Interactions between Memory and New Learning: Insights from fMRI Multivoxel Pattern Analysis.

Abstract

Declarative memory—long-term memory for events and facts—is a key form of cognition that depends on distributed neural coding. Given the rich, multifaceted nature of life events, their neural representations (episodic memory “engrams”) typically incorporate a broad set of cortical and subcortical regions whose coding properties underlie event features (Paller and Wagner, 2002; Rugg et al., 2002; Tulving, 2002; Schacter et al., 2007). With continued experience, representations of individual events may form a foundation for more generalized semantic knowledge about the world (van Kesteren et al., 2012). A fundamental theoretical question is how existing memories interact with encoding of new experiences to enable formation of integrated knowledge structures. The distributed nature of memory content in the brain, both locally (i.e., across neurons within a region) and across relevant cortical and subcortical regions, creates challenges for measurement of mnemonic content across various stages of memory encoding and retrieval. By combining non-invasive imaging techniques (e.g., functional magnetic resonance imaging–fMRI) with multivariate pattern-analyses (MVPA), such representational content can be decoded from distributed patterns of brain activity (Polyn et al., 2005; Norman et al., 2006; Rissman and Wagner, 2012). Moreover, quantitative measures of mnemonic representations can be related to behavioral performance measures, thus informing mechanistic models of memory. At a macroscopic level, mnemonic representations of events are distributed across perceptual, motor, affective, and associative brain regions (Tulving and Markowitsch, 1997). Episodic memory retrieval entails the reinstatement or reconstruction of information encoded in memory (for reviews see Danker and Anderson, 2010; Ben-Yakov et al., 2015). MVPA provides a means of measuring distributed neural representations, and quantifying reinstatement processes (Norman et al., 2006; Rissman and Wagner, 2012). Importantly, a myriad of externally and internally generated retrieval cues can drive reinstatement of existing memory traces during encoding of related information. Such reinstatement may support the formation of more generalized knowledge through integration of new with old memories (Shohamy and Wagner, 2008; Preston and Eichenbaum, 2013). As such, elements of new memories that overlap with prior experiences can trigger reinstatement and integration processes allowing for extension and strengthening of existing associative knowledge structures, or “schemas” (Tse et al., 2007; van Kesteren et al., 2012). The medial temporal lobe (MTL)—with the hippocampus at its core—is the most prominently studied region in memory research (Burgess et al., 2002; Squire et al., 2004; Eichenbaum et al., 2007). The hippocampus serves as an integrative hub for the binding of disparate neocortical representations of event features into unified memories (Eichenbaum et al., 2004; Andersen, 2007). Through creating flexibly addressable memory traces that link to the driving cortical representations of event content, the hippocampus can support subsequent reactivation of a remembered event's feature representations in the neocortex during retrieval. MVPA techniques can index expressions of distributed memory representations and processes in MTL as they unfold, as well as probe reinstatement and integration processes in content-selective cortical regions (Polyn et al., 2005; Johnson et al., 2009; Staresina et al., 2012; Gordon et al., 2014; Sigman et al., 2014). Beyond the MTL, other cortical areas have been posited to contribute to across-event integration. In particular, the integration of associated memories is thought to also depend on computations within the medial prefrontal cortex (mPFC), a prefrontal region intimately connected with the hippocampus and suggested to be involved in the building of knowledge structures (van Kesteren et al., 2012; Preston and Eichenbaum, 2013). Recent evidence from direct neuronal recordings in non-human models of memory has linked hippocampus and mPFC population coding to the expression of schema knowledge (McKenzie and Eichenbaum, 2011; McKenzie et al., 2014; Richards et al., 2014). In humans, MVPA provides a powerful means to assess how mPFC and the hippocampus underlie integration of newly learned experiences with existing memories, and critically, to link this integration process with cortical reinstatement (Dudai and Eisenberg, 2004; Kuhl et al., 2010; Nadel et al., 2012). Here we review how MVPA, applied to fMRI-data, is leveraged to address fundamental questions about reinstatement and subsequent integration of memory representations in the human brain. We discuss a framework in which reinstatement of prior knowledge during new learning can facilitate formation of integrated knowledge across experiences, highlight evidence for potentially disruptive effects of such processes on other expressions of memory (e.g., memory for episodic details), and suggest future research directions.