Neural Correlates of Verbal Working Memory: An fMRI Meta-Analysis
Emch, von Bastian, Koch · 2019
공개 학술지 Frontiers in Human Neuroscience 에 실린 총설 논문입니다(2019년). CC BY 4.0 으로 공개된 글이고, 그림 · 표 · 참고 문헌 표시를 뺀 본문을 실었습니다.
출처 Emch et al. (2019), Frontiers in Human Neuroscience · 라이선스 CC BY 4.0 · 원문
이 장은 원문만 있습니다. 문장별 해석은 앞 장부터 차례로 붙입니다.
3장 · Neural Correlates of Verbal Working Memory: An fMRI Meta-Analysis — Discussion
The present comprehensive meta-analysis across 42 whole-brain vWM fMRI tasks showed vWM processing to be based on a fronto-parieto-cerebellar network and to involve also subcortical regions such as the cingulate, left insula and right lenticular nucleus. Thus, the present results corroborate previously discussed networks, but also provide evidence for the involvement of additional regions that have been neglected in the past in the discussion of vWM processing.
Dual-Selection Model
A tentative explanation of the results is provided by the dual-selection model. Nee et al. (2013) proposed this model based on a meta-analysis of 36 event-related fMRI studies aimed at understanding the executive processes of WM. According to this model the caudal superior frontal sulcus (SFS) is associated with a spatial selection while the mid-lateral PFC is especially sensitive to non-spatial content, matching the “where” and “what” based selections, respectively. This proposal was further corroborated by a previous meta-analysis of 24 experiments based on an n-back task (Owen et al., 2005). The results of the present meta-analysis lend further support to the dual-selection model given that we also found activation of the mid-lateral PFC (bilateral IFG, right MFG, and medial part of the left SFG). The left SFG appeared to be a heterogeneous region. The fact that we found a noticeable bilateral prefrontal activation in this meta-analysis suggests that the assumption of a strongly left-lateralized verbal WM activation in PFC should be reconsidered. However, it needs to be clarified that we did not include studies systematically comparing spatial vs. non-spatial WM. Therefore, these conclusions need to be drawn with caution because the mere fact that we found the same activation does not fully support the dual-selection model.
Phonological Loop
We did not find activation of the left supramarginal gyrus, which is known to be important for the phonological store, but instead in the region where the supramarginal gyrus is located, the left inferior parietal cortex. The activation of this region was found to be heterogeneous, which tends to be in line with the hypothesis of Buchsbaum and D'Esposito (2008). They argue that the phonological store does not precisely correspond to a single specific functional brain region, but rather is associated with several brain regions that underlie neural processes from perception and production of speech. Surprisingly, the present meta-analysis did not reveal any activation in the Wernicke area although this is an essential area for the comprehension and/or production of verbal material (Binder, 2015). This area is assumed to comprise mainly the posterior part of the superior temporal gyrus as well as the occipito-parieto-temporal junction including the angular gyrus. However, the exact location of the Wernicke area is still a matter of debate also due to its comprehensive and partly heterogeneous functionality in the context of verbal processing. Moreover, the fact that we used the AAL atlas, which comprises relatively large brain regions, might also explain why we were not able to isolate activation of this specific and somewhat ill-defined region. The fact that we did not find any activation in the right parietal cortex was also expected, since this region is assumed to serve spatial rehearsal. Hence, as opposed to the bilateral activation in the prefrontal cortex, activation in the parietal cortex turned out to be strongly left-lateralized, presumably due to modality. As predicted, we also found activation in the left IFG containing the Broca's area, as well as in the left SMA, which are components central to the articulatory loop. The fact that we did not find any activation in the premotor cortex could also be due to the atlas used. The AAL atlas does not contain this region because the labeled SMA embeds both the premotor cortex and the pre-SMA. Therefore, we cannot exclude that there was specific activation of the premotor cortex. In addition, there was activation in the left rolandic operculum, which is caudally adjacent to Broca's area. It has been demonstrated that this brain area is involved in speech production (Koelsch et al., 2009) and speech prosody processing (Wu et al., 2017). To the best of our knowledge, this is the first WM meta-analysis isolating activation specifically in this area. This finding may indicate that the majority of the participants used an overt rehearsal strategy during the vWM tasks. Further studies testing the impact of the opportunity for rehearsal during vWM tasks on brain activation are however needed to confirm this hypothesis.
Attentional Control
Many conceptualizations of WM include an attentional control mechanism that supports dealing with interferences such as from other items in memory. The right IFG has been proposed to be an important region for attentional control (Aron et al., 2003; Forstmann et al., 2008). Specifically, Aron et al. 's (2003) data strongly suggest that response inhibition is uniquely located in the right IFG, in particular in its triangular part. The data were acquired by studying patients with lesions of the right frontal lobe during a go/no-go task. Forstmann et al. (2008) found a direct linkage between structural and functional properties of the right IFG, and its role in response inhibition. Another fMRI study (Aron and Poldrack, 2006) found that the IFG targets the subthalamic nucleus (STN) and regions in its vicinity. The STN sends excitatory projections to the globus pallidus externus, which, in turn, suppresses the thalamo-cortical output; this is assumed to lead to an inhibition of the initiated response. Finally, a strongly right-lateralized network comprising the right IFG, the STN, and also the pre-SMA, is recruited during response suppression (Aron, 2007). It remains unclear, however, whether the right IFG triggers the STN directly or via the pre-SMA (Aron et al., 2014). We found a strong activation of the right IFG, especially in the triangular part, the right SMA—which also includes the pre-SMA in this atlas –, and the right pallidum, giving support to the idea that these areas constitute a network subserving response inhibition in the context of vWM processing. Indeed, a substantial body of behavioral research has found that attentional control as employed in response inhibition tasks is related to WM capacity (Kane and Engle, 2003; Unsworth and Engle, 2007, but see Rey-Mermet et al., 2019). Notably, heterogeneity analyses confirmed the stability of these networks indicating that activation in these regions is not likely due to a possible publication bias. However, it is important to mention that the selected studies did not manipulate attention. The fact that we found activation in the same areas that mediate response inhibition in other experimental contexts does not completely mean that they do so in the context of vWM.
In addition to the IFG, the angular gyrus has been found to be activated in the context of response inhibition (Wager et al., 2005). The angular gyrus is located in the posterior part of the inferior parietal lobule and has been found to be activated in a variety of tasks (Seghier, 2013). Some anatomical studies (Makris et al., 2005, 2009; Uddin et al., 2010) define the angular gyrus as an important seed point, given its strong interaction with temporo-frontal subsystems as well as regions such as hippocampus, caudate, and precuneus. It is a key component of the default-mode network and shows activation in most tasks demanding information retrieval (Spaniol et al., 2009; Kim, 2010). The role of this region in memory retrieval is plausible given its strong connectivity with the hippocampus. To the best of our knowledge, this is the first time that the right angular gyrus appears in a vWM meta-analysis. Considering that this region has been reported to be important for inhibition and retrieval we conclude that the activation of the angular gyrus in the present meta-analysis may predominantly reflect the employment of attentional control during information retrieval. Although we cannot exclude the possibility of this region's activity being found due to presence of publication bias in the selected literature. Further studies allowing for a separate analysis of the retrieval process are however needed to further explore this assumption.
Cerebellar and Subcortical Activations
It has been shown that cerebellum is connected not only to motor areas, but also to prefrontal cortical areas (Schmahmann, 1996); this suggests an involvement of the cerebellum in higher-order cognitive processes. A distinct cross-cerebro-cerebellar circuitry for vWM has been proposed with predominant involvement of right cerebellum, especially the lobule VI (Ng et al., 2016). In accordance with this proposal, earlier studies already pointed at the relevance of the right cerebellum in the context of vWM. Using inhibitory continuous theta burst stimulation (cTBS) Tomlinson et al. (2014) found that participants were less accurate during a verbal version of the Sternberg task if a trial was preceded by a stimulation of the right cerebellar hemisphere. Moreover, patients with right-sided cerebellar lesions have been found to be impaired in verbal memory, whereas patients with left cerebellar lesions turned out to be slower in a visuospatial task (Hokkanen et al., 2006). All these findings suggest a lateralized function of the cerebellum with its right hemisphere contributing mainly to verbal and its left hemisphere to visuospatial processing. Moreover, a meta-analysis (Stoodley and Schmahmann, 2009) analyzing cerebellum neuroimaging studies found that regions involved in vWM studies overlap with those involved in language tasks which is in agreement with domain-specific storage modules as in Baddeley's model. It corroborates the idea that vWM is more right-lateralized with a strong activation occurring mostly at the junction lobule VI/Crus I. Our results showing a significantly stronger activation in the right cerebellum (crus I) support this hypothesis. A case study of a right cerebellar hemispherectomy in an 18-years-old patient reported that the patient suffered from a disproportionate impairment of the rehearsal system, while the phonological store was preserved (Silveri et al., 1998). This could be due to anatomical connections between Broca's area, left SMA, right lobule VI and crus I of the cerebellum (Schmahmann, 1991). However, in the present meta-analysis, we did not differentiate between those processes and, thus, we cannot further investigate whether the right cerebellum is mainly involved in rehearsal. Still, our analysis provides clear evidence for the relevance of the right cerebellum, especially crus I, in the context of vWM processing. Further studies disentangling the different vWM processes are warranted to elucidate the specific function of the right cerebellum in vWM.
Apart from the cerebellum, a number of additional subcortical areas are assumed to be relevant for vWM. Thus, basal ganglia regions, especially the caudate and the lenticular nucleus, have been found to be activated during encoding and maintenance phases during vWM tasks (Lewis et al., 2004; Chang et al., 2007; Moore et al., 2013). Although in the present meta-analysis basal ganglia activation was restricted to the right lenticular nucleus, it remains unclear whether the activation is ascribable to these processes. Again, we were unable to distinguish between the different vWM processes given the available data. Caudate, putamen and capsular regions are known to receive afferents from the left pre-SMA region, which is involved in vWM (Inase et al., 1999). Crosson et al. (2003) found that basal ganglia activity was accompanied by activation of the left pre-SMA during a word production task. They hypothesized that the increase of right basal ganglia activation serves to suppress the non-dominant right frontal cortex, whereas the increase of the left basal ganglia activation serves to enhance the language processing of the left dominant hemisphere. Against the background of these findings, the basal ganglia can be assumed to interact closely with the frontal cortex and to serve as a selective gating mechanism for the prefrontal cortex (Frank et al., 2001). From this perspective, the findings showing the basal ganglia to be active only during encoding and maintenance phases seem plausible, because selective gating plays a major role for these processes. However, as we did not study the phases separately, we cannot rule out that these activations also reflect attentional processes in addition to pure vWM processes. Moreover, six publications compared activation conditions with a simple baseline (e.g., fixation of a cross hair). Hence, we cannot rule out that some parts of the subcortical activation were due to motor activity (i.e., button press in the activation conditions vs. no button press in the baseline condition). In the present meta-analysis we also found left pre-SMA activation, but a conclusion about their influence on basal ganglia is unwarranted without any connectivity data. In addition to the pre-SMA activation, our meta-analysis demonstrated significant activation in the anterior cingulate which has been found to be activated during vWM tasks before (Bedwell et al., 2005; Narayanan et al., 2005). It should be noted that in the AAL atlas the significant cluster was labeled as median cingulate, which is part of the anterior cingulate. It is striking, however, that a majority of vWM did not find an involvement of the anterior cingulate. Hence, future studies should make an attempt to clarify the specific contribution of the different parts of the cingulate to vWM.
Age, Load, and Mean RT as Influencing Factors
Age-related changes in vWM are not fully understood because of a lack of longitudinal data. A recent longitudinal study found the activation of left prefrontal cortex (i.e., MFG and parts of the IFG) to be reduced during a vWM manipulation task in older people (Rieckmann et al., 2017). Somewhat in accordance with this finding, the present meta-analysis demonstrated a negative association between activation in the left and right IFG—including Broca's area—and age. In addition, we found a negative association between activation in the right MFG and age. It is known that cortical thickness, surface area, and volume of this region decrease with age (Lemaitre et al., 2012) which may, to some degree, explain this finding. Moreover, the right MFG plays a central role in reorienting attention from exogenous to endogenous attentional control (Japee et al., 2015). Our results are in agreement with the ontogenetic model of brain development according to which those brain regions that are the last to mature are the first to be affected by aging (Raz et al., 2005). Of note, all the other regions exhibiting a negative association between activation and age were localized in the left hemisphere. We found this negative association in the left insula, which—as stated above—plays a relevant role in the context of rehearsal, in the left putamen, which is involved in the active filtering of irrelevant material allowing us to focus on relevant material (Moore et al., 2013) in the left rolandic operculum, important for overt rehearsal, and in the left superior temporal gyrus, parts of which are critically involved in phonological storage. The fact that these regions important for different vWM processes showed a negative association with age might explain why older people tend to exhibit worse vWM performance, although it should be kept in mind that we did not take into account any longitudinal data or individual subject performance. Since these associations were detectable mainly in the left hemisphere and age-related changes were not restricted to the right PFC, our results seem to speak against the HAROLD model (Cabeza et al., 2002). Overall, the present results do not provide any evidence for a decrease of this lateralization with age, as claimed in other studies (Reuter-Lorenz et al., 2000; Cabeza et al., 2002, 2004). The fact that we found a decreased frontal activation with increasing age could either mean that the brain is not as adaptive as proposed by the earlier discussed STAC model or indicate that task demands were too high for elderly people leading to a “breakdown” of frontal activation instead of a compensatory increase. In order to draw any further conclusion it would be helpful to study the activation of these regions in elderly people taking also into account their individual task performance (e.g., accuracy). However, only three studies selected for the meta-analysis specifically included older populations; thus, the age-range was clearly undersampled in the current meta-analysis and the power to reliably assess the influence of age was too low. This might also explain why our results do not seem to be in line with the HAROLD model. Therefore, more empirical data comparing older and younger populations are necessary in order to find out more about specific age differences in activation during vWM. A better understanding of these age-related differences would pave the way for creating more sophisticated methods to preserve or enhance cognitive function in elderly populations.
Höller-Wallscheid et al. (2017) hypothesized the decreased lateralization across the PFC to be independent of age, but to depend on the subjective difficulty of WM tasks. In line with this hypothesis we found a bilateral activation across the PFC in the load effect meta-analysis. Our results are also in accordance with the load effect meta-analysis performed by Rottschy et al. (2012). The CRUNCH model states that the extent of cortical activation depends on the task load. Our results support this model, since we found a positive correlation between activation in several cortical regions (e.g., frontal areas) and task load. Apart from PFC areas we also found activation in the parietal cortex (IPG and left SPG) to be influenced by load, as reported in a previous study by Braver et al. (1997). Likewise, activation of the right lobule VI and crus I of the cerebellum turned out to depend on the difficulty of the vWM tasks. This is in accordance with a previous study which showed these parts of the cerebellum to respond to changes in vWM load (Kirschen et al., 2005). As stated before, there are anatomical connections between these parts of the cerebellum and frontal areas. Hence, the increased input from frontal regions involved in the articulatory system during a load manipulation could also reflect the increased activation of the right cerebellum. In addition, we found an association between load and activation in the fusiform gyrus. Tsapkini and Rapp (2010) pointed out that lesions of the left fusiform gyrus were significantly associated with reading and spelling deficits. In light of this finding, the positive correlation between load and activation in the left fusiform gyrus in the present study might indicate that a majority of people may have used overt rehearsal as a strategy to cope with increasing task difficulty.
It has long been recognized that RT is sensitive to manipulations of any kind of WM load (Just and Carpenter, 1992). Therefore, RT can also be viewed as a measure of load. A previous study found a significant positive correlation between RT and fMRI signal in nine subjects in the MFG and the left IFG (Braver et al., 1997). The present meta-analysis partially corroborates these findings showing both the left MFG and the left precentral gyrus to be positively associated with RT. However, our results seem to contradict a study by Honey et al. (2000), which reported that posterior parietal cortical activation was predicted by a prolonged RT in a vWM task. Importantly though, as states earlier, activation in the parietal cortex is influenced by load, which—in turn is related to RT. Moreover, the present meta-analysis revealed a positive association between left precentral gyrus activation and RT. This finding is plausible considering that the left precentral gyrus constitutes a major part of the primary motor area and its activation is contralateral to the side of the hand movement. Hence, increased activation in the primary motor cortex might facilitate faster responding. It should be emphasized that RT information was not available for all studies. Moreover, RT depends on many other factors such as number of responses alternatives, type of discrimination or delay time. Therefore, results of this factor should be treated with caution.
Although we performed a meta-analysis in which only fMRI studies were included, there has been a previous meta-analysis in which they selected both fMRI and PET studies to isolate the neural correlates of human working memory (Wager and Smith, 2003). Although they found some support for left frontal cortex dominance in vWM tasks, this was only for tasks with low executive demand. These results support our finding regarding the lateralization, i.e., the higher the difficulty on the task, the less lateralization of PFC activation is to be expected.
Finally, we expected the type of fMRI paradigm to be a significant moderator as demonstrated in a previous meta-analysis (Rottschy et al., 2012). This expectation was not met by the data. This could be due to a strong overlap in task activation, with potentially existing subtle quantity differences being too weak to be significant. We also found that gender did not affect activation associated with vWM tasks. However, we cannot exclude that gender differences would emerge when controlling for effects of sex hormones. Hence, future studies are required that carefully consider these potentially confounding factors.
Limitations
In the present meta-analysis we discussed a number of relevant networks based on fMRI activity, such as the attentional system, but we did not take into account brain connectivity. However, the localization of brain areas is just the first step toward a more comprehensive understanding of the neural correlates of vWM. Analyses based on temporal dynamics, such as EEG or single-unit recordings, are essential to build a more integrative view. Another limitation regards our cerebellum findings. There is strong reason to assume that we did not find any inferior cerebellum activation because some of the scans included in the present meta-analysis did not cover the whole cerebellum due to methodological limitations (e.g., trade-off between brain coverage and repetition time).
Conclusions
We used a coordinate-based meta-analysis to integrate the current literature on vWM in healthy humans. We found activation of the established fronto-parietal network and the right cerebellum, especially crus I, and lobule VI. Our results support the dual-selection model, according to which a mid-lateral PFC activation occurs due to verbal input. Moreover, our results illustrate that we should not underestimate the activation of subcortical regions that play an important role for response inhibition. Age, mean RT, and load moderate vWM task activation and, thus, should be taken into consideration in future research. Especially the influencing factor of age should be further analyzed since the sample included in the present meta-analysis consists of primarily young people. Mean reaction time, moreover, could be influenced by many other factors. Further, more fine-grained studies are needed to gain a better understanding of the neural correlates underlying processes involved in vWM including encoding, maintenance, and retrieval.
이 글은 위 출처의 원문을 그 라이선스(CC BY 4.0)에 따라 옮겼습니다. 참고문헌 목록과 인용 번호, 그림 · 표 · 각주 · 수식 · 부록은 빼고 절별로 나누어 실었으며, 그 밖의 문장은 원문 그대로입니다. 원 저작자와 발행 기관은 이 화면을 만들거나 검토하지 않았고, 이곳의 해석과 낱말 풀이는 공식 번역이 아닙니다.
