본문 바로가기
읽기 · 학술 · 공개 라이선스 원문

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

장 3개 올림 이 장 1549낱말
← 앞 장 다음 장 →

이 장은 원문만 있습니다. 문장별 해석은 앞 장부터 차례로 붙입니다.

2장 · Neural Correlates of Verbal Working Memory: An fMRI Meta-Analysis — Materials and Methods

Literature Search and Inclusion of Studies

An exhaustive literature search was conducted on whole-brain fMRI studies on vWM from January 2000 to December 2017. We searched the databases PubMed, Scopus and Web of Science for English-language studies with the combination of the following key words: “n-back,” “DMTS,” “Sternberg,” ”delayed matched to sample,” “delayed match to sample,” plus “verbal working memory,” “fMRI,” “healthy.” The Brainmap database was also searched with their respective search criteria (Subjects Size is more than 10, Experiments Paradigm Class is Delayed Match to Sample/n-back, Experiments Imaging Modality is fMRI, Conditions Stimulus is Visual Letters, and Subjects Handedness is Right). Further studies (11 publications) were identified through chasing citations from the selected studies (see Figure 1 for flowchart diagram). The “Meta-analysis of Observational Studies in Epidemiology” (MOOSE) guidelines Stroup et al. (2000) were used for the literature search and selection of studies. All articles were identified, selected and coded by a single investigator (M.E.). The same investigator double-checked the manually extracted peak coordinates and effect size values from the selected studies.

The criteria for inclusion were whole-brain analyses with reported results in a standard reference space (Talairach or MNI), inclusion of more than 10 healthy subjects and studies with clear boundaries between inclusion and exclusion criteria. Studies were excluded if they only included region-of-interests (ROI) analyses, did not report peak coordinates, investigated between- or within-group effects of pharmacological treatment, disease, living conditions, or used reward trials or emotional retrieval. We also excluded studies that re-analyzed previously reported data to avoid overestimating the effects. Positron emission tomography (PET) experiments were also not included in this meta-analysis due to methodological differences (e.g., differences in temporal resolution between PET and MRI) and differences in the underlying physiology (i.e., BOLD contrast vs. glucose mechanism).

Comprehensive Meta-Analysis

We first conducted a meta-analysis of all the vWM studies. The demographic and study characteristics are shown in Table 1. The vast majority of the selected studies used the SPM software (https://www.fil.ion.ucl.ac.uk/spm/software/) to perform their fMRI analyses (83.3% of studies) indicating a clear bias toward this software package. Coordinates and t-values included in the analysis are shown in Supplementary Table 1. When only p- or z-values were reported, they were transformed into t-values taking into account the sample size per study. The influence of gender (% female), mean age, type of fMRI task (DMTS including the Sternberg task or n-back), mean RT, and the type of threshold used in the study (uncorrected vs. corrected) were studied with meta-regressions. The majority of studies corrected for multiple comparisons by controlling the false-discovery rate (FDR), except for one study that used the family-wise error rate (FWE) and another one that used Bonferroni correction. Those studies presenting results with peak coordinates at p < 0.005 or p < 0.001 uncorrected (33.3%) controlled for the cluster-size with different thresholds (5, 8, 10, 17, or 25 contiguous voxels). On a side note—although IQ and years of education have been shown to be associated with WM performance (Fukuda et al., 2010; Boller et al., 2017), we could not assess these factors here because the majority of studies did not provide any information on IQ or years of education.

There are several established fMRI vWM paradigms: n-back, Sternberg, and delayed matching to sample (DMTS) tasks (Kirchner, 1958; Sternberg, 1966; Paule et al., 1998). N-back tasks include a sequential presentation of stimuli. Subjects have to decide whether the current stimulus is the same as the one n positions before (e.g., the previous one in a 1-back condition or the one two positions back in a 2-back condition). In Sternberg tasks, a set of stimuli is presented simultaneously that need to be maintained over a certain period which is followed by a single probe stimulus for which participants need to decide whether it was part of the set or not. In DMTS tasks, a single stimulus is presented. After the maintenance period, a set of multiple probes is presented from which participants need to recognize the single stimulus they had to memorize. While n-back tasks are normally presented in the form of a block-design, DMTS and Sternberg tasks are presented in an event-related design.

Load-Effect Meta-Analysis

To assess the neural correlates of increasing vWM load (i.e., the difficulty of the fMRI task), we performed a load-effect meta-analysis. We only included studies in which there was a contrast between higher and lower vWM loads, such as 3-back vs. 1-back or 3-back vs. 2-back. The selected studies are shown in Supplementary Table 2.

Meta-Analytical Approach: ES-SDM

We used the anisotropic effect-size version of seed-based d' mapping software (http://www.sdmproject.com) to conduct coordinate-based meta-analyses. The software uses a voxel-based meta-analytic approach. First, a strict selection of the reported peak coordinates of gray matter differences was applied by only including the studies containing whole-brain analyses. This is essential in order to avoid biased results from some neuroimaging studies, in which more liberal statistical thresholds were used for some ROIs relative to the rest of the brain. Peak coordinates in MNI or Talairach and effect size values were manually extracted from each contrast of interest in each study. All p- or z-values were transformed into t-values using SDM web utilities. Second, a map for the activation in gray matter was created for each study using the Automated Anatomical Labeling (AAL) atlas partitioned into 116 brain regions (Tzourio-Mazoyer et al., 2002). If a study included more than one contrast of interest, we adjusted for multiple contrasts by combining the created images of each contrast into one image for the final analyses. The ES-SDM software re-creates the maps from the studies by converting the t-value of each peak to Hedge's g (Alegria et al., 2016). Third, an anisotropic non-normalized Gaussian kernel was applied by assigning different values to the different neighboring voxels based on the spatial correlation between them (Radua et al., 2014). At the end, we obtained a mean map by a voxelwise calculation of the mean of the study maps, weighted by the square root of the sample size, so that studies with larger sample sizes contributed more strongly (Radua and Mataix-Cols, 2009).

To assess the robustness of the main findings, we performed a whole-brain Jackknife analysis. Jackknife analysis consists of repeating the statistical analyses several times by discarding one study each time thus demonstrating the stability of the results (Müller et al., 2018). Heterogeneity of effect sizes and publication bias were assessed with the I² index and Egger's test (Egger et al., 1997; Müller et al., 2018). The I² index provides the proportion of variability across studies that is due to true heterogeneity relative to that from sampling error (Higgins and Thompson, 2002). Egger's tests were used to test for asymmetry of funnel plots, serving as an indicator of publication bias (see Supplementary Figure 1 for examples).

Statistical significance was determined with random-effects models. We used the default threshold for the calculated mean (voxel-level p < 0.005 uncorrected, peak height threshold 1, minimum cluster extent 10 contiguous voxels) (Radua and Mataix-Cols, 2009). To control for multiple testing in the several meta-regressions we used a more conservative threshold, Bonferroni-corrected threshold of p < 0.001.

Results

Comprehensive Meta-Analysis (42 Studies)

The mean map of brain regions of the whole-brain meta-analysis for vWM is shown in Figure 2. The majority of studies reported only task-positive activation. We observed extended activation patterns in the frontal lobe including left superior frontal gyrus (SFG), medial frontal gyrus, right middle frontal gyrus (MFG), right inferior frontal gyrus (IFG), triangular, orbital and opercular part of the right IFG, orbital and opercular part of the left IFG, bilateral SMA, bilateral precentral gyrus, and left rolandic operculum. There was also activation in parietal areas including left post-central gyrus, right angular gyrus, and left inferior parietal gyri (IPG). Moreover, there was activation in the bilateral median cingulate, the left insula, the right lenticular nucleus (i.e., putamen and pallidum) and in bilateral cerebellum (crus I).

Robustness analyses showed that these results were preserved in all studies. Egger's tests indicated that there were some regions for which there was evidence of heterogeneity: left SFG, left SMA, left precentral gyrus, left post-central gyrus, right angular gyrus, left IPG, right median cingulate, left insula, and right cerebellum (crus I) (see Table 2).

Meta-regression analyses confirmed that mean age and mean RT moderated activation in some brain regions. Mean age was associated with decreased activation in the left rolandic operculum, left insula, left superior temporal gyrus (STG), left IFG (opercular part), left heschl gyrus, left post-central gyrus, left lenticular nucleus (putamen), and the right MFG. Mean RT was positively associated with activation in the left precentral gyrus and the left MFG (see Figure 3 and Table 3). None of the other meta-regression analyses yielded any significant results.

Load-Effect Meta-Analysis (16 Studies)

We found activation in several frontal areas: right SFG (dorsolateral and medial part), left SFG (medial part), right MFG, right IFG (triangular part), left IFG (triangular and opercular part), right SMA, bilateral precentral gyrus, bilateral rolandic operculum. Moreover, there was activation in several parietal areas (left post-central gyrus, left angular gyrus, left SPG, and bilateral IPG) as well as in the left anterior cingulate gyri, bilateral median cingulate gyri, left fusiform gyrus, and right cerebellum (crus I and hemispheric lobule VI) (see Table 4).

Jackknife analyses showed that the findings were preserved across studies, except for the right rolandic operculum and left fusiform gyrus, which were no longer detectable after discarding two papers. We only observed heterogeneity in the right inferior parietal gyrus (see Table 4).

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.