Hippocampus consolidates memory in the upstate of cortical sleep slow oscillations

16 Cortical slow oscillations (SOs), a hallmark of non -rapid eye movement (NonREM) sleep, have 17 been proposed to support systems memory consolidation by organizing hippocampal -cortical 18 communication. However, whether consolidation requires hippocampal memory processing 19 during SO-defined windows is unclear. Here, we used closed -loop optogenetics to transiently 20 inhibit dorsal hippocampal activity in adult rats (N = 12) during NonREM sleep following object-21 place association learning, either during cortical SO upstates or outside SOs, compared with a 22 no-stimulation control. Inhibition during SO upstates completely abolished expression of memory 23 at retrieval, despite preserved sleep architecture and intact cortical SO and spindle dynamics. By 24 contrast, inhibition outside S Os preserved memory and only slightly reduced performance 25 compared to the no-stimulation control. Memory impairment from hippocampal inhibition was 26 largely mediated by SO upstates nesting spindles. Our findings provide novel evidence that sleep-27 dependent systems consolidation requires precisely timed hippocampal-neocortical dialogue. 28 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 2 Main 29 Sleep supports the consolidation of newly acquired memor ies 1,2. Consolidation during sleep is 30 commonly thought of as a systems consolidation pro cess based on the dialogue between the 31 hippocampus and neocortical networks during non -rapid eye movement (NonREM) sleep 3–6. 32 Within this framework, cortical slow oscillations (SOs) are assumed to favor the neuronal replay 33 of newly encoded memory representations in hippocampal networks, thereby facilitating the 34 transfer of hippocamp ally reactivated information as well as memory storage in neocortical 35 networks 7–9. SOs emerge in the neocortex as prominent, discrete events consisting of a 36 hyperpolarized downstate followed by a depolarized upstate, recurring at approximately 0.1 -4 37 Hz 10,11. At the behavioral level, SOs have been consistently linked to enhanced memory 38 performance 2,12,13. At the neural level, the depolarizing SO upstate defines temporal windows 39 during which activity across widespread neuronal populations becomes synchronized, likely 40 driven by the highly coordinated neuronal silence during the preceding downstate 14,15. 41 Accordingly, SOs are assumed to als o provide periods of enhanced coordination between 42 hippocampal and neocortical neural activity 16–18. Indeed, within hippocampal networks, neurons 43 encoding recent experiences are repeatedly reactivated in conjunction with sharp-wave ripples 44 which preferentially occur in close proximity to the SO downstate 19–21. Recent work has identified 45 a subset of high-amplitude, long-duration sharp wave-ripples that preferentially emerge during the 46 SO upstate and are associated with enhanced memory reactivation in both the hippocampus and 47 prefrontal cortex 22. The SO upstate also promotes the generation of thalamo -cortical sleep 48 spindles 11,23,24, which facilitate sharp wave-ripple occurrence during the upstate 6,25 and are 49 themselves critical for memory consolidation 26,27. 50 Whereas the role of a hippocampal-cortical dialogue in memory consolidation has been 51 inferred based on correlative findings, direct causal demonstrations remain limited. In a seminal 52 study, Maingret et al. 28 showed that electrical stimulation to the neocortex timed to hippocampal 53 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 3 sharp wave-ripples enhances hippocampal-cortical coupling as well as later memory retrieval in 54 rats. Here, we provide complementary novel evidence for a causal role of the hippocamp al-55 neocortical dialogue in memory consolidation by adopting the converse approach , i.e., by 56 optogenetically silencing hippocampal activity timed to online detected neocortical SO upstates. 57 Object-location memories were preserved at retrieval 3 hours after encoding when the 58 hippocampus was silenced during post -encoding NonREM sleep outside SOs or in a 59 no-stimulation control condition. In contrast, hippocampal silencing during SO upstates entirely 60 abolished the expression of object-location memory. Mediation analysis indicates that the memory 61 deficit is largely explained by disruption of spindles nest ing within SO upstates, implicating SO-62 spindle-coupled hippocampal processing as the principal mechanism organizing memory 63 consolidation. 64 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 4

65 Experiments were performed in male rats (N = 12) which were chronically implanted with EEG 66 screw electrodes (of which the 0.1-4 Hz filtered EEG over left frontal cortex was used for online 67 detection of SO events) and optic fibers positioned bilaterally above the dorsal CA1 (expressing 68 the red -shifted light -activated chloride pump Jaws ) for optogenetic inhibition of hippocampal 69 activity (Fig. 1a and Extended Data Fig. 1a). Animals were repeatedly subjected to a standard 70 object-place recognition (OPR) task consisting of a 10-min encoding phase, a 3 -hour retention 71 interval during which the animals slept in a resting box, and a 5 -min retrieval phase (Fig. 1b). 72 Closed-loop optogenetic inhibition of the hippocampus was delivered either during the SO upstate 73 (In-Phase condition), outside periods of online -detected SOs (O ut-of-Phase condition), or not 74 delivered despite SO detection (No-Stimulation control condition, Fig. 1c), using a within-subject-75 design (Extended Data Fig. 1b). The efficacy of hippocampal inhibition by Jaws activation was 76 tested in three additional animals and confirmed that light delivery acutely produced a robust and 77 consistent suppression of neuronal firing rates (Wilcoxon signed-rank test: V = 6486.5, p < 0.001), 78 with 84.5% of recorded neurons showing suppressed activity during illumination (Fig. 1d,e , 79 Extended Data Fig 1c,d). 80 81 Hippocampal inhibition during cortical SO upstates disrupts OPR memory consolidation 82 Memory performance during the OPR retrieval phase was quantified using the discrimination 83 ratio, reflecting the relative exploration time of the object in the novel versus familiar location. 84 Replicating previous reports 29–32 animals exhibited robust memory performance in the No -85 Stimulation control condition, as indicated by a high discrimination ratio during the first three and 86 full five minutes of the test phase (mean ± SEM: 0.45 ± 0.08 and 0.51 ± 0.07, respectively; t(11) = 87 5.33 and t(11) = 7.07 , p < 0.001 against chance level; Fig. 1 f). In contrast, optogenetic ally 88 inhibiting the hippocampus In-Phase, i.e., during the SO upstate s of post -encoding NonREM 89 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 5 sleep, effectively abolished memory performance (mean ± SEM: -0.09 ± 0.09 and -0.12 ± 0.09, 90 respectively for three and five minutes; t(11) = -1.04, p = 0.32 and t(11) = -1.25, p = 0.24 against 91 chance level; t(20.00) = 4.44, p < 0.001 and t(20.60) = 5.29, p < 0.001 for post-hoc comparison 92 against the No-Stimulation condition). Notably, memory performance in the Out-of-Phase 93 inhibition condition remained significantly above chance (mean ± SEM: 0.20 ± 0.08 and 0.26 ± 94 0.08, respectively for three and five minutes; t(8) = 2.44, p = 0.04 and t(8) = 3.42, p = 0.009 against 95 chance level), and also was significantly higher than in th e In-Phase condition (t(18.80) = 2.44, 96 p = 0.05 and t(18.90) = 3.12, p = 0.006, respectively for three and five minutes ), but modestly 97 reduced compared to the No-Stimulation condition ( t(18.60) = -2.06, p = 0.054 and 98 t(18.20) = -2.40, p = 0.03, respectively for three and five minutes). The overall effect of inhibition 99 condition was additionally confirmed by an analysis that controlled for the order of stimulation 100 conditions (χ²(2) = 45.06, p < 0.001; Extended Data Fig. 1e). Importantly, we did not find any 101 differences between conditions in control measures, including distance travelled and total object 102 exploration time during encoding and retrieval, as well as total sleep duration, total inhibition time, 103 mean inhibition time, and inhibition density during the retention interval (all p > 0.05, Fig. 1g and 104 Extended Fig. 1f), thereby excluding confounding effects of nonspecific arousal-related factors on 105 memory performance. These results indicate that the timing of optogenetic hippocampal inhibition 106 during NonREM sleep critically determines its impact on memory consolidation, with the SO 107 upstate representing the most sensitive window. 108 109 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 6 110 Fig. 1. Timing of hippocampal inhibition during cortical SOs determines its impact on memory 111 consolidation. a, Left, Chronic implantation of frontal, parietal and occipital EEG screw electrodes 112 (reference/ground), bilateral optic fibers above dorsal hippocampus, and neck EMG electrodes. Right, 113 Bilateral optic fiber placement above dorsal CA1 expressing the r ed-light-activated chloride pump Jaws -114 GFP. b, Object-place recognition task. During encoding, rats explored two identical objects for 10 min. 115 During the 3 h retention interval, cortical SOs were detected online during N onREM sleep to trigger LED 116 illumination. During the 5 min retrieval test, one object was displaced to a novel location. c, Closed-loop 117 hippocampal inhibition during retention: No-Stimulation control (NoSTIM, top), inhibition during the SO 118 upstate (IN, middle), or during N onREM sleep outside of SOs (OUT, bottom). Representative EEG traces 119 illustrate LED timing (red bars). d, Left, Suppression of hippocampal spiking in one rat under urethane 120 anesthesia (spikes of 16 single units summed across 593 trials). Right, Normalized spike-count differences 121 between inhibition and baseline windows (units sorted by suppression magnitude). e, Top, Firing rates of 122 116 single units (N = 3 rats) recorded under urethane anesthesia decrease during LED illumination. Bottom, 123 Proportion of suppressed units and unchanged units. f, Mean ± SEM cumulative discrimination ratios during 124 retrieval shows better spatial memory in the No-Stimulation control condition than in the Out-of-Phase-125 inhibition condition, whereas inhibition during the SO upstate abolishes memory performance (*p< 0.05, **p 126 < 0.01, ***p< 0.001 for condition comparisons; ##p< 0.01, ###p< 0.001 against chance). g, Control measures: 127 total object exploration time and distance traveled during encoding and retrieval (left, right), as well as total 128 sleep duration and total inhibition time during retention (middle), were comparable across conditions. 129 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 7 Macro s leep archite cture as well as cortical SO and spindle dynamics are preserved 130 despite hippocampal inhibition 131 Memory consolidation has been shown to correlate with total sleep time and the amount of 132 NonREM sleep 33,34, and the magnitude of EEG slow -wave activity during N onREM sleep 35, 133 suggesting that both sleep duration and depth contribute to consolidation. Accordingly, we 134 examined whether closed -loop hippocampal inhibition altered overall sleep architecture during 135 the post -encoding interval. Across the 3 -hour retention period, animals spent approximately 136 88.34 ± 21.17 min awake, 81.77 ± 18.92 min in NonREM sleep and 6.82 ± 3.96 min in REM sleep 137 (mean ± SD), with no differences between conditions (all p > 0.05, Fig. 2b). Likewise, the average 138 duration of individual Wake, N onREM, and REM epochs (mean ± SD: 94.78 ± 36.81, 139 86.48 ± 24.62, and 118.52 ± 34.66 s) as well as latencies to NonREM and REM sleep onset 140 (mean ± SD: 19.64 ± 18.92 and 91.70 ± 3.96 min, respectively) were comparable across 141 conditions (all p > 0.05, Extended Data Fig. 2a). Spectral power distributions during NonREM and 142 REM sleep (from left frontal EEG) were indistinguishable between conditions as well (all p > 0.05, 143 Fig. 2c). 144 Considering evidence that hippocampal activity, particularly the occurrence of sharp 145 wave-ripples, may influence the emergence of cortical SOs in a bottom -up manner 26,36,37, we 146 examined whether closed -loop optogenetic inhibition of the dorsal CA1 altered cortical SOs, 147 spindles, or their temporal coupling. As sharp-wave ripples and associated hippocampal memory 148 replay occurs time locked to SOs 20,38, stronger changes in SO and spindle dynamics might have 149 been expected for the In-Phase inhibition of CA1 compared to the Out -of-Phase condition . 150 However, the three experimental conditions did neither differ in SO density or negative-to-positive 151 peak amplitude, nor in spindle density or absolute peak amplitude (all p >  0.05, Fig. 2d and 152 Extended Data Fig. 2b). Also, SO-spindle coupling remained unaffected: The percentage of SOs 153 coupled to spindles, defined by a spindle peak occurring within 1 s following SO detection, was 154 comparable across conditions, as were the percentage of coupled spindles (all p > 0.05, Fig. 2e). 155 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 8 Likewise, phase-amplitude coupling, i.e., the exact SO phase at which spindle peaks occurred , 156 was close to the SO positive maximum and comparable in all three conditions (Watson-Williams 157 Test, p = 0.74, Fig. 2f). These results for the frontal EEG were replicated for parietal EEG channels 158 located in closer proximity to the hippocampus (Extended Data Fig . 2c-e). Together, these 159 findings demonstrate that closed -loop hippocampal inhibition - whether applied during SO 160 upstates or outside of SOs - did not substantially alter cortical SO or spindle dynamics , or any 161 relevant parameters of sleep macro-architecture. 162 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 9 163 Fig. 2. Preserved macro sleep architecture and cortical oscillatory dynamics despite hippocampal 164 inhibition. a, Representative hypnograms of one animal during the 3 h retention interval with raw frontal 165 EEG, EMG, and optogenetic inhibition times (red dots). Top, No -Stimulation control condition (NoSTIM); 166 Middle, In-Phase inhibition (IN); Bottom, Out -of-Phase inhibition (OUT). EEG scale: ±500 µV ; EMG scale: 167 ±1000 µV. b, Mean ± SEM total time spent in each brain state during retention were comparable across 168 conditions (condition comparison: Wake: χ²(2) = 4.99, p = 0.08, NonREM: χ²(2) = 3.47, p = 0.18, REM: χ²(2) 169 = 1.91, p = 0.39). c, Mean spectral power during NonREM (top) and REM (bottom) sleep epochs did not 170 differ between conditions. d, Mean density of online-detected cortical SOs (left) and sleep spindles (right) 171 during the retention interval were comparable across conditions. e, SO-spindle coupling, defined as spindle 172 onsets occurring within 1 s after online SO detection, was comparable across conditions, both for the 173 percentage of SOs followed by spindles (left) and the percentage of spindles coupled to SOs (right). f, Mean 174 SO phase at which spindle power peaked. Purple dots indicate individual animals; grey lines indicate the 175 mean phase of maximal spindle power. In all conditions, spindle power peaked near the SO positive 176 maximum (0°, Watson-Williams Test for condition comparison: p = 0.74). 177 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 10 Memory impairment following closed -loop hippocampal inhibition during SO upstates is 178 mediated by cortical spindles 179 SOs often nest a spindle in their upstate and there is some evidence that , rather than directly 180 impacting hippocampal memory processing, the influence of SO upstates may be conveyed via 181 thalamically generated spindles 6,11,36. For example, optogenetic induction of spindles temporally 182 grouped hippocampal ripples, independent of the SO phase in which they were induced 39. 183 Indeed, time-frequency plots aligned to the rising flank of online-detected SOs revealed a robust 184 increase in spindle -frequency power (10 -16 Hz) during the SO upstate in the No -Stimulation 185 control condition (Extended Data Fig. 3a) as well as in the In-Phase condition, in which the rising 186 SO flank was equivalent to the inhibition onset, whereas in the Out -of-Phase condition, this 187 increase in spindle power preceded the inhibition window (all clusters p < 0.05; Fig. 3a). To 188 examine whether the impairment of OPR memory following hippocampal inhibition during the SO 189 upstates reflected a suppression of spindle-mediated, rather than direct influences of cortical SOs 190 on hippocampal networks, we performed a multilevel mediation analysis . This analysis tested 191 whether the difference in retrieval performance between the In -Phase and Out -of-Phase 192 conditions can be explained by the percentage of spindles that overlapped with the inhibition 193 window. Indeed, the two conditions differed in the percentage of spindles affected by hippocampal 194 inhibition (β = 0.14 ± 0.03, t = 5.48, p < 0.001, Fig. 3b,c), and a higher fraction of inhibited spindles 195 predicted poorer retrieval performance ( β = −2.20 ± 0.85, t = −2.59, p = 0.02). The mediation 196 analysis revealed a significant indirect, spindle-mediated effect (average causal mediation effect, 197 ACME = −0.30, 95% CI −0.57 to −0.0 7, p = 0.008), which accounted for approximately 84% of 198 the total effect (p = 0.01). In contrast, the direct effect of inhibition condition after accounting for 199 spindle disruption was not significant ( β = −0.07 ± 0.15, t = −0.47, p = 0.64). Consistent effects 200 were also observed in parietal EEG (Extended Data Fig. 3b -d). Although these results support 201 the view that spindles in the SO upstate mediate the effect of In-Phase inhibition of hippocampal 202 CA1 on memory consolidation, they do not rule out that spindles themselves, i.e., in absence of 203 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 11 SOs are able to trigger consolidation of OPR memory. However, memory performance did neither 204 correlate with the number of solitary spindles, i.e., spindles occurring in the absence of a SO, in 205 any of the three experimental conditions (all R 0.39, uncorrected for multiple 206 comparisons), nor with the fraction of spindles occurring during the inhibition window in the Out-207 of-Phase condition (R = -0.15, p = 0.71). Taken together, these findings point to hippocampal -208 cortical interactions supporting OPR memory consolidation being primarily mediated by 209 hippocampal processing during SO-spindle events. 210 211 212 Fig. 3. Memory impairment induced by hippo campal inhibition during SO upstates is spindle -213 mediated. a, Time-frequency representations locked to the inhibition onset in the In-Phase (left) and Out-214 of-Phase (right) conditions. Dashed lines indicate significant positive (black) and negative (white) spectral 215 clusters relative to baseline. b, The fraction of all spindles overlapping with the hippocampal inhibition 216 window was significantly high er in the In -Phase than in the Out -of-Phase condition ( t(16.30) = 5.34, 217 p = 0.054, ***p< 0.001). c, Mediation model relating inhibition condition (In -Phase versus Out -of-Phase), 218 the percentage of inhibited spindles, and memory performance. Path coefficients represent unstandardized 219 fixed-effect estimates (estimate ± s.e.) from linear mixed-effects models with animals included as a random 220 intercept. Solid arrows indicate significant paths; the dashed arrow denotes the direct effect of condition 221 after accounting for the mediator (*p< 0.05, ***p< 0.001). 222 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 12

223 According to active systems consolidation theory, memory consolidation during sleep relies on a 224 coordinated dialogue between neocortical and hippocampal networks , with neocortical slow 225 oscillations (SOs) providing temporal windows for hippocampal reactivation and information 226 transfer to long-term cortical stores 2,4. While there is a large body of evidence indicating causal 227 contributions to memory consolidation of both neocortical SOs 12,40,41 as well as hippocampal 228 memory reactivations 42–44, evidence for the causal importance of a hippocampal-neocortical 229 dialogue mediating consolidation during sleep remains scarce 28. Here, we pr ovide such 230 evidence, showing that effective consolidation of spatial object -location memory requires 231 hippocampal activity during SO upstates . Inhibiting dorsal CA1 during SO upst ates (In-Phase 232 inhibition) in post-encoding NonREM sleep completely abolished behavioral expression of object-233 location memory at a later retrieval test. With inhibitions of CA1 outside of SOs (Out-of-Phase 234 inhibition) object-place memory was preserved, although slightly diminished in comparison with 235 the No-Stimulation control condition. The mediation analysis further suggests that the effect of 236 SOs on hippocampus-dependent consolidation of memories is primarily conveyed through 237 spindles nesting into the SO upstat e. The findings extend prior work by providing direct causal 238 evidence that temporally precise hippocampal activity during the SO upstate is required for sleep-239 dependent memory consolidation. 240 The impairing effect on spatial memory consolidation was specific to hippocampal 241 inhibition during SO upstates. Brief optogenetic inhibition of CA1 neuronal activity was triggered 242 in real time by closed-loop detection of cortical SOs during post-encoding sleep. The 1-s inhibition 243 window was chosen to encompass the entire SO upstate including the period of maximal spindle 244 power. Although inhibition during this window abolished memory consolidation, it alter ed neither 245 sleep macro-architecture nor the incidence or coupling of SOs and spindles , ruling out that the 246 observed memory impairment was due to nonspecific confounds of inhibition. 247 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 13 SOs originating from neocortical networks can travel to the hippocampus to directly impact 248 memory processing in these networks 45–48. However, neocortical SOs may also ind irectly affect 249 hippocampal memory processing through spindles nesting in the SO upstate 6,36,39. Our mediation 250 analysis indicates that the memory impairment produced by hippocampal inhibition during SO 251 upstates was primarily linked to disruption of coupled SO -spindle events . Importantly, this 252 mediation does not imply that spindles act independently of SOs, but rather that they constitute a 253 key intermediary mechanism through which SO-defined windows shape hippocampal processing. 254 In line with this, s pindle activi ty occurring outside SOs did not predict behavi oral memory 255 performance at retrieval, nor did spindle activity present during periods of hippocampal inhibition 256 in the Out-of-Phase condition. This pattern aligns with previous work in which thalamic spindles 257 were optogenetically induced to examine their role in memory consolidation 39. In those 258 experiments, spindles were the primary factor timing hippocampal ripples associated with memory 259 replay, regardless of whether they were induced during an online-detected SO upstate or outside 260 any SO. Notably, however, intact spatial memory at later retrieval required that spindle induction 261 coincided precisely with the SO upstate, whereas spindles induced outside SOs did not enhance 262 memory. Together, these findings support the view that the temporally organizing influence of 263 SOs on hippocampal memory reactivations is primarily conveyed through coupled thalamic 264 spindles. Effective hippocampo-to-neocortical transfer of reactivated memory information and its 265 long-term storage into neocortical circuits , however, require th at spindle -timed hippocampal 266 memory reactivations occur during the excitable upstates of neocortical SOs 49. 267 Inhibiting hippocampal activity during N onREM sleep outside of SOs slightly reduced 268 memory performance, but retrieval remained above chance . In this Out-of-Phase condition, 269 hippocampal inhibition was delayed (by 1-2 s) relative to SO detection and terminated upon the 270 occurrence of a subsequent SO or transition into wake or REM sleep. Hippocampal inhibition in 271 this condition closely matched with that in the experimental In-Phase condition differing only in 272 the timing of the inhibition relative to cortical SOs (Fig. 1g and Extended Data Fig. 1f). The slight 273 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 14 decrease in memory in this Out -of-Phase condition was unexpected , but might hint at 274 hippocampal processing outside of SOs that adds to memory stabilization. For example, spindles 275 might be implicated in memory consolidation independent of co-occurring SOs 16,27,50,51, although 276 our correlational analyses did not reveal any significant link between spindle activity during Out-277 of-Phase hippocampal inhibition an d memory performance . Alternatively, the optogenetic 278 inhibition may have affected intrahippocampal processes that support memory consolidation 279 independently of both SOs and spindles . Recent work, for example, indicates that memory 280 reactivation during post -learning N onREM sleep is also regulated by interneuron -mediated 281 mechanisms within the hippocampus that operate outside periods of sharp-wave ripples and help 282 balance replay-related activity 52. Finally, the memory decrease in the Out -of-Phase condition 283 could also r eflect a limitation of our closed -loop approach, which detected SOs only when 284 exceeding a certain amplitude criterion and, thus, might have missed smaller, more local SOs 285 that nevertheless could impact hippocampal memory processing 53,54. 286 In sum, our findings identify the SO upstate as a critical temporal window of hippocampal-287 cortical interaction underlying effective memory consolidation during sleep. At the same time, they 288 point to contributions of hippocampal activity outside of SOs. To what extent these contributions 289 likewise reflect hippocampal-cortical interactions warrants further study. 290 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 15

291 Animals 292 Nineteen adult male Long-Evans rats (Janvier, Le Genest-Saint-Isle, France), aged 9-12 weeks 293 at the start of the experiment, were used in this study. The rats were housed in groups of 2-4 per 294 cage, with ad libitum access to food and water throughout the experiment. They were maintained 295 on a 12 -hour light/dark cycle (lights on at 6:00 am). Prior to the experiment, the animals were 296 handled daily for 10 -15 minutes over five consecutive days. All experimental procedures were 297 conducted in accordance with European animal protection laws and policies and were approved 298 by the Baden-Württemberg state authorities. 299 300 Surgical procedures 301 Animals underwent two surgical procedures: (i) virus injection four weeks prior to behavioral 302 testing, and (ii) implantation of EEG/EMG electrodes and optic fibers one week before testing, or, 303 in three animals, an acute recording procedure (see Acute Recordings). 304 Before each surgery, rats received an intraperitoneal injection of anesthetic mixture (0.005 mg/kg 305 fentanyl, 2 mg/kg midazolam, and 0.15 mg/kg medetomidine). Surgeries were performed under 306 general isoflurane anesthesia (induction: 1 -2%; maintenance: 0.8-1.2% in 0.35 L/min O ₂). Rats 307 were placed in a stereotaxic frame, body temperature was maintained at 33 -36 °C with a 308 feedback-controlled heating pad, and eyes were protected with ophthalmic ointment, before the 309 skull was exposed. 310 During the first surgery, animals were bilaterally injected with 500 nL of viral vector (AAV5-311 hSyn-Jaws-KGC-GFP-ER2; Addgene Plasmid #65014; diluted 1:4 in sterile PBS; resulting titer: 312 ~1.75 x 1012 vg/mL) targeting the dorsal CA1 (AP: -3.8 mm, ML: ±2.4 mm, DV: -2.3 mm, relative 313 to Bregma), to enable AAV -mediated expression of Jaws, a red -shifted light-activated chloride 314 inward pump. Expression was driven by the human synapsin promoter, which supports robust 315 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 16 transgene expression across neuronal populations but do es not distinguish between excitatory 316 and inhibitory neurons. The virus was delivered at 0.1 µL/min via a sharpened glass pipette 317 (Wiretrol II, Drummond Scientific; tip diameter <25 µm), which remained in place for 10 min post-318 injection to prevent backflow , then was automatically withdrawn at 0.2 µm/s using a motorized 319 micromanipulator (MP -285, Sutter Instruments) . Craniotomies were sealed with silicone 320 elastomer (Kwik -Cast, World Precision Instruments) and cold -polymerizing dental resin 321 (Palapress, Kulzer), and the wound was sutured. 322 For the second surgery, the sealant and dental resin were removed after skull exposure. 323 Five stainless-steel screw electrodes (Plastics One) were implanted: two frontal (AP: +2.6 mm, 324 ML: ±1.5 mm), two parietal (AP: -1.5 mm, ML: ±2.5 mm), and one occipital (AP: -10.0 mm, ML: 0 325 mm), the latter serving as reference and ground. Two 400 -µm diameter optic fibers (NA 0.5; 326 CFML15L10, Thorlabs) mounted in guide cannulae (OGL, Thorlabs) were bilaterally implanted 327 above dorsal CA1 (AP: -3.8 mm, ML: ±2.4 mm, DV: -1.8 mm). Two stainless-steel wire electrodes 328 were inserted bilaterally into the neck muscles for EMG recording. All electrodes were connected 329 to a Mill-Max pedestal (Mill-Max Mfg. Corp.) and secured to the skull with dental resin. After each 330 surgery, rats received subcutaneous carprofen (5 mg/kg) and were allowed to recover for at least 331 7 days before the start of behavioral testing. 332 333 Acute recordings 334 To validate optogenetic inhibition, three animals underwent acute urethane -anesthetized 335 recordings 4-6 weeks after viral injection. A two-shank, 32-channel sharpened silicon probe (P2-336 ASSY-116, Cambridge Neurotech) was mounted on a custom adapter coupled to a stepper motor 337 actuator (ZST225B, Thorlabs) equipped with a 400-µm diameter optical fiber (NA 0.5; Thorlabs), 338 coupled with a 625 nm LED (M625F2, Thorlabs). This configuration allowed the optical fiber to be 339 positioned <200 µm from the probe shanks and to be moved independently using Kinesis software 340 (Thorlabs). 341 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 17 Anesthesia was induced by stepwise intraperitoneal injection of urethane (1.5 g 342 urethane/5 ml sterile saline; 0.005 ml per g body weight) until no reflexes were observed. Animals 343 then received subcutaneous carprofen (5 mg/kg) and were secured in a stereotaxic frame. The 344 skull was exposed, and sealant and dental resin were removed above the left hemisphere. A 345 screw electrode was implanted at AP = -10.0 mm, ML = 0 mm relative to bregma and connected 346 to the silicon probe to serve as reference and ground. The exposed cortical surface was covered 347 with saline, and the probe-fiber assembly was slowly lowered into the brain under visual guidance 348 to prevent bending of the probe shanks. Once the probe/fiber tips were positioned above CA1 349 (AP = -3.8 mm, ML = -2.4 mm, DV = -1.8 mm), the assembly was left in place for 1 h to allow the 350 tissue to settle. The probe was then advanced in 50-µm steps at 2 µm/s using the stepper motor 351 until auditory monitoring of the neural activity from the probe indicated proximity to the CA1 352 pyramidal layer, follo wed by 20 -µm steps until clear spiking activity was detected across all 353 channels. Recordings were then initiated while delivering 0.5-1 s red LED light pulses (10-20 mW 354 at the fiber tip) with randomized 3-5 s inter-trial intervals to avoid rhythmic entrainment of spiking 355 activity and 5-minute breaks after 150 light pulses to let the tissue reset. After a 1 h recording 356 session, the probe-fiber assembly was retracted, and the animal was transcardially perfused (see 357 Histology). 358 359 Apparatus and Objects 360 The object-place recognition (OPR) task was conducted in a square open-field arena (80×80×40 361 cm) made of gray PVC. The arena was dimly lit (20-30 lux) and supplemented with constant white 362 noise (60 dB). A camera (Logitech C920) was mounted above the arena for video recording. 363 Distal spatial cues included the camera, posters affixed to the walls, hanging objects (e.g., 364 spheres, egg boxes), and surrounding curtains. Adjacent to the arena, a resting box (35×35×45 365 cm) made of stainless steel and filled with bedding material was used to house the animals during 366 the post-encoding phase. The resting box was enclosed within a Faraday cage, and the animal 367 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 18 was monitored using a second camera. Three pairs of glass objects of different shapes and sizes 368 (height: 15-30 cm; base di ameter: 7-12 cm), each filled with colored sand, were used. Objects 369 were sufficiently heavy to prevent displacement by the rats. To minimize olfactory cues, both the 370 arena and the objects were thoroughly cleaned with 70% ethanol after each trial. 371 372 Object-place recognition task 373 Before the experiment, animals were habituated to the experimental context over three 374 consecutive days (Extended Data Fig. 1b). Each day, rats first acclimated to the testing room for 375 20 min in their home cage before being placed into an empty IVC cage containing an unfamiliar 376 object (distinct from those used in the experiment) for 5 min to familiarize them with the presence 377 of objects. Subsequently, animals were placed for 10 min into the empty open-field arena, facing 378 a different wall on each habituation day. Afterward s, rats were connected to the EEG and optic 379 fiber cables and transferred to the resting box, where they remained undisturbed for 4 h with free 380 access to water (but not food). 381 For the encoding phases, rats were brought back to the testing room and, after a 20 -min 382 acclimation period, placed in the open -field arena containing two identical objects positioned 383 equidistant from two corners. Following 10 min of exploration, the animals were connected to the 384 EEG and optic fiber cables and placed in the resting box for a 3 -h retention interval. After this 385 period, rats were disconnected and left undisturbed in the resting box for 5 min to allow grooming 386 before the retrieval phase. During retrieval, the animals we re reintroduced into the arena, which 387 contained the same two objects, but one object was relocated to a different corner relative to the 388 encoding session. Animals were subjected to the OPR task either twice (n = 3) or three times (n 389 = 9), each using a diff erent combination of objects and locations within the arena. The first two 390 OPR sessions were separated by 2 days, and the third by 3 -7 days, to minimize potential 391 interference between tests. 392 393 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 19 Electrophysiological recordings 394 During behavioral experiments, EEG and EMG signals were acquired via the Mill -Max pedestal 395 and EEG tether (HS -18, Neuralynx) connected to a Digital Lynx SX acquisition system 396 (Neuralynx). Signals were sampled at 1 kHz using Cheetah software (Neuralynx). For acute 397 recordings, a 32-channel silicon probe was connected to the same acquisition system through a 398 buffered 36-channel headstage (HS-36, Neuralynx), and raw data were sampled at 32 kHz. 399 400 Online detection of slow oscillations 401 Cortical SOs during NonREM sleep were detected online using the left frontal EEG electrode. To 402 determine individualized detection thresholds, EEG recordings from the second habituation 403 session were first sleep-scored (see Offline sleep scoring), and SOs were identified offline (see 404 Offline detection of slow osci llations and sleep spindles ). The mean negative amplitude across 405 offline-detected SOs was then used as an individual negative threshold for online detection during 406 the retention interval (mean ± SD: -120.12 ± 32.13 µV). During the retention interval, the raw EEG 407 signal was digitally bandpass-filtered between 0.1 and 4 Hz (1st-order Butterworth filter, forward 408 only). An SO event was detected when two criteria were met: (i) the filtered signal crossed the 409 individual threshold in the negative direction within 150 ms following a positive -to-negative zero 410 crossing (falling flank), and (ii) the signal subsequently crossed one -third of the individual 411 threshold in the positive direction within 200 ms (rising flank). 412 413 Closed-loop optogenetic inhibition during online-detected slow oscillations 414 To restrict optogenetic inhibition to NonREM sleep, frontal EEG and EMG signals were bandpass 415 filtered (0.1-40 Hz and 80 -300 Hz, respectively; 1st-order Butterworth) and visually monitored 416 together with the video recording to i dentify NonREM epochs using the same criteria as offline 417 scoring (see Offline sleep scoring). Closed-loop SO detection and LED stimulation were enabled 418 only during these periods. 419 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 20 In the In-Phase inhibition condition, a 1 s LED pulse was triggered immediat ely after an 420 SO was detected. No further SO detections were permitted during the 1 s stimulation window. In 421 the Out-of-Phase-inhibition condition, the LED pulse was triggered after a random delay of 1.5 -422 2.0 s following SO detection. If a new SO was detecte d before the scheduled pulse onset, the 423 pulse was postponed by 1 s. When the filtered EEG crossed the individual negative threshold 424 within 150 ms following a positive -to-negative zero crossing (falling flank) while the LED was 425 active, the pulse was immedia tely terminated, and the missed stimulation time was added to 426 subsequent LED pulses (up to a maximum pulse duration of 2 s). If multiple LED pulses were 427 postponed (e.g., during SO trains), rescheduled pulses were separated by random intervals of 428 200-700 ms to avoid continuous light delivery exceeding 2 s. This closed -loop design ensured 429 that both the number and total duration of LED stimulations were matched across conditions. In 430 the No-Stimulation control condition, SOs were detected online, but the LED re mained off. LED 431 power output was kept at 10 -20mW at the fiber tip (Optical Power Meter, PM20, Thorlabs) and 432 was delivered bilaterally. 433 434 Histology 435 After completion of the experiments, animals were deeply anesthetized and transcardially 436 perfused with 4% paraformaldehyde (PFA) in phosphate-buffered saline. Brains were removed, 437 post-fixed in 4% PFA for at least 24 hours, and then sectioned coronally on a vibratome at 50-80 438 μm thickness. Sections were mounted with an antifade mounting medium containing DAPI 439 (Vectashield). Optic fiber placement above the hippocampus and Jaws expression was verified 440 post-hoc by GFP fluorescence in dorsal CA1 using a light microscope (Leica DMi8). 441 442 Behavioral assessment 443 Memory performance was quantified by manually scoring object exploration during encoding and 444 retrieval from video recordings analyzed in ANY -maze (Stoelting Europe) . Exploration was 445 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 21 defined as the rat being within 1 cm of an object with its nose directed toward it and actively 446 sniffing; leaning on the object without sniffing or being farther than 1 cm away was not counted. 447 All videos were scored by the same blinded, experienced experimenter. Memory retrieval in the 448 object-place recognition task was assessed using a discrimination ratio (DR), calculated as: 449 𝐷𝑅 = 𝑇𝑛𝑜𝑣𝑒𝑙 − 𝑇𝑓𝑎𝑚𝑖𝑙𝑖𝑎𝑟 𝑇𝑛𝑜𝑣𝑒𝑙 + 𝑇𝑓𝑎𝑚𝑖𝑙𝑖𝑎𝑟 450 where 𝑇𝑛𝑜𝑣𝑒𝑙 and 𝑇𝑓𝑎𝑚𝑖𝑙𝑖𝑎𝑟 denote the total exploration time directed toward the object in the novel 451 and familiar location, respectively. A positive discrimination ratio indicates successful memory for 452 the spatial change, whereas a ratio near zero reflects no exploration preference. 453 454 Electrophysiological data analyses 455 Offline sleep scoring 456 Offline sleep stage classification was performed manually using 10 s epochs, based on one frontal 457 and one parietal EEG channel in combination with the EMG signal. Scoring followed standard 458 criteria 55. Wakefulness was characterized by predominant low -amplitude, high-frequency EEG 459 activity accompanied by increased EMG tone. NonREM was defined by high-amplitude delta EEG 460 activity (<4 Hz) and red uced EMG activity. REM sleep was identified by dominant theta EEG 461 activity (4-8 Hz), minimal EMG tone, and the presence of phasic muscle twitches. 462 463 Offline detection of slow oscillations and sleep spindles 464 SOs and spindles were detected offline as described previously 30. For offline SO detection, EEG 465 signals from N onREM sleep were bandpass -filtered between 0.1 -4 Hz (3rd -order Butterworth). 466 SOs were defined by two consecutive positive-to-negative zero crossings separated by 0.5-2.0 s. 467 From all detected events, the 33% with the largest negative peak amplit udes were selected. 468 Spindles were detected after filtering between 10 and 16 Hz ( 6th-order Butterworth). The 469 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 22 smoothed absolute Hilbert transform of the signal was used to identify events exceeding (i) 1.5 470 SD of the mean NonREM level for 0.5-2.5 s, (ii) 2 SD for 0.25-2.5 s, and (iii) at least once 2.5 SD 471 within the same event. 472 473 SO-spindle events 474 The co-occurrence of online-detected SOs and offline-detected sleep spindles was quantified by 475 calculating the rate of spindle onsets occurring within 1 s after an online SO detection, allowing 476 the distinction between solitary and coupled SO -spindle events . To characterize SO -spindle 477 phase-amplitude coupling, the EEG signal was bandpass filtered in the narrow SO range (0.95 -478 1.05 Hz; 3rd-order Butterworth filter), and the instantaneous phase was extracted from the Hilbert-479 transformed signal. Spindle peaks we re identified from the absolute value of the Hilbert -480 transformed EEG signal, bandpass-filtered in the spindle band, for each offline-detected spindle. 481 If the maximum amplitude of a given spindle occurred within a ±1 s window around an online SO 482 detection, the corresponding SO phase was extracted and stored for further analysis. 483 484 Spectral analyses 485 To estimate spectral power across sleep stages, the continuous EEG signal was segmented into 486 4 s epochs and power spectra were computed over 1-45 Hz with a frequency resolution of 0.05 Hz 487 using fast Fourier transform with a Hanning taper in FieldTrip 56. For time -frequency analyses, 488 ±5 s segments centered on online -detected SOs, or on the inhibition onset in the delayed -489 inhibition condition, were analyzed using a multitaper convolution approach with a Hanning taper. 490 Frequency-specific time windows of seven cycles per frequency were used to estimate power 491 from 5 to 40 Hz in steps of 0.5 Hz. Power values were normalized on a per -animal basis to the 492 average power in a baseline window from -1.5 to -0.5 s relative to SO detection, or from -1 to 0 s 493 relative to inhibition onset in the Out-of-Phase condition. 494 495 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 23 Spike sorting 496 Spike sorting was performed on acute hippocampal recordings obtained under urethane 497 anesthesia, using SpikeInterface 57. Raw extracellular recordings were bandpass filtered between 498 300 and 6000 Hz, globally re-referenced, and whitened prior to spike detection. Automated spike 499 sorting was carried out using MountainSort5 with default parameters. Resulting units were 500 subjected to automated quality-based curation, and units were retained only if they met all of the 501 following criteria: signal -to-noise ratio >5, inter -spike interval violation ratio 0.7. All retained units were subsequently visually inspected using the 503 SpikeInterface graphical user interface, and units exhibiting non -physiological waveforms, 504 unstable firing rates, or clear contamination by noise were excluded. 505 506 Validation of optogenetic inhibition 507 To quantify optogenetic inhibition under anes thesia, spikes from all recorded single units (N = 508 116; 3 animals) were counted across inhibition trials (N = 1469) within the inhibition window and 509 a matched baseline window (0.5-1 s). Suppression was calculated as the normalized difference 510 between inhibition and baseline spike counts ( -1 = strong suppression; 0 = no change; >0 = 511 increased firing). For statistical analysis, firing rates were averaged across trials per unit and 512 compared between windows using a Wilcoxon signed-rank test. 513 514 Statistical analyses 515 Statistical analyses were performed using custom scripts in R 58 or MATLAB ( Version 2023b). 516 Animals were excluded a priori for (1) insufficient Jaws -GFP expression, (2) incorrect optic fiber 517 placement above CA1, or (3) low encoding exploration (<1 s per object). Based on these criteria, 518 three animals were excluded for insufficient expression and one for enlarged ventricles. Statistical 519 comparisons across conditions were conducted using linear mixed models, fitted using the lme4 520 package 59, with rat as a random intercept and group factors as fixed effects. To compare memory 521 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 24 performance, for instance, inhibition condition (in-phase vs. out-of-phase vs. no-stimulation) and 522 condition order (whether an inhibition condition was tested first, second, or third in the within 523 design) were compared using the formula: 524 525 𝐷𝑅 ~ (𝐼𝑛ℎ𝑖𝑏𝑖𝑡𝑖𝑜𝑛 𝐶𝑜𝑛𝑑𝑖𝑡𝑖𝑜𝑛 ∗ 𝐶𝑜𝑛𝑑𝑖𝑡𝑖𝑜𝑛 𝑂𝑟𝑑𝑒𝑟) + (1| 𝐴𝑛𝑖𝑚𝑎𝑙) 526 527 where DR indicated the discrimination ratio over the 5 min test interval. The significance of factors 528 was assessed by stepwise removal of the respective main effect or interaction from the model 529 and comparison of nested models using likelihood -ratio tests. Behavioral control parameters - 530 including total distance traveled, sleep duration, and total object exploration time during encoding 531 and retrieval phases - were analyzed analogously. Post-hoc comparisons were performed using 532 two-sided Welch’s t -tests. Correlations were assessed using Spearman’s rank coefficients to 533 account for the small sample size. 534 SO-spindle phase -amplitude coupling was assessed using Rayleigh tests for non -535 uniformity of circular distributions, as implemented in the Circular Statistics Toolbox 60, and group 536 differences were evaluated using Watson -Williams tests as implemented in the circular 537 package 61. Time-frequency representations were compared across conditions and against 538 baseline using dependent-samples t-tests with cluster-based permutation correction (Monte Carlo 539 method, 5,000 permutations, two-sided) as implemented in FieldTrip 56. Mediation analysis was 540 performed by estimating indirect and direct effects using quasi-Bayesian Monte Carlo simulation 541 as implemented in the mediation framework 62. Linear mixed -effects models with animal as a 542 random intercept were fitted for the total, mediator, and outcome paths, and mediation effects 543 were estimated from 1,000 simulations. For a ll analyses a p < 0.05 was considered significant. 544

were visualized using ggplot2 63 and ggpubr 64 packages. 545 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 25 Data availability 546 All data and code required to reproduce the statistical analyses and plots in the paper are 547 available at the following repository: https://github.com/MaxHarkotte/Hippocampus-consolidates-548 memory-in-the-upstate-of-cortical-sleep-slow-oscillations. Any further materials will be made 549 available upon reasonable request. 550 551 Author contributions 552 M.H., M.I., J.B. and N.N. designed the study. M.H., J.B. and N.N. wrote the paper. M.H. collected 553 and analyzed the data. 554 555 Acknowledgments 556 We thank Francesco Gobbo for sharing protocols for the use of Jaws in rats. We are grateful to 557 Edward S. Boyden for making the viral constr uct available. We also thank Ilona Sauter, Daniel 558 Gramling and Klaus Vollmer for technical support. 559 560 Funding 561 This study was supported by grants from the Deutsche Forschungsgemeinschaft to J.B. ( FOR 562 5434) and the European Research Council to J.B. (ERC AdG 883098 SleepBalance). M.I. and 563 N.N. are supported by the Hertie Foundation (Hertie Network of Excellence in Clinical 564 Neuroscience). 565 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 26

566 1. Stickgold, R. Sleep-dependent memory consolidation. Nature 437, 1272–1278 (2005). 567 2. Lutz, N. D., Harkotte, M. & Born, J. Sleep’s contribution to memory formation. Physiological 568 Reviews https://doi.org/10.1152/physrev.00054.2024 (2025) 569 doi:10.1152/physrev.00054.2024. 570 3. Buzsáki, G. The Hippocampo-Neocortical Dialogue. Cereb Cortex 6, 81–92 (1996). 571 4. Diekelmann, S. & Born, J. The memory function of sleep. Nat Rev Neurosci 11, 114–126 572 (2010). 573 5. Klinzing, J. G., Niethard, N. & Born, J. Mechanisms of systems memory consolidation during 574 sleep. Nature Neuroscience 22, 1598–1610 (2019). 575 6. Staresina, B. P., Niediek, J., Borger, V., Surges, R. & Mormann, F. How coupled slow 576 oscillations, spindles and ripples coordinate neuronal processing and communication during 577 human sleep. Nat Neurosci 26, 1429–1437 (2023). 578 7. Rothschild, G., Eban, E. & Frank, L. M. A cortical–hippocampal–cortical loop of information 579 processing during memory consolidation. Nat Neurosci 20, 251–259 (2017). 580 8. Brodt, S., Inostroza, M., Niethard, N. & Born, J. Sleep-A brain-state serving systems 581 memory consolidation. Neuron 111, 1050–1075 (2023). 582 9. Puentes-Mestril, C., Roach, J., Niethard, N., Zochowski, M. & Aton, S. J. How rhythms of 583 the sleeping brain tune memory and synaptic plasticity. Sleep 42, zsz095 (2019). 584 10. Massimini, M., Huber, R., Ferrarelli, F., Hill, S. & Tononi, G. The Sleep Slow Oscillation as a 585 Traveling Wave. J. Neurosci. 24, 6862–6870 (2004). 586 11. Steriade, M. Grouping of brain rhythms in corticothalamic systems. Neuroscience 137, 587 1087–1106 (2006). 588 12. Marshall, L., Helgadóttir, H., Mölle, M. & Born, J. Boosting slow oscillations during sleep 589 potentiates memory. Nature 444, 610–613 (2006). 590 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 27 13. Miyamoto, D. Optical imaging and manipulation of sleeping-brain dynamics in memory 591 processing. Neuroscience Research 181, 9–16 (2022). 592 14. Steriade, M., Timofeev, I. & Grenier, F. Natural Waking and Sleep States: A View From 593 Inside Neocortical Neurons. Journal of Neurophysiology 85, 1969–1985 (2001). 594 15. Volgushev, M., Chauvette, S., Mukovski, M. & Timofeev, I. Precise Long-Range 595 Synchronization of Activity and Silence in Neocortical Neurons during Slow-Wave Sleep. J. 596 Neurosci. 26, 5665–5672 (2006). 597 16. Sirota, A., Csicsvari, J., Buhl, D. & Buzsáki, G. Communication between neocortex and 598 hippocampus during sleep in rodents. Proceedings of the National Academy of Sciences 599 100, 2065–2069 (2003). 600 17. Helfrich, R. F. et al. Bidirectional prefrontal-hippocampal dynamics organize information 601 transfer during sleep in humans. Nat Commun 10, 3572 (2019). 602 18. Niknazar, H., Malerba, P. & Mednick, S. C. Slow oscillations promote long-range effective 603 communication: The key for memory consolidation in a broken-down network. Proceedings 604 of the National Academy of Sciences 119, e2122515119 (2022). 605 19. Wilson, M. A. & McNaughton, B. L. Reactivation of Hippocampal Ensemble Memories 606 During Sleep. Science 265, 676–679 (1994). 607 20. Buzsáki, G. Hippocampal sharp wave-ripple: A cognitive biomarker for episodic memory and 608 planning. Hippocampus 25, 1073–1188 (2015). 609 21. Schreiner, T. & Staudigl, T. Electrophysiological signatures of memory reactivation in 610 humans. Philos Trans R Soc Lond B Biol Sci 375, 20190293 (2020). 611 22. Robinson, H. L. et al. Large sharp-wave ripples promote hippocampo-cortical memory 612 reactivation and consolidation during sleep. Neuron 114, 226-236.e6 (2026). 613 23. Staresina, B. P. et al. Hierarchical nesting of slow oscillations, spindles and ripples in the 614 human hippocampus during sleep. Nat Neurosci 18, 1679–1686 (2015). 615 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 28 24. Klinzing, J. G. et al. Spindle activity phase-locked to sleep slow oscillations. NeuroImage 616 134, 607–616 (2016). 617 25. Clemens, Z. et al. Temporal coupling of parahippocampal ripples, sleep spindles and slow 618 oscillations in humans. Brain 130, 2868–2878 (2007). 619 26. Fernandez, L. M. J. & Lüthi, A. Sleep Spindles: Mechanisms and Functions. Physiological 620 Reviews 100, 805–868 (2020). 621 27. Peyrache, A. & Seibt, J. A mechanism for learning with sleep spindles. Philos Trans R Soc 622 Lond B Biol Sci 375, 20190230 (2020). 623 28. Maingret, N., Girardeau, G., Todorova, R., Goutierre, M. & Zugaro, M. Hippocampo-cortical 624 coupling mediates memory consolidation during sleep. Nature Neuroscience 19, 959–964 625 (2016). 626 29. Binder, S. et al. Sleep enhances memory consolidation in the hippocampus-dependent 627 object-place recognition task in rats. Neurobiology of Learning and Memory 97, 213–219 628 (2012). 629 30. Sawangjit, A. et al. The hippocampus is crucial for forming non-hippocampal long-term 630 memory during sleep. Nature 564, 109–113 (2018). 631 31. Oyanedel, C. N. et al. Role of slow oscillatory activity and slow wave sleep in consolidation 632 of episodic-like memory in rats. Behav Brain Res 275, 126–130 (2014). 633 32. Bolsius, Y. G. et al. Recovering object-location memories after sleep deprivation-induced 634 amnesia. Current Biology 33, 298-308.e5 (2023). 635 33. Takashima, A. et al. Declarative memory consolidation in humans: A prospective functional 636 magnetic resonance imaging study. Proceedings of the National Academy of Sciences 103, 637 756–761 (2006). 638 34. Sawangjit, A., Oyanedel, C. N., Niethard, N., Born, J. & Inostroza, M. Deepened sleep 639 makes hippocampal spatial memory more persistent. Neurobiology of Learning and Memory 640 173, 107245 (2020). 641 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 29 35. Wilhelm, I. et al. Sleep selectively enhances memory expected to be of future relevance. J 642 Neurosci 31, 1563–1569 (2011). 643 36. Oyanedel, C. N., Durán, E., Niethard, N., Inostroza, M. & Born, J. Temporal associations 644 between sleep slow oscillations, spindles and ripples. Eur J Neurosci 52, 4762–4778 645 (2020). 646 37. Sanda, P. et al. Bidirectional Interaction of Hippocampal Ripples and Cortical Slow Waves 647 Leads to Coordinated Spiking Activity During NREM Sleep. Cereb Cortex 31, 324–340 648 (2021). 649 38. Peyrache, A., Khamassi, M., Benchenane, K., Wiener, S. I. & Battaglia, F. P. Replay of rule-650 learning related neural patterns in the prefrontal cortex during sleep. Nat Neurosci 12, 919–651 926 (2009). 652 39. Latchoumane, C.-F. V., Ngo, H.-V. V., Born, J. & Shin, H.-S. Thalamic Spindles Promote 653 Memory Formation during Sleep through Triple Phase-Locking of Cortical, Thalamic, and 654 Hippocampal Rhythms. Neuron 95, 424-435.e6 (2017). 655 40. Ngo, H.-V. V., Martinetz, T., Born, J. & Mölle, M. Auditory Closed-Loop Stimulation of the 656 Sleep Slow Oscillation Enhances Memory. Neuron 78, 545–553 (2013). 657 41. Kim, J., Gulati, T. & Ganguly, K. Competing Roles of Slow Oscillations and Delta Waves in 658 Memory Consolidation versus Forgetting. Cell 179, 514-526.e13 (2019). 659 42. Girardeau, G., Benchenane, K., Wiener, S. I., Buzsáki, G. & Zugaro, M. B. Selective 660 suppression of hippocampal ripples impairs spatial memory. Nat Neurosci 12, 1222–1223 661 (2009). 662 43. Ego-Stengel, V. & Wilson, M. A. Disruption of ripple-associated hippocampal activity during 663 rest impairs spatial learning in the rat. Hippocampus 20, 1–10 (2010). 664 44. Gridchyn, I., Schoenenberger, P., O’Neill, J. & Csicsvari, J. Assembly-Specific Disruption of 665 Hippocampal Replay Leads to Selective Memory Deficit. Neuron 106, 291-300.e6 (2020). 666 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 30 45. Hahn, T. T. G., Sakmann, B. & Mehta, M. R. Phase-locking of hippocampal interneurons’ 667 membrane potential to neocortical up-down states. Nat Neurosci 9, 1359–1361 (2006). 668 46. Hahn, T. T. G., Sakmann, B. & Mehta, M. R. Differential responses of hippocampal subfields 669 to cortical up–down states. Proceedings of the National Academy of Sciences 104, 5169–670 5174 (2007). 671 47. Isomura, Y. et al. Integration and Segregation of Activity in Entorhinal-Hippocampal 672 Subregions by Neocortical Slow Oscillations. Neuron 52, 871–882 (2006). 673 48. Swanson, R. A. et al. Topography of putative bi-directional interaction between hippocampal 674 sharp-wave ripples and neocortical slow oscillations. Neuron 113, 754-768.e9 (2025). 675 49. Niethard, N., Ngo, H.-V. V., Ehrlich, I. & Born, J. Cortical circuit activity underlying sleep 676 slow oscillations and spindles. PNAS 115, E9220–E9229 (2018). 677 50. Siapas, A. G. & Wilson, M. A. Coordinated interactions between hippocampal ripples and 678 cortical spindles during slow-wave sleep. Neuron 21, 1123–1128 (1998). 679 51. Mölle, M., Eschenko, O., Gais, S., Sara, S. J. & Born, J. The influence of learning on sleep 680 slow oscillations and associated spindles and ripples in humans and rats. Eur J Neurosci 681 29, 1071–1081 (2009). 682 52. Karaba, L. A. et al. A hippocampal circuit mechanism to balance memory reactivation during 683 sleep. Science 385, 738–743 (2024). 684 53. Vyazovskiy, V. V., Riedner, B. A., Cirelli, C. & Tononi, G. Sleep Homeostasis and Cortical 685 Synchronization: II. A Local Field Potential Study of Sleep Slow Waves in the Rat. Sleep 30, 686 1631–1642 (2007). 687 54. Nir, Y. et al. Regional Slow Waves and Spindles in Human Sleep. Neuron 70, 153–169 688 (2011). 689 55. Neckelmann, D., Olsen, Ø. E., Fagerland, S. & Ursin, R. The Reliability and Functional 690 Validity of Visual and Semiautomatic Sleep/Wake Scoring in the Møll-Wistar Rat. Sleep 17, 691 120–131 (1994). 692 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 31 56. Oostenveld, R., Fries, P., Maris, E. & Schoffelen, J.-M. FieldTrip: Open Source Software for 693 Advanced Analysis of MEG, EEG, and Invasive Electrophysiological Data. Comput Intell 694 Neurosci 2011, 156869 (2011). 695 57. Buccino, A. P. et al. SpikeInterface, a unified framework for spike sorting. eLife 9, e61834 696 (2020). 697 58. R Core Team. R: A Language and Environment for Statistical Computing. R Foundation for 698 Statistical Computing (2024). 699 59. Bates, D., Mächler, M., Bolker, B. & Walker, S. Fitting Linear Mixed-Effects Models Using 700 lme4. Journal of Statistical Software 67, 1–48 (2015). 701 60. Berens, P. CircStat: A MATLAB Toolbox for Circular Statistics. Journal of Statistical 702 Software 31, 1–21 (2009). 703 61. Agostinelli, C. & Lund, U. R package ‘circular’: Circular Statistics. (2025). 704 62. Tingley, D., Yamamoto, T., Hirose, K., Keele, L. & Imai, K. mediation: R Package for Causal 705 Mediation Analysis. Journal of Statistical Software 59, 1–38 (2014). 706 63. Wickham, H. ggplot2. WIREs Computational Statistics 3, 180–185 (2011). 707 64. Kassambara, A. ggpubr: ‘ggplot2’ Based Publication Ready Plots. (2025). 708 65. Stujenske, J. M., Spellman, T. & Gordon, J. A. Modeling the Spatiotemporal Dynamics of 709 Light and Heat Propagation for In Vivo Optogenetics. Cell Reports 12, 525–534 (2015). 710 711 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 32 Extended Data Figures 712 713 Extended Data Fig. 1. Validation of hippocampal inhibition and memory performance controlling for 714 testing order. a, Histology. Bilateral optic fibers positioned above dorsal CA1 in animals expressing the 715 red-light–activated chloride pump Jaws-GFP (n = 12). b, Experimental timeline. Animals were handled for 716 5 days prior to the first surgery, during which Jaws-GFP was injected bilaterally into the hippocampus. After 717 3 weeks of recovery, a second surgery was performed to implant EEG and EMG electrodes together with 718 bilateral optic fibers above the dorsal hippocampus. Following a 1 -week recovery period, animals were 719 repeatedly tested in the object -recognition task; the first two tests wer e separated by 2 days and the third 720 by 3-7 days to reduce interference. Animals were subsequently transcardially perfused. c, Light delivery. 721 Estimated light spread within the hippocampus from a 625-nm LED pulse delivered through a 400-µm optic 722 fiber. Light propagation was modeled using a previously published MATLAB toolbox 65. d, Acute recordings. 723 Top left: A two-shank, 32-channel sharpened silicon probe (P2 -ASSY-116, Cambridge Neurotech; grey) 724 mounted to a 400 -µm optic fiber (white) was used during acute recordings . Top right : Histological 725 verification of probe shank placement in CA1 expressing Ja ws-GFP following acute recording. Bottom, 726 representative bandpass -filtered trace (300 –3000 Hz, 4th -order Butterworth) showing three trials of 727 hippocampal inhibition (red). e, Memory performance corrected for testing order. To confirm whether the 728 effect of closed-loop optogenetic inhibition locked to online-detected SOs (No-Stimulation vs. In-Phase vs. 729 Out-of-Phase) did not depend on the order in which the conditions were tested, estimated marginal means 730 were derived from a linear mixed-effects model including testing order as a factor (in addition to inhibition 731 condition as well as minute across the retrieval phase as fixed factors and animal as random intercept) . 732 While a significant effect of testing order was found (χ²(2) = 8.84, p = 0.012), the main effect of inhibition 733 condition on memory performance remained significant ( χ²(2) = 97.45, p < 0.001). No effect was found for 734 the minute during the retrieval phase, from which the discrimination ratio (DR) was calculated (χ²(4) = 0.93, 735 p = 0.92), although minute was retained in the model for completeness . Bars indicate the estimated 736 marginal means across all five minutes of the retrieval phase , adjusting for testing order, and error bars 737 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 33 represent SEM. Symbols above bars denote model-based tests of estimated marginal means against zero 738 (##p< 0.01, ###p< 0.001); brackets indicate pairwise condition contrasts (***p< 0.001). f, Inhibition across 739 conditions. Inhibition density (left) and average duration of one inhibition (right) were comparable between 740 the In-Phase and Out-of-Phase conditions. 741 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 34 742 Extended Data Fig. 2. Control measures of macro sleep architecture and sleep-oscillatory activity 743 show no difference despite hippocampal closed -loop inhibition. a, Left, Average epoch duration of 744 Wake, NonREM, and REM sleep. Right, Sleep latency to NonREM, and REM sleep. All measures were 745 comparable across conditions (all p > 0.05). b, Negative-to-positive peak amplitude of online detected SOs 746 (left) and absolute peak amplitude of sleep spindles (right) detected on the left frontal EEG electrode were 747 comparable across conditions. c, Spindle density and peak amplitudes were comparable across conditions 748 also when detected in parietal EEG, i.e., closer to the hippocampal inhibition. d, SO-spindle coupling was 749 comparable across conditions in the parietal EEG. e, Mean SO phase at which sp indle peak powered in 750 parietal EEG. Purple dots indicate individual animals; grey lines indicate the mean phase of maximal spindle 751 power. In all conditions, spindle power peaked near the SO positive maximum (0 °, Watson-Williams Test 752 for condition comparison: p = 0.13). 753 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint 35 754 Extended Data Fig. 3. Spindle-band p ower increase during the SO upstate under undisturbed 755 conditions and spindle -mediated memory impairment in parietal EEG. a, Time-frequency 756 representation locked to the online-detection of a SO in the No-Stimulation control condition for frontal EEG. 757 Dashed lines indicate significant positive (black) and negative (white) spectral clusters relative to baseline. 758 b, Same as (a), but for parietal EEG. c, Same as (a), but time-frequency representations are locked to the 759 inhibition onset in the In -Phase (left) and Out -of-Phase (right) conditions for parietal EEG. d, The fraction 760 of all parietal spindles overlapping with the hippocampal inhibition window was significantly higher in the 761 In-Phase than in the Out -of-Phase condition (t(16.20) = 5.58, ***p < 0.001). e, Mediation model relating 762 inhibition condition (SO in -phase versus Out-of-Phase), the percentage of inhibited spindles in parietal 763 EEG, and memory performance. Path coefficients represent unstandardized fixed -effect estimates 764 (estimate ± s.e.) from linear mixed-effects models with animals included as a random intercept. Solid arrows 765 indicate significant paths; the dashed arrow denotes the direct effect of condition after accounting for the 766 mediator (*p< 0.05, ***p< 0.001). 767 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted April 17, 2026. ; https://doi.org/10.64898/2026.04.17.719155doi: bioRxiv preprint