REM by acting at the basal forebrain..." (Line 163). This statement is based on a cited study that reported REMS suppression (and increased NREMS time) after TNFα injection in the subarachnoid space of the basal forebrain. It is unclear to me why this statement is included when ICV and IV TNFα administration also reduce REMS (Shoham et al, Am J Physiol, 1987). Given these data and this statement is not being tested, it does not seem like it needs to be included. It should also be noted a previously reported (but not cited) global TNFα KO mouse (Szentirmai and Kapás, Brain Behav Immun, 2019) also showed increased REMS and REMS bouts, but this seemed to be a dark period phenotype (NREMS and Wake time, bout frequency, and bout duration were unaffected). This is an interesting detail to at least include in the second paragraph of the Discussion. Answer: To comply with this comment, we have now removed the sentence “We thus anticipate…” (line 163, now 160), and we have modified the second paragraph of the discussion so as to include the dark-period specificity described in Szentirmai and Kapás that we now cite.
    1. It is unclear to me why the authors believe ATP in these studies has a neuronal origin (Lines 106, 132, 218) when other cell types also release ATP. Is this because of NMDA treatment? If so, NMDA receptors are also expressed on other cell types like astrocytes (Verkhratsky and Chvátal, Neurochemical Research, 2020). Answer: We do not believe and we did not write that ATP has a neuronal origin. Indeed, we wrote:

    -line 104: “We next identified the signaling pathway between neuron and microglia”.

    -line 130: “ATP released downstream neuronal activity activates microglial P2RX7”

    -line 218: “…microglia sense neuronal activity through an ATP/P2RX7 signaling pathway”.

    However, to rule out any misinterpretation, we have now added a sentence that explicitly recall the possible involvement of other cell types:

    Line 132 “Noteworthy, our results do not exclude the possible involvement of other cell types acting between neurons and microglia”

    1. Because the authors rationalize investigating memory consolidation based on micTNFα-KO changes in NREM SWA, I am curious if the authors considered parsing NREM SWA into slow oscillations and delta waves as Vaidyanathan et al (eLife, 2021) did. The reason for this is because slow oscillations are shown to be associated with memory consolidation, but delta waves are associated with weakening memories.

    Answer: The parsing between delta waves and slow oscillations (SO) in the Vaidyanathan et al. article is based on a quantile separation of the size of the events (the top 15% of events are called SO while the rest may qualify as delta waves events; this is quite different from our definition which is based on deviations larger than 3 times the estimated standard deviation of slow fluctuations during Wake); it is worth noting that applying the definition from Vaidyanathan et al. to compare groups may introduce a bias in the interpretation if the rate of slow waves is modified between groups of animals. We have performed the parsing of slow events according to Vaidyanathan et al. and found similar changes for Slow Oscillations as using our definition (wider and “slower” SO). The same effect was observed in delta waves, suggesting that microglial TNFα affects both types of slow waves similarly. __ __

    1. For the complex wheel task, micTNFα-KO mice seem to start and end with better performance compared to tCTL on S1 (although it is not clear if this difference was statistically evaluated). Would the conclusions from this experiment change if data were normalized to account for the apparent better starting performance? Answer: Despite a trend for a better performance of micTNFα-KO at S1, no significant difference in the mean performance was found between controls and micTNFα-KO mice at S1. We show below the mean performance at S1 and S2 for controls and micTNFα-KO mice. Moreover, learning within each session (both at S1 and S2) is not altered in micTNFα-KO (figure 5c) revealing that the ability to learn is not affected in microglia TNF-KO mice neither in S1 nor in S2 suggesting that memory impairment across sessions is not the result of saturation of learning capacity.

    Mean performance in the two sessions (S1 and S2) of the complex wheel learning task in control and micTNFα-KO as measured by the mean time on the complex wheel (in seconds) of all trials in each session. ***p Nevertheless, we acknowledge that the apparent better starting performance could lead to misinterpretation of the results, and so as suggested by this reviewer, we normalized the data to the average of the last 3 trials in session 1 and computed using the normalized values the performance improvement (figure 5d) and consolidation (figure 5e). The same results were obtained. We thus interpret that the differences in memory consolidation between S1 and S2 (fig 5d, e) do not stem from changes in baseline performance.

    Results on the complex wheel task following normalization to performance in S1. For each mouse, latency to fall off the complex wheel in each S1 and S2 trials was normalized to the average of the last 3 trials in S1. The normalized values were used to plot the graphs as in figure 5:

    Left, latency to fall in S1 and S2;

    middle, performance improvement;

    right, consolidation of motor learning.

    We have now modified figure 5 accordingly.

    1. Many of the molecular studies emphasized a layer-specific effect in L1 vs. L5. It would be helpful if the authors could link (at least in the Discussion) this cortical-layer specificity with reported microglial TNFα effects on sleep parameters and memory consolidation.

    Answer: According to this comment, we have now proposed a mechanism for the L1 vs L5 specificity in the discussion:

    Line 254: “The layer 1 vs layer 5 specificity may arise from the molecular difference of GABAergic synapses across the somato-dendritic arbour as proposed. Alternatively, but not exclusively, it may result from a differential expression of TNF-R1 along the cortical layers and/or from layer-specific behavior of microglia”.

    We hope that it will help understand our hypothesis of a link between microglial TNFα effect on upper layer synapses with the effect on sleep parameters and memory consolidation that are proposed from line 277 onwards.

    Minor Comments:

    1. For the experiments investigating TNF receptor (TNFR) involvement (fig. S5), it would have been interesting to see the response to human recombinant TNFα which interacts with TNFR1 but not TNFR2 whereas mouse recombinant TNFα interacts with both receptors (Lewis et al, PNAS, 1991).

    Answer: The differential effect of human and mouse TNFα was not known by us. We do agree with this reviewer that the proposed experiment would have been interesting and would further confirm the data shown in Supp fig 5b showing an involvement of TNFR1 but not of TNFR2.

    1. "Synapse plasticity in the sleeping brain likely supports crucial functions of sleep" (Line 33). I believe it is more accurate to instead state sleep supports synapse plasticity. The sentences immediately following also provide examples of sleep/wake mediating plasticity.

    Answer: We have now replaced the sentence in line 29 for “Sleep drives plasticity of excitatory synapses”.

    1. "Moreover, microglia depletion also abolished the reduction of synaptic AMPA receptor subunit GluA2 at ZT6 (fig. S1)..." (Lines 70 -71). The data in this figure shows an increase GluA2, but the data cited showed decreased excitatory transmission. This discrepancy is not discussed.

    Answer: We agree with this reviewer that the we did not discuss the increased GluA2 at synapses upon microglial depletion. However, we believe that this aspect is beyond the scope of this study and although showing the data seemed important, we believed that discussing these data might lose the reader. We have now rephrased this sentence so as not to mislead the reader (line 68):

    “Moreover, microglia depletion also reversed the reduction of synaptic AMPA receptor subunit GluA2 at ZT6”

    1. If you normalize within treatment (e.g., PLX, SAP, minoc, 4-OHT, apy, PPADS, A74, PSB) rather than to controls (e.g., normalize PLX-iLTP to PLX-bsl instead of CTL-bsl) in Figures 2b-f, 3b-c, do you get the same results? Similarly, if you normalize PLX treatment to CTL ZT18 in Figure 1c-d, do you get the same general outcome?

    Answer: As requested by this reviewer, we have normalized to treatment rather than to control for all the experiments shown in figure 2b-f. The graph below with this normalization shows that the same results are obtained.

    Concerning figures 1c-d and 3b-c, the normalization suggested cannot be performed because CTL and PLX treated mice or mutant mice were not processed for immunohistochemistry simultaneously, and so intensity values are not comparable.

    1. Is there a significance marker missing in Figure 2g for bls vs. BzATP?

    Answer: The significance marker it is not missing. Even though the individual experiments performed consistently show an increase in CaMKII T286 phosphorylation with BzATP treatment, the difference does not reach statistical significance between bsl and BzATP using a nested one-way ANOVA followed by Sidak’s multiple comparison test (p=0.09). There was however a typo in the legend of figure 2g about the statistical test used, which has been corrected (lines 611).

    1. It is unclear if the fig. 5a callout is the correct one at Line 143. If it is correct, then it is confusing (to me) in the way it is currently associated with the text.

    Answer: at line 143 (now 142), the callout is fig S5a (supplementary figure 5a that confirms TNFα depletion) and not 5a.

    1. There is a missing reference in Line 279 after "NOR."

    Answer: we have now added the reference.

    __Reviewer #3 __

    Summary: This paper tries top address how microglia-related TNFa modulate the REM sleep and motor behavior consolidation, using layer I gabaergic synapse enhancement (possible by SOM interneurons).

    The methods and results are solid/convincing and easily to be followed and they seem logically for the conclusion which they state.

    However there are major issues which need to be addressed before formal submission.

    Reviewer #3 (Significance (Required)):

    These works have serious limitations to be addressed before submitting to formal journals.

    (1) The in vivo sleep experiments with micTNFa-KO (fig. 4 and extended table S1, Fig S8 a/b) indicate that layer I GABAergic synapse potentiation modulates the start of down-state of slow-waves, which supposes to affect the NREM sleep. Controversially, NREM sleep is not affected (with SW down state duration increased in KO mice) and only REM sleep is influenced. This is opposite to the literature that GABAergic synapses in the cortex or thalamus determine the power of slow-waves and The fast transition to down state suggests that cortical neurons should be less active for REM sleep compared with NREM sleep.

    Answer This reviewer pointed out that in extended table 1, micTNFα-KO mice spent more time in REM sleep over 24 hours. However, we believe that this increase is marginally significant because there was no parallel decrease in the time spent in Wake or NREM sleep. Accordingly, the mean duration of REM sleep is not affected (extended table 1) and when measuring the amount of REM sleep in 2-hour bins, figure 4a shows no difference between micTNFα-KO and tCTL mice. Additionally, figure 4 shows that there is no change in the electrophysiological features of REM sleep. On the other hand, the electrophysiological parameters of NREM sleep, such as the duration and slope of slow waves, are affected, as pointed out by this reviewer. Therefore, we do not view our findings as controversial. While we have not tested this hypothesis, we do not exclude that the lack of synaptic plasticity observed in micTNFα-KO mice is linked to the prolonged REM sleep.

    (2) REM sleep can consolidate motor learning. However, the data (Fig. 5) do not consistently support this, although the authors cite the literature to support their results with complex motor learning.

    Answer: In line with the comment of this reviewer, Li et al. 2017 (ref 13) study showed that REM sleep deprivation impairs the performance improvement in a motor learning tasks. However, and as noticed by this reviewer, the amount of REM sleep is increased in micTNFα-KO and we are not aware of any study that has shown the consequences of increased REM sleep on motor learning consolidation. The floor-texture recognition task that we have used to show an alteration of consolidation in micTNFα-KO was shown to be NREM sleep-dependent (Miyamoto et al 2016, ref 54).

    (3) NMDA-iLTP needs to be further addressed since NMDA with CNQX in oganotypic slices supposes not be able to activate NMDARs that need co-activation of AMPARs to depolarize to remove the Mg-blockade before NMDAR activation in neurons. To further strengthen this result, ACh should be co-applied with NMDA (if not AMPA) since only REM sleep is enhanced in this study indicating that ACh should be involved to activate neurons during rem sleep, more relevant to REM sleep enhancement.

    Answer: we followed the iLTP protocol in culture described by Petrini et al. (reference 26) and verified an increase in GABAAR accumulation at synapses. Furthermore, in preliminary experiments (not shown), we found that this effect depended on CaMKII, as previously reported by Chiu et al. who used NMDA only on acute slices (reference 27). These findings indicate that the application of NMDA+CNQX on organotypic slices replicates the synaptic effects observed in primary cultures (in which NMDA+CNQX is used) and acute slices (only NMDA). While investigating a novel protocol of synaptic plasticity using Ach and its possible connection to REM sleep is intriguing, we believe it would be more appropriate to explore this in a separate study.

  7. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #3

    Evidence, reproducibility and clarity

    Summary: This paper tries top address how microglia-related TNFa modulate the REM sleep and motor behavior consolidation, using layer I gabaergic synapse enhancement (possible by SOM interneurons).

    The methods and results are solid/convincing and easily to be followed and they seem logically for the conclusion which they state.

    However there are major issues which need to be addressed before formal submission.

    Significance

    These works have serious limitations to be addressed before submitting to formal journals.

    1. The in vivo sleep experiments with micTNFa-KO (fig. 4 and extended table S1, Fig S8 a/b) indicate that layer I GABAergic synapse potentiation modulates the start of down-state of slow-waves, which supposes to affect the NREM sleep. Controversially, NREM sleep is not affected (with SW down state duration increased in KO mice) and only REM sleep is influenced. This is opposite to the literature that GABAergic synapses in the cortex or thalamus determine the power of slow-waves and The fast transition to down state suggests that cortical neurons should be less active for REM sleep compared with NREM sleep.
    2. REM sleep can consolidate motor learning. However, the data (Fig. 5) do not consistently support this, although the authors cite the literature to support their results with complex motor learning.
    3. NMDA-iLTP needs to be further addressed since NMDA with CNQX in oganotypic slices supposes not be able to activate NMDARs that need co-activation of AMPARs to depolarize to remove the Mg-blockade before NMDAR activation in neurons. To further strengthen this result, ACh should be co-applied with NMDA (if not AMPA) since only REM sleep is enhanced in this study indicating that ACh should be involved to activate neurons during rem sleep, more relevant to REM sleep enhancement.
  8. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #2

    Evidence, reproducibility and clarity

    Summary:

    Pinto et al. sought to determine the role of microglia in sleep and plasticity. Using a combination of in vivo experiments in mice and organotypic slices, they report a molecular circuit whereby microglia respond to ATP via the purinergic receptor P2X7R to release tumor necrosis factor α (TNFα). TNFα then acts in its soluble form at TNF receptor 1 (TNRF1) to increase synaptic enrichment of GABAA receptors (GABAAR) in layer 1 (L1) of cortex (but not L5) during the light phase. These results suggest that microglial TNFα mediates cortical inhibitory synapses in a layer- and time-of-day-specific manner. The authors also showed selectively disrupting microglia TNFα alters slow wave activity (SWA) during non-rapid eye movement sleep (NREMS) and impairs memory consolidation.

    Major Comments:

    The authors performed several well-designed and controlled studies to uncover microglia regulation of GABAAR enrichment at synapses during the light period uncovering a nicely presented molecular circuit that included upstream and downstream mechanisms. They also nicely probed this circuit in vivo with a conditional, microglial-specific depletion of TNFα to determine the role of microglial TNFα in sleep and memory consolidation. Although the experiments were well done, the authors made some conclusions that cannot be determined by the experiments presented in this manuscript. Specifically, there are several claims that some of the phenomena occurred "during sleep" or are "sleep-dependent" when the experiments were not designed to test these claims. I provide more detailed comments below:

    1. There are several instances where the authors state the experiments occurred "across the 24 h light/dark cycle" (Lines 42, 139), "during the sleep/wake cycle" (Lines 87, 242, 248), or "during sleep" (Lines 155, 220, 254). These statements are imprecise and can lead to erroneous interpretations of the data. For molecular studies, data were collected at a light period timepoint (Zeitgeber Time (ZT) 6) and a dark period timepoint (ZT18). While I appreciate the comparisons of the light and dark phases, 2 timepoints are not sufficient to claim that phenomena were tested across the light-dark cycle. More importantly, though, it is not accurate to claim outcomes from data collected during ZT6 occurred "during sleep" (or ZT18 outcomes occurred during wake). Although mice sleep more in the light period vs. the dark period, they are polyphasic sleepers and thus can be awake at ZT6 and asleep at ZT18. Therefore, statements should be edited for accuracy to instead state that phenomena were observed at ZT6/ZT18 or light/dark periods. In addition, any figures (e.g., Figure S1) using x-axis labels of "W" and "S" should be relabeled as "ZT18" and "ZT6," respectively.
    2. The authors claim that microglial TNFα plays a role in sleep-dependent memory consolidation (Title and Lines 20, 22, 178, 198, 224, 276, 288) based on a series of experiments using tests previously shown to have a sleep-dependent consolidation component. However, the authors did not assess sleep-dependent consolidation in the micTNFα-KO and the tCtl mice, and thus this conclusion cannot be drawn. This is because the experimental paradigms did not include sleep deprivation. Claims that outcomes are sleep-dependent need to be shown as absent/impaired after sleep deprivation especially in mutant (and control) lines that have not been previously tested in this context. As such, claims of sleep-dependent memory consolidation (including in the title) should be removed OR new experiments including sleep deprivation should be included.
    3. "This shows that P2RX7 and microglial TNFα drive daily fluctuations in CaMKII Thr286-phosphorylation and are required for sleep-dependent GABAAR synaptic upregulation in L1 during the light phase" (Lines 144 - 146). Similar to the above comment, it cannot be definitively concluded the P2X7R or microglial TNFα are required for sleep-dependent GABAAR synaptic upregulation because sleep deprivation studies were not conducted in the P2rx7-KO or micTNFα-KO mice. Furthermore, there is no analysis (or citation) of P2rx7-KO mice sleep-wake expression nor has the micTNFα-KO sleep data been presented at this point to make any determinations on how (possibly perturbed) sleep-wake expression in these mice could affect the stated outcomes.
    4. There are some details regarding data analysis that are lacking:
      • a. How were bouts defined for each arousal state?
      • b. It seems more details are needed for EEG spectra analysis. From what values was the median derived and over what time period? How was each spectral bin normalized and over what time period? What data (i.e., from what time period and duration) are shown in Figure 4b? Same question for Figure 4c-d? Were these time periods the same for controls and mutants given that NREM SWA changes across the light-dark cycle?
      • c. How was NREM delta power normalized and analyzed and over what time period?
    5. Claims that GABAAR enrichment at synapses is sleep-dependent is based primarily on the data presented in Figure 1d reporting no increase in cortical GABAAR after sleep deprivation. A previous study (not cited) showed sleep deprivation increased GABAAR expression in CaMKIIα+ neurons in barrel cortex (Del Cid-Pellitero et al., Front Syst Neurosci, 2017). It would be helpful if the authors cited and discussed this study.
    6. Some sentences/conclusions are overstatements:
      • a. "...discarding the possibility that lack of synaptic GABAARs enrichment upon PLX3397 treatment results from perturbed sleep during the light phase" (Lines 68 - 69). Only sleep time is reported to make this claim, but bout frequency, bout duration, and EEG spectra could be perturbed with this manipulation. This claim should be edited for accuracy or additional data (e.g., bout and spectral analysis) should be presented. In addition, Line 68 should be edited to state that "...microglia depletion does not alter sleep time during the light phase..." unless additional analyses are provided.
      • b. "TNFα, which is mostly if not exclusively produced by microglia in the brain..." (Lines 93 -94). Although microglia are a major source of TNFα, there is evidence other brain cell types also release TNFα. In addition, the citation provided does not support this exclusivity claim.
      • c. "We thus anticipate that microglial TNFα may control REM by acting at the basal forebrain..." (Line 163). This statement is based on a cited study that reported REMS suppression (and increased NREMS time) after TNFα injection in the subarachnoid space of the basal forebrain. It is unclear to me why this statement is included when ICV and IV TNFα administration also reduce REMS (Shoham et al, Am J Physiol, 1987). Given these data and this statement is not being tested, it does not seem like it needs to be included. It should also be noted a previously reported (but not cited) global TNFα KO mouse (Szentirmai and Kapás, Brain Behav Immun, 2019) also showed increased REMS and REMS bouts, but this seemed to be a dark period phenotype (NREMS and Wake time, bout frequency, and bout duration were unaffected). This is an interesting detail to at least include in the second paragraph of the Discussion.
    7. It is unclear to me why the authors believe ATP in these studies has a neuronal origin (Lines 106, 132, 218) when other cell types also release ATP. Is this because of NMDA treatment? If so, NMDA receptors are also expressed on other cell types like astrocytes (Verkhratsky and Chvátal, Neurochemical Research, 2020).
    8. Because the authors rationalize investigating memory consolidation based on micTNFα-KO changes in NREM SWA, I am curious if the authors considered parsing NREM SWA into slow oscillations and delta waves as Vaidyanathan et al (eLife, 2021) did. The reason for this is because slow oscillations are shown to be associated with memory consolidation, but delta waves are associated with weakening memories.
    9. For the complex wheel task, micTNFα-KO mice seem to start and end with better performance compared to tCTL on S1 (although it is not clear if this difference was statistically evaluated). Would the conclusions from this experiment change if data were normalized to account for the apparent better starting performance?
    10. Many of the molecular studies emphasized a layer-specific effect in L1 vs. L5. It would be helpful if the authors could link (at least in the Discussion) this cortical-layer specificity with reported microglial TNFα effects on sleep parameters and memory consolidation.

    Minor Comments:

    1. For the experiments investigating TNF receptor (TNFR) involvement (fig. S5), it would have been interesting to see the response to human recombinant TNFα which interacts with TNFR1 but not TNFR2 whereas mouse recombinant TNFα interacts with both receptors (Lewis et al, PNAS, 1991).
    2. "Synapse plasticity in the sleeping brain likely supports crucial functions of sleep" (Line 33). I believe it is more accurate to instead state sleep supports synapse plasticity. The sentences immediately following also provide examples of sleep/wake mediating plasticity.
    3. "Moreover, microglia depletion also abolished the reduction of synaptic AMPA receptor subunit GluA2 at ZT6 (fig. S1)..." (Lines 70 -71). The data in this figure shows an increase GluA2, but the data cited showed decreased excitatory transmission. This discrepancy is not discussed.
    4. If you normalize within treatment (e.g., PLX, SAP, minoc, 4-OHT, apy, PPADS, A74, PSB) rather than to controls (e.g., normalize PLX-iLTP to PLX-bsl instead of CTL-bsl) in Figures 2b-f, 3b-c, do you get the same results? Similarly, if you normalize PLX treatment to CTL ZT18 in Figure 1c-d, do you get the same general outcome?
    5. Is there a significance marker missing in Figure 2g for bls vs. BzATP?
    6. It is unclear if the fig. 5a callout is the correct one at Line 143. If it is correct, then it is confusing (to me) in the way it is currently associated with the text.
    7. There is a missing reference in Line 279 after "NOR."

    Significance

    The role of non-neuronal cells in sleep and sleep-related processes like learning and memory is relatively unexplored and this is especially true of microglia. The authors present a nicely done series of rigorous experiments that reveal a microglial-centric molecular circuit of inhibitory synaptic modulation that differs between the light and dark periods and plays a role in NREM slow wave activity and memory consolidation. However, most of the claims of sleep dependency of the reported phenomena have not been directly tested in this manuscript and thus await additional experiments or re-framing of the stated conclusions. Re-framing these conclusions without claims of sleep dependency still provides very interesting and informative data about the mechanistic role of microglia in inhibitory synapse plasticity and memory consolidation that may be of interest to researchers interested in learning and memory, synaptic plasticity, and sleep.

  9. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #1

    Evidence, reproducibility and clarity

    Summary:

    Using both in vivo and ex vivo approaches in mice, the authors showed that microglia in layer 1 (L1) of the frontal cortex modulate the level of GABAA receptors at L1 GABAergic synapses depending on the light/dark phase and, more specifically, sleep/wake state (high during sleep). This was shown to be mediated through purinergic signaling via microglial P2RX7 receptors followed by microglial release of TNFa then CaMKIIa phosphorylation in neurons. Microglia-selective TNFa-KO mice showed normal sleep/wake cycles except for an increase in REM sleep amount, but cortical EEG slow waves during NREM sleep, a measure of sleep propensity, were slower in the delta range (1-4 Hz), but without any change in delta power. These animals also showed deficits in memory consolidation in two out of three memory tasks used.

    Major comments:

    1. The paper is clearly written and easy to follow, with virtually no typos or editorial errors. Both the introduction and discussion are informative and well-referenced.
    2. The experiments are generally carefully designed including appropriate controls and comparison groups (e.g., L1 vs. L5; GABAAR vs. AMPAR; PLX vs. saporin vs. minocycline). The results are presented appropriately and often in detail including supplementary figures and tables.
    3. One issue concerns the conclusion that microglial TNFa signaling shapes slow waves during NREM sleep (e.g., title; lines 148, 175-176; 180; 222-223; 288) on the basis of the data shown in Fig. 4b-d. Slow waves normally consist of two components, < 1Hz (slow oscillations) and 1-4 Hz (delta waves), and Fig. 4b shows a modest slowing in delta range (from ~2.2 Hz to ~1.7 Hz, from reading the graph for means). Importantly, there was no change in the spectral density in delta range. In the opinion of this reviewer, this is a modest effect and its significance and impact remain to be investigated. Arguably, the effects on memory consolidation could have been a result of microglial TNFa gene deletion elsewhere in the brain of these KO mice. Modification of the claim on slow waves should be considered.
    4. VGAT was used to identify GABAergic synapses in conjunction with GABAA receptors. Of various GABAergic interneurons, somatostatin (SOM)-containing GABAergic interneurons are known to be crucial for generating slow waves during NREM sleep through their axon terminals that target and concentrate in L1 (e.g., Funk et al., 2017, ref. 17). However, not all GABAA receptors in L1 would be associated with the inputs from SOM-containing GABA interneurons. For example, there are parvalbumin-containing GABA interneurons and their activation has been reported to DECREASE slow waves (Funk et al. 2017). This is relevant and should be discussed in relation to the results.
    5. To follow up on the above, it is unclear why NeuN was used to delineate cell bodies (Fig. 1e). In fact, SOM-containing GABA neurons (see above) have been shown to inhibit pyramidal neurons through presynaptic inhibition of excitatory inputs as well as postsynaptic inhibition of dendrites, but not cell bodies, of pyramidal neurons (see Funk et al. 2017 for references). Some discussion along this line would be useful and potentially important. In addition, it would have been interesting to add an immunolabel for SOM to identify SOM-containing axon terminals associated with VGAT (Figs 1, 2), and this could be done for parvalbumin (see above) terminals as well; however, this analysis is optional and not required.

    Minor comments:

    1. It appears that n's are not consistently reported. Please check.
    2. The Y-axis does not start from zero in some graphs. Although this might be a matter of preference, it can be misleading.
    3. In the supplementary information PDF, under Immunohistochemistry (IHC): "In direct IHC" in the first line of the paragraph should be "Indirect IHC".

    Significance

    There is increasing evidence for and interest in the role of microglia in modulating synapses and neuronal circuits underlying various behaviors including sleep. However, the role of cortical microglia in sleep or NREM slow waves (a measure of sleep propensity or sleep need) is unclear. GABAergic synapses in cortical L1 are known to play an important role in NREM slow waves. Thus, the evidence that microglial P2X7-TNFa signaling enriches synaptic GABAA receptors selectively in L1 (not in L5) and during sleep is important and novel.

    On the other hand, in this reviewer's opinion, the effect of microglia-selective TNFa gene deletion on slow waves during NREM sleep seems modest (minor slowing but no change in power) and it is also unclear whether the effects on memory consolidation in two out of three tasks were due to the observed change in slow waves, or other alterations that are likely in these KO mice.

    The reviewer's expertise: sleep neurobiology. Familiar with microglia and much of the techniques, but not ex vivo techniques.

  10. Version published to 10.1101/2022.02.21.481254v2 on bioRxiv
  11. Version published to 10.1101/2022.02.21.481254v1 on bioRxiv