changes in spatial and temporal context (absent that carefully track and manipulate the 6. Schapiro, A. C. et al. Human hippocampal replay during rest prioritizes weakly learned information any retrieval driving context reinstatement). influence of the on cortical and predicts performance. Nat. Commun. 9, But replay seems to be more persistent and representations15 — we think the evidence 3920 (2018). 7. Ji, D. & Wilson, M. A. Coordinated memory adaptive than this, as it can occur as frequently already points to replay having a critical and replay in the visual cortex and hippocampus for a remote spatial context as for the current active role in driving consolidation across during . Nat. Neurosci. 10, 100–107 (2007). 8 2 . Wilber, A. A. et al. Laminar organization of encoding environment ; has been observed 10 hours memory systems. and memory reactivation in the parietal cortex. Neuron 95, 1406–1419 (2017). after exposure to a novel environment, with 1 2 James W. Antony * and Anna C. Schapiro * 9. Zhang, H., Fell, J. & Axmacher, N. Electrophysiological stronger activity during sleep than wake 1Princeton Neuroscience Institute, Princeton University, mechanisms of human . periods3; and, critically, can occur more for Princeton, NJ, USA. Nat. Commun. 9, 4103 (2018). 4 5 2Department of Psychology, University of 10. Khodagholy, D., Gelinas, J. N. & Buzsáki, G. infrequently experienced , gradually learned Learning-enhanced​ coupling between ripple Pennsylvania, Philadelphia, PA, USA. and weakly encoded6 information. These oscillations in association cortices and hippocampus. *e-mail:​ [email protected]; Science 358, 369–372 (2017). findings may not be strictly inconsistent with [email protected] 11. Xia, F. et al. Parvalbumin-​positive interneurons mediate neocortical-​hippocampal interactions that CB, but they are not motivated by it; additional are necessary for memory consolidation. eLife 6, mechanisms would be needed to explain why e27868 (2017). context is more strongly reinstated in these 1. Yonelinas, A. P. et al. A contextual binding theory 12. Antony, J. W. et al. Sleep spindle refractoriness of episodic memory: systems consolidation segregates periods of memory reactivation. Curr. Biol. situations, especially during sleep. reconsidered. Nat. Rev. Neurosci. 20, 364–375 28, 1736–1743 (2018). We think there is strong evidence that (2019). 13. Ngo, H. V. et al. Auditory closed-​loop stimulation 2. Karlsson, M. P. & Frank, L. M. Awake replay of remote of the sleep slow oscillation enhances memory. sleep benefits memory beyond the reduction experiences in the hippocampus. Nat. Neurosci. 12, Neuron 78, 545–553 (2013). of contextual interference, and that this 913–918 (2009). 14. Maingret, N. et al. Hippocampo-​cortical coupling 3. Giri, B. et al. Hippocampal reactivation extends for mediates memory consolidation during sleep. active process drives systems consolidation several hours following novel experience. J. Neurosci. Nat. Neurosci. 19, 959–964 (2016). (as defined above). If sleep primarily bene­ 39, 866–875 (2019). 15. Mao, D. et al. Hippocampus-​dependent 4. Gupta, A. S. et al. Hippocampal replay is not a simple emergence of spatial sequence coding in retrosplenial fits hippocampus-​dependent memory by function of experience. Neuron 65, 695–705 (2010). cortex. Proc. Natl Acad. Sci. USA 115, 8015–8018 reducing interference or through local 5. van de Ven, G. M. et al. Hippocampal offline (2018). reactivation consolidates recently formed cell consolidation processes, specific active assembly patterns during sharp wave-​ripples. Competing interests cortical events and hippocampal–cortical Neuron 92, 968–974 (2016). The authors declare no competing interests. interactions during sleep should not be robustly and causally related to later memory. However, cortical replay coincides with hippocampal replay7 and high-​frequency Reply to ‘Active and effective replay-​associated bursts called ripples8, and this coupling is associated with later replay: systems consolidation memory9. Hippocampal and neocortical ripples coincide and their coupling increases reconsidered again’ with learning10, and disrupting the coupling between hippocampal ripples and cortical sleep spindles impairs memory retention11. Andrew P. Yonelinas , Charan Ranganath, Arne D. Ekstrom In addition, optimal replay relies on the and Brian J. Wiltgen potential for spindles to occur12, and arti­ ficially boosting individual slow oscilla­ In their Correspondence article (Active and sharp wave–ripple (SWR) activity in the tions increases spindle power and improves effective replay: systems consolidation recon­ hippo­­campus correlates (in some instances) memory13. These processes seem to promote sidered again. Nat. Rev. Neurosci. https:// with activity in the cortex3, are consistent with systems consolidation: timing optogenetic (2019))1, other theoretical accounts such as CB. That is, stimulations of the neocortex precisely to Antony and Schapiro agree that the concept because episodic memory involves hippocam­ hippocampal ripples enhances endogenous of contextual binding (CB) described in our pal binding of item information and context hippocampal–neocortical coupling and alters Opinion article (A contextual binding theory information that is in the cortex, residual or neocortical neuronal spiking patterns that of episodic memory: systems consolidation potentiated encoding activity across the hip­ support behaviour14. reconsidered. Nat. Rev. Neurosci. 20, 364–375 pocampus and cortex should be observed Yonelinas et al. argue that replay primarily (2019))2 is crucial for explaining episodic after the nominal encoding event is over, even reflects prior memory formation rather than memory, and that many aspects of standard if the hippocampus is not actively training driving subsequent memory transformation. systems consolidation theory are probably the cortex at the time. Furthermore, although However, they acknowledge that post-​ incorrect. They also point to some recent sleep experimental manipulations of SWRs in the encoding hippocampal activity may cause and replay studies that might be taken as evi­ hippocampus and/or neocortex can affect local cellular consolidation or re-​encoding dence for a form of systems consolidation that subsequent memory4, the studies performed that could sometimes “lead to the formation maintains that the hippocampus rapidly trains to date do not show whether those manipula­ of strong neocortical semantic representations the cortex during offline periods of sleep or tions prevented the hypothesized transfer of that could support decontextualized memory rest. Although we agree that these findings are information from the hippocampus to the cor­ for remote events”. This latter mechanism fits compatible with a modified form of systems tex, or whether they affected the hippocampal well into our conceptualization of systems consolidation, we contend that the studies or cortical representations themselves. For consolidation; we contend that this is a feature, summarized by Antony and Schapiro do not instance, it would be reasonable to infer that rather than a side effect, of replay. Although necessitate such an account. interfering with hippocampal function dur­ more work is needed for full confidence For example, demonstrations that replay ing an SWR could disrupt a hippocampal (or in this contention — such as experiments correlates with subsequent memory, or that cortico-hippocampal) memory trace, or that

NAture Reviews | Neuroscience volume 20 | August 2019 | 507 Correspondence

reactivating hippocampal memory traces dur­ on replay and its potential role in systems 2Department of Psychology, University of Arizona, ing a delay could affect subsequent memory consolidation has focused largely, if not entirely, Tucson, AZ, USA. *e-mail:​ [email protected] performance, even if the hippocampus is not on reactivation of place cell sequences during active in training the cortex. sleep and quiet wakefulness, and have focused Although the present evidence does not on examining activity observed in well-​learned 1. Antony, J. W. & Schapiro, A. C. Active and effective replay: systems consolidation reconsidered again. compel a systems consolidation account, spatial environments. It would be important Nat. Rev. Neurosci. future studies could provide more definitive to know whether one can observe this type of s41583-019-0191-8 (2019). 2. Yonelinas, A. P. et al. A contextual binding theory evidence. For example, studies show that as replay in tests of memory in humans that are of episodic memory: systems consolidation little as 60 minutes of sleep leads to consid­ more analogous to those described in theories reconsidered. Nat. Rev. Neurosci. 20, 364–375 (2019). erable reductions in forgetting. If these rapid of systems consolidation. Regardless of whether 3. Khodagholy, D., Gelinas, J. N. & Buzsáki, G. Learning-​ effects of sleep reflect systems consolidation, future studies of post-​encoding activity are enhanced coupling between ripple oscillations in association cortices and hippocampus. Science 358, hippocampal lesions should produce accel­ ultimately found to provide stronger support 369–372 (2017). erated rates of forgetting over delays that for context models or systems consolidation 4. Xia, F. et al. Parvalbumin-​positive interneurons mediate neocortical-​hippocampal interactions that include sleep. Studies reviewed in our paper models, we agree with Antony and Schapiro are necessary for memory consolidation. eLife 6, 5 do not support this prediction , but perhaps that such studies will continue to have an e27868 (2017). future studies will show otherwise. important impact on the field. 5. McKee, R. D. & Squire, L. R. Equivalent forgetting rates in long-​term memory for diencephalic and J. Neurosci. 12 Another important question for future 1 1 medial temporal lobe amnesia. , Andrew P. Yonelinas *, Charan Ranganath , 3765–3772 (1992). research will be to determine whether replay Arne D. Ekstrom2 and Brian J. Wiltgen1 observed in rodents is related to putative 1Center for Neuroscience, University of California, Competing interests replay reported in humans. Rodent research Davis, CA, USA. The authors declare no competing interests.

508 | August 2019 | volume 20