Episodic memory is a form of memory defined by consciously accessible memory for specific events (Tulving, 1983). Episodic memory is a fundamental type of memory that allows us to “time travel” back through subjective time to remember and re-experience past events (Tulving, 2002). Through retrieval of relevant information from past events, episodic memory can guide behavior in service of our current and future goals (Conway & Pleydell-Pearce, 2000). Episodic memory and semantic memory (factual knowledge) together comprise declarative memory, consciously accessible memory for events and facts (Squire, 2004). A wealth of experimental evidence indicates that declarative memory function is supported by the structures in the medial temporal lobe, including the hippocampus, entorhinal cortex, perirhinal cortex, and parahippocampal cortex (Squire, 2004; Moscovitch et al., 2016).

Autobiographical memory (AM), defined as memory for specific events from one’s own past, combines both episodic and semantic memory content (Cabeza & St. Jacques, 2007; Rubin, 2005; Piolino et al., 2009). For example, an AM of a trip to a cafe might include vivid episodic recollection of a social interaction, together with semantic knowledge about the café and new people who were met there. Autobiographical memories are fundamentally important to one’s sense of self and play important roles in a wide variety of domains ranging from social behavior, problem solving, and emotion regulation (Bluck, 2003; Cabeza & St. Jacques, 2007; Nelson & Fivush, 2020).

A large and growing literature of neuroimaging studies has investigated the neural basis of AM retrieval (for reviews see Cabeza & St Jacques, 2007; Daselaar et al., 2008; Rugg & Vilberg, 2013; Moscovitch et al., 2016). The brain regions reported to be active during AM retrieval differ considerably across studies, reflecting a wide range of factors such as differences in the tasks used to elicit retrieval, differences in control tasks, and the relatively limited statistical power of many fMRI studies. As with many cognitive domains, the findings from multiple studies need to be combined and synthesized to yield a comprehensive view of the regions that are consistently involved in AM retrieval. Although narrative reviews that are qualitative in nature can be informative, quantitative meta-analyses have several advantages over qualitative reviews (Albajes-Eizagirre & Radua, 2018; Laird et al., 2005; Mü;ller et al., 2018; Radua & Mataix-Cols, 2012).

Quantitative meta-analysis methods such as Activation Likelihood Estimation (ALE) and Seed-based d Mapping (SDM) are coordinate based meta-analysis (CBMA) methods that identify regions of the brain that are consistently activated across numerous studies (Albajes-Eizagirre & Radua, 2018; Laird et al., 2005; Witteman et al., 2019; Boccia et al., 2019; Addis et al., 2016). Because the original neuroimaging data from individual studies are typically not available, CBMA methods use the peak coordinates of activations reported in each published study to estimate the original functional activation maps. These estimated activation maps are then pooled and analyzed at the voxel level to yield meta-analytic summaries of regions that are consistently activated across studies.

Previous quantitative meta-analyses have been conducted to identify the common regions associated with AM retrieval (Spreng et al., 2009; Addis et al., 2016; Boccia et al., 2019). However, these previous studies have limitations in terms of sample size, study selection criteria, and analytic method (see Discussion for a detailed discussion). To address these issues, we examined the neuroimaging correlates of AM retrieval in the current study, using the largest sample of neuroimaging studies to date (50 studies), improved study selection criteria, and a different analysis method, Seed-based d Mapping with Permutation of Subject Images (SDM-PSI) (https://www.sdmproject.com).

A key advantage of SDM over other CBMA methods such as ALE is that SDM takes into account variations in the effect size of each reported activation maximum (based on their maximum t values), better representing the original neuroimaging results, whereas ALE does not explicitly factor in effects sizes and treats all statistically significant maxima equivalently. Other advantages of SDM include its sophisticated random-effect modeling of subject-level data (which increases reliability and performance of the analysis) and the use of threshold-free cluster enhancement (TFCE) to control for multiple comparisons, a method that has advantages relative to cluster-based thresholding methods (Albajes-Eizagirre et al., 2019). Finally, because the results of neuroimaging meta-analyses are frequently used for many purposes such as providing regions of interest for fMRI or brain stimulation studies, or to interpret new neuroimaging findings, our goal was to make the meta-analysis results maximally useful by making all the SDM meta-analysis results images freely available online (on the Mendeley repository https://data.mendeley.com/datasets/w9p86fndr7), in contrast to previous meta-analyses of AM retrieval, which have not provided public open access to their results images.

In addition to characterizing the neuroimaging correlates of AM retrieval, we also addressed how two important methodological variables may affect these neuroimaging correlates. One of these differences concerns whether the memories being retrieved have been pre-rehearsed prior to neuroimaging scanning or alternately, are retrieved for the first time during scanning. Most AM retrieval studies cue memory retrieval with a method where single word cues are presented and participants are instructed to retrieve an AM that is related either directly or indirectly to the cue (the Galton-Crovitz cueing method; Crovitz & Schiffman, 1974; Galton, 1879). In some neuroimaging studies, these cues are presented to participants prior to scanning and participants retrieve AM memories. This cue-rehearsed retrieval method is advantageous because experimenters know in advance which memories will likely be retrieved to the cues during scanning, affording greater experimental control. However, due to concerns over the possible effects of prior rehearsal on sequent memory retrieval processes, other neuroimaging studies have presented participants with retrieval cues for the first time in the scanner, with no prior rehearsal (Cabeza & St. Jacques, 2007). We will refer to studies in which AM retrieval is cued with no prior rehearsal as studies that use the cue-novel retrieval method. Prior rehearsal has been suggested to have a particularly prominent effect on the initial memory search processes involved in AM retrieval (Nadel et al., 2007). One previous study found that activation in areas supporting episodic and semantic memory retrieval decreased with increasing amounts of memory rehearsal (Svoboda & Levine, 2009). However, the impact of this key methodological factor has yet to be systematically examined in a meta-analytic study. Accordingly, we examined this question by comparing the activations from neuroimaging studies using cue-rehearsed retrieval to those from studies using cue-novel retrieval.

A second important methodological variable we examined is the type of control task used to compare with the primary AM retrieval task. As with any neuroimaging study, the type of control task used can have a considerable influence on the activation results. A particularly important consideration is the degree to which the chosen control task recruits cognitive processes that overlap with those involved in AM retrieval. A key difference in the control tasks used in neuroimaging studies examined in this study is the extent to which perceptual/attentional vs. semantic processing is engaged during the control task. Several studies have used visuo-attention tasks such as pressing a button in the direction of an arrow or performing a simple arithmetic task, whereas another large set of studies have used semantic memory retrieval tasks such as generating category exemplars. We predicted that AM retrieval studies that used semantic retrieval tasks as control conditions would be associated with decreased activation in regions associated with semantic memory retrieval (due to subtraction of semantic memory retrieval activations associated with the semantic control task), relative to studies that used visuo-attention control tasks (which involved minimal or no semantic memory retrieval). To examine this issue, we compared the meta-analysis results between studies that used visuo-attention tasks vs. semantic control tasks.

In summary, the goal of the current study was to update the current understanding of the neural basis of AM retrieval using the largest set of neuroimaging studies of AM retrieval to date and a powerful meta-analytic approach (SDM), to provide an updated, more representative summary of the regions involved in AM retrieval. While we expected our results to have broad similarities with prior neuroimaging meta-analysis studies and qualitative reviews of AM retrieval (e.g., medial temporal regions such as the hippocampus, prefrontal and parietal regions, among several other regions), given the differences in approach and the studies included, we expected our meta-analysis to potentially yield different results from prior meta-analytic studies. Finally, we addressed two novel questions about how two important methodological factors (type of retrieval task and type of control task) influence these meta-analysis results.