The aim of this study was to contrast the theory of separate memory stores with the theory of working memory as activated long-term memory, by investigating the behavioral and neuroanatomical correlates of working and episodic memory in a stroke population. To this end, we used a task design in which working memory and episodic memory are assessed based on the same encoding phase. We used behavioral and neuroimaging data to investigate 1) the relation between visual working memory and episodic memory performance in stroke patients and older adults and 2) anatomical correlates of visual memory function using multivariate voxel-based, atlas-based, and track-based approaches. We found that discriminability in working memory and episodic memory were independent at the behavioral level. In contrast, response bias was correlated between working memory and episodic memory in stroke patients. LSM analyses suggested there might be independent regions that are associated with working memory and episodic memory performance.
The key issue in the ongoing debate on the multicomponent model of memory versus the view of working memory as activated long-term memory, is the need of a separate and independent short-term memory store (Baddeley et al. 2019; Cowan 2019; Norris 2017; Norris 2019; Oberauer 2009; Shallice and Papagno 2019). According to the multicomponent model, a separate store and mechanism is needed to construct new representations and actively maintain relational information (Norris 2017, 2019). The theory of activated long-term memory states that this can be achieved by rapid new learning, in which new associations can be formed as new long-term memory trace. While the multicomponent model of memory explains long-term memory deficits as the failure to encode a representation into long-term memory, the theory of activated long-term memory interprets this as a failure of consolidation of rapidly formed new long-term memory traces (Cowan 2019). If rapidly formed representations underlie associative memory, interference or a deficit in consolidation explains low performance on the subsequent memory task but does not explain low performance solely on the working memory task. Our results suggest that there might be separate representations in working memory and episodic memory as discriminability is not correlated between the tasks and some patients show selective impairment. Response bias on the other hand might rely on common neural substrates in working memory and episodic memory as this does correlate between the tasks in stroke patients.
Results from the LSM analyses show independent regions that are stronger associated with working memory and episodic memory performance. Lesions in the arcuate fasciculus in the right hemisphere were more strongly associated with discriminability in working memory than in subsequent memory, while lesions in the frontal operculum in the right hemisphere were more strongly associated with criterion setting in subsequent memory than in working memory. As we included the scores for discriminability and criterion on the other task as covariate, we can state that there is a stronger association for one task than for the other with lesion status in these regions. The arcuate fasciculus connects the perisylvian cortex of the frontal, parietal, and temporal lobes. In the left hemisphere the three segments of the arcuate form the perisylvian language network, which is extensively studied (e.g. Bonakdarpour et al. 2019; Catani et al. 2005). The left anterior segment has been associated with the phonological loop, specifically with order errors (Papagno et al. 2017). The right arcuate fasciculus has been studied less extensively, but available studies associated lesions in this region with spatial neglect (Catani and Thiebaut de Schotten 2008; Machner et al. 2018), visuospatial processing (Rolland et al. 2018), and visual working memory (Chechlacz et al. 2014; Matias-Guiu et al. 2018). We found discriminability on the working memory task, compared to the episodic memory task, to be stronger related to lesions in the anterior and long segment of arcuate fasciculus based on multivariate and atlas-based analyses. Track-based analyses demonstrated an association with the posterior segment of the arcuate fasciculus only for discriminability on the working memory task but not the subsequent memory task. As the working memory task is based on temporal order, our findings converge with previous results for verbal order information in the left anterior segment of the arcuate fasciculus. The posterior segment of the arcuate fasciculus connects Wernicke’s areas to the inferior parietal lobe. Previous studies have identified the right inferior parietal lobe to be involved in reorienting attentional focus to memory representations of previously attended stimuli (Kizilirmak et al., 2015). Based on our results and previous findings we suggest that the right arcuate fasciculus might be associated with the visuospatial sketchpad. The different results concerning the different segments of the arcuate fasciculus are likely due to the difference between a binary atlas (CAT) and a probabilistic atlas (Tractotron). The cluster identified with the multivariate analyses overlaps with the anterior and long segment of the arcuate fasciculus based on the CAT atlas, but the probabilistic atlas used by Tractotron shows that this cluster also overlaps with the posterior segment. All analyses indicate that the right arcuate fasciculus is more involved in working memory than subsequent memory. DTI analyses in a future study should give more insight in the role of different segments of the right arcuate fasciculus in working memory. As track-based analyses provide better evidence for behavioral correlations for white matter lesions, this might be an indication that specifically the posterior segment of the right arcuate fasciculus is essential for visual working memory.
Criterion setting was stronger associated with the frontal operculum for subsequent memory compared to working memory. It is interesting to note that for criterion setting we only found an association with lesion status in the frontal operculum for the subsequent memory task while criterion was correlated between the two tasks at the behavioral level. Even though the correlation was statistically significant, the correlation was weak. The correlation might be explained by a third factor influencing response bias on both tasks even if they have different neural substrates. A possible factor related to response bias is age (for a meta-analysis see Fraundorf et al., 2019). The frontal operculum has been described as essential in exerting control over cognitive processes (Takayasu et al., 2011). It was shown to be related to selective attention and to regulate activity in occipitotemporal areas involved in the processing of different classes (faces, houses, bodies) of visual stimuli (Takayasu et al., 2011). A second study found evidence for activation of the frontal operculum during interference tasks that required response inhibition (Wager et al., 2005). A third study showed that resisting bias based on irrelevant previous information was associated with activation of the frontal operculum (Scholl et al., 2015). These findings converge with our results which suggest that damage to the frontal operculum might result in a stronger response bias.
The results from our LSM analyses should be interpreted with caution because the associations between memory performance and lesion location were no longer significant after correction for lesion volume. Larger lesion volume was associated with lower discriminability on the working memory task and stronger response bias in the subsequent memory task. The finding that lesion volume is associated with performance does not nullify the result that specific regions in the brain are stronger related with the one memory task compared to the other.
Our results partly converge with a previous study in stroke patients on discriminability and criterion setting in verbal recognition memory. Like in our study, Biesbroek et al. (2015) reported that the right inferior frontal gyrus/frontal operculum is crucial for criterion setting. This study indicated that the left medial temporal lobe, left temporo-occipital structures, both thalami, and the right hippocampus are associated with discriminability (Biesbroek et al. 2015). Two main differences should be pointed out, the verbal versus visual nature of the task and the distribution of lesions. Lesion symptom-mapping studies rely heavily on the total lesion prevalence distribution resulting in differences between studies. Previous studies have shown different neural correlates for verbal and visual memory (e.g. Donolato et al. 2017).
The advantage of studying stroke patients is that due to the sudden nature of the brain damage, it is acceptable to infer causal relations (Karnath et al. 2019; Rorden and Karnath 2004). A critical comment is that people with stroke might have a higher vascular burden that is related to memory function (Van Leijsen et al. 2019). There might be a selection bias in the sample with patients with mild symptoms and small lesions being more likely to participate in research. This has a consequence for the distribution of lesions across the brain, though this is party inherent to the population studied. Brain lesions due to stroke are determined by the vascular tree resulting in vulnerable lesion sites and intercorrelation between voxels. Even though there might be locations in the brain crucial to a specific task that are rarely affected by stroke, these would not be considered as main associates for post-stroke memory deficits. A limitation remains that we can only draw conclusions on the voxels/ROIs with sufficient lesion coverage and that some areas typically associated with memory, like the hippocampus, were not included in the analyses.
The aim of the study was to investigate how stroke patients can give insight into shared and distinct processes in working memory and episodic memory. Due to the limited lesion coverage that is, however, typical for a stroke sample (Zhao et al., 2018), we cannot make any claims on hippocampal/medial temporal lobe structures that may or may not be involved in both visual working memory and episodic memory. However, the current study does give an indication that other brain regions are also associated with working memory performance (the right arcuate fasciculus) and with criterion setting in episodic memory (the right frontal operculum). Furthermore, the behavioral data provide moderate evidence that discriminability in visual working memory and episodic memory are unrelated, supported by a lack of correlation and by selective impairments. They also give strong evidence that criterion setting in working memory and episodic memory are correlated.
With the task design we used, we aimed to assess working memory and episodic memory in one task design, using the same stimuli, the same encoding phase and comparable binding demands. The difficulty is assessing two different processes in a comparable task with limited confounding factors differentiating between which processed is tapped into. It is important to stress that both tasks involve context binding. The 2-back task is a temporal order binding task, so not based solely on object recognition as all objects appeared twice in the same block. The subsequent memory task assessed spatial binding. A few limitations concerning the task design should be mentioned. First, although several studies indicate that contextual binding for time and space relies on the hippocampus (e.g. Eichenbaum 2017; Yonelinas et al. 2019), they might not have fully overlapping neural correlates. A recent study showed that different subregions within the hippocampus were differently associated with object-location, object-time and object-object associations in development from childhood into adolescence based on structural MRI in 171 subjects (Lee et al. 2020). Furthermore, an fMRI study of 16 healthy subjects showed activations in specific areas for spatial order (parahippocampus) and temporal order processing (Brodmann area 10 within the prefrontal cortex), in addition to general hippocampal involvement for source retrieval (Ekstrom et al. 2011). Given these results, we cannot fully rule out the possibility of stroke selectively affecting different types of binding. Second, hyper-binding might differently affect the working memory and subsequent memory task. In the ageing literature hyper-binding refers to the inability of older adults to inhibit irrelevant information resulting in lower performance on a working memory task but enhanced performance when the previously irrelevant information is subsequently tested (e.g. Campbell et al. 2010). However, in our design we do not expect this to have a large influence. Even though location was not relevant during the 2-back task, the information was not conflicting and could even be used as a cue as a target could only be in the same location as two trials previously. Secondly, hyper-binding only occurs under fully implicit instructions (Campbell and Hasher 2018). In our task, participants are made explicitly aware of the link between the tasks. Campbell and Hasher (2018) showed that the effect of hyper-binding in older adults disappears when made aware of the connection between the tasks. Finally, our previous study in which we studied the effect of age on memory with this task design, did not show an advantage for older adults on the subsequent memory task (for a more extensive discussion on the task design see Lugtmeijer et al. 2019). A third difference is in task encoding, the subsequent memory task is unexpected. While encoding for working memory is typically shallow and based on rehearsal, encoding for a planned long-term retention task is more elaborative, which is beneficial for episodic memory but less essential for working memory (Cowan 2019; Craik and Watkins 1973). While this might result in associations with different neural substrates than typically found in explicit episodic memory tasks, these instructions ensure that the encoding strategy is not different between the tasks. Therefore, this design is more sensitive to detecting possible overlapping substrates for working memory and episodic memory. A second possible difference in encoding might be verbalization. The working memory task can be supported by verbal labelling of the objects (e.g. apple, car, apple). The participants were instructed that the second appearance of an object was always on the same location as the first so location could be used as a cue (e.g. apple right lower corner) but as location was irrelevant to the working memory task, it is unknown whether participants included that in their verbal label. Furthermore, our LSM analyses do not indicate a dominant role for verbalization. We identified right hemispheric counter parts of typical language areas to be associated with working memory performance.
For clinical cognitive assessment it is relevant to take into account that stroke patients might have an altered response bias, especially because our results show that stroke can affect response bias towards a more liberal and a more conservative bias.
In conclusion, stroke can result in both working memory and episodic memory deficits. This study indicates that discriminability in working memory and episodic memory are two distinct processes, while criterion setting might be a shared process. LSM analyses suggested that independent regions are stronger associated with visual working memory (right arcuate fasciculus) and criterion setting in episodic memory (frontal operculum).