Variable specificity of memory trace reactivation during hippocampal sharp wave ripples

Abstract

Hippocampal sharp wave-ripples (SWR) are thought to mediate brain-wide reactivation of memory traces in service of memory consolidation. However, rather than the faithful replay of neural activity observed during a specific experience, reactivation in both the hippocampus and downstream regions is more variable. We suggest that variable reactivation is a unifying feature of recurrent brain circuits. In the hippocampus, self-organized activation during offline states is constrained by existing attractor manifolds, or maps, and may be biased toward particular mapped locations by salient experience, which results in the appearance of experience-specific replay. Similarly, the impact of SWR-associated reactivation on downstream regions is not a simple transfer of hippocampal representational content. Rather, the response of downstream regions depends on a transformation function, defined by both the feedforward and local circuit architecture, as well as the 'listening state' of the downstream region. We hypothesize that SWRs act as a multiplexed signal, the mnemonic specificity of which is largely determined by this transformation function, and discuss the implications of this framing for theories of systems consolidation.

Figure 1

Reading from SWR activity. Leveraging SWRs to broadcast information from the hippocampus assumes two stages of reading. In the primary reading stage, information from a memory trace existing in the hippocampal network is read out, in the form of a SWR. In the secondary reading stage, information is broadcast through feedforward connections to a DSR network. The information broadcast depends on the feedforward connectivity and its alignment with the SWR activity. This broadcast is received (purple spikes), and interacts with the ongoing local activity in the DSR network (black spikes). This interaction depends on local properties of the DSR network, including connectivity and state.

Figure 2

Specificity of SWR activity. The relevant information contained in a SWR exists on a theoretical spectrum of specificity: at the most general (left) the content of a SWR is completely random and unrelated to previous experience or local network architecture. At the most specific (right) the activity exactly replicates that which occurred during an experience. The range of possibilities is restricted to a physiologically plausible domain that accounts for the constraints imposed by the local network and SWR dynamical events. Towards the middle is a network attractor manifold subspace that acknowledges the low(er) dimensional space that activity is likely to occupy, which results from network architecture and local smooth manifolds corresponding to represented environmental variables. Experience may bias which or where an attractor is activated.

Figure 3

Reading multiple signals from one SWR. (a) Series of spikes across a population during a SWR. (b) Set of filters implemented by properties of DSR. Feed forward projections determine signal broadcast from SWR. Local weights in DSR network filter the subspace of inputs accessible to a network. (Note: In this column, we do not depict the local DSR network weights themselves, simply the subspaces of SWR activity accessible to the DSR given the local weights). DSR state determines the DSR’s responsiveness to inputs. (c) Resulting DSR responses to single SWR event given diverse DSR features. (d) DSRs can access different channels of information contained in the same SWR activity. Numerous aspects of secondary reading can vary across DSRs, granting them unique access to information contained in a SWR. A unitary SWR event can thus broadcast unique information to each DSR.