Dynamic behavior of the locus coeruleus during arousal-related memory processing in a multi-modal 7T fMRI paradigm
- Gordon Center for Medical Imaging, Department of Radiology, Massachusetts General Hospital/Harvard Medical School, United States
- Faculty of Health, Medicine and Life Sciences, School for Mental Health and Neuroscience, Alzheimer Centre Limburg, Maastricht University, Netherlands
- Faculty of Psychology and Neuroscience, Department of Cognitive Neuroscience, Maastricht University, Netherlands
- Center for Neuroscience Imaging Research, Institute for Basic Science and Department of Biomedical Engineering, Sungkyunkwan University, Republic of Korea
- Techna Institute & Koerner Scientist in MR Imaging, University Health Network, Canada
Figures
Figure 1 with 2 supplements
Experimental design.
(A) Task procedure. The salivettes indicate the time points of collection of saliva samples. For the 7T fMRI part, anatomical scans and a baseline 6 min resting state with a fixation cross were collected first. Then a memory task was completed, consisting of encoding, consolidation (a resting-state with a fixation cross) and recollection. (B) Encoding consisted of the presentation of 90 face-name pairs (45 emotional- 45 neutral valence; permission of use obtained from PERT96 group). Recollection consisted of 135 trials (45 new faces (23 with emotional valence)) for which participants had to decide if the face was presented before. Upon endorsement, they were asked to indicate the name that was presented with the face during encoding. Throughout the fMRI experiment, breathing and pulse rate data were collected.
Figure 1—figure supplement 1
Sagittal and axial slices of the T1-weighted and fMRI average in the MNI space at the level of the hippocampus and brainstem.
Note: The first column shows the matched MNI template. The middle column shows the average of individual T1-weighted scans warped to the MNI space. The right column shows the average of the individual fMRI scans warped to the MNI space. Note that for both averages relevant features such as the dentate gyrus, the substantia nigra and much of the internal structure of the brainstem (of a similar size to the locus coeruleus) can be clearly differentiated (red arrows), attesting to the quality of the registrations. See also the response letter to the reviewers for individual registration examples.
Figure 1—figure supplement 2
Comparison of the temporal signal-to-noise ratio (tSNR) per participant across fMRI pipelines.
Note: Compared to the ‘spatially normalized’ pipeline, the ‘FIX’ pipeline provided a sizable gain in tSNR, while including explicit physiological time series provided a modest tSNR increase (median ‘FIXed’=39.78 (IQR = 35.862, 42.88), median ‘FIXed + explicit Resp’=44.74 (IQR = 39.893, 51.931), median ‘FIXed + explicit Phys’=46.39 (IQR = 41.402, 54.159), median ‘Spatially normalized’=23.48 (IQR = 21.657, 26.156)), but at the cost of removing 22 statistical degrees of freedom This is also reflected in the intraclass correlation coefficient between the pipelines: spatially normalized versus FIXED: ICC = 0.443; FIXED versus FIXED + explicit Resp: ICC = 0.96; and FIXED + explicit Resp versus FIXED + explicit Phys: ICC = 1.00.
These ICC values show a high level of similarity between the tSNR of the FIXED, FIXED + explicit Resp and FIXED + explicit Phys pipelines.
Figure 2 with 1 supplement
Behavioral performance (N = 27).
(A) Hit rate for emotional and neutral conditions during recognition (paired Wilcoxon test: Z = −1.17, p=0.244). (B) False alarm rate for emotional and neutral conditions during recognition (paired Wilcoxon test = 3.81, p<0.001). (C) Recognition bias for emotional and neutral conditions (paired t-test, p=0.019). (D) Uncorrected recollection rate for emotional and neutral conditions (p=0.95). (E) Emotional condition: scatter-plot between recollection rate and bias c (robust linear regression: β = −30.99,t = −9.77, p<0.001). (F) Neutral condition: scatter-plot between recollection rate and bias c (robust linear regression: β = −23.78, t = −6.35, p<0.001). (G) Corrected recollection rate for emotional and neutral conditions (paired Wilcoxon test: p<0.001). Boxplots inside the violinplots show median, quartiles (boxes) and range (whiskers). Shaded regions depicts the 95% confidence interval.
-
Figure 2—source data 1
Behavioral performance on the face-name association task.
Individual hit rates, false alarms, bias scores for the recognition, recollection scores (raw and bias-adjusted) are provided for the neutral and emotional valence condition.
- https://cdn.elifesciences.org/articles/52059/elife-52059-fig2-data1-v2.rdata
Figure 2—figure supplement 1
Schematic representation of the procedure used to calculate bias-free emotional recollection scores.
Note: Recognition – and hence the recognition bias - was based on the correct or incorrect recognition of faces. Recollection involved the identification of the name belonging to the face (see also Figure 1). The work flow depicted here visualizes the calculation of the normalized emotional recollection scores used in the manuscript. Terms with ‘emotional, neutral’ in subscript refer to a separate calculation for emotional and neutral with the provided formula.
Figure 3
Associations between autonomic tone measures and memory performance across the task stages.
(A) sAA change during the task (encoding, consolidation and recollection) relative to the baseline resting-state. A significant increase in sAA was observed between consolidation and baseline (linear mixed effects model, p=0.033, N = 21, 68 observations). (B) Association between ΔsAA (consolidation – baseline) and emotional memory performance (correlation, r = 0.452; p=0.059, N = 18). (C) Distributions of the rMSSD during the task (encoding, consolidation and recollection) relative to the baseline resting-state (no significant differences, linear mixed effects model, p=0.23, N = 18, 63 observations). (D) Repeated measures correlation plot between ΔsAA and rMSSD across all task stages (Rrm = −0.387; p=0.018, N = 17, 56 observations). Boxplots inside the violinplots show median, quartiles (boxes) and range (whiskers). Shaded region in the scatterplot depicts the 95% confidence interval.
-
Figure 3—source data 1
Autonomic tone and memory performance across the entire paradigm.
Individual ΔsAA, rMSSD values per task stage. The emotional memory scores provided for panel B show the proportion of adjusted emotional scores versus neutral scores as indicated in the methods.
- https://cdn.elifesciences.org/articles/52059/elife-52059-fig3-data1-v2.rdata
Figure 4 with 2 supplements
Associations between LC BOLD and autonomic tone measures using repeated measures correlations.
(A–B) LC component visualization. (A) Group average t-map from the dual-regression showing a robust bilateral LC component, thresholded at p<0.01 FWE, after denoising with FIX (N = 24). (B) LC template from all scans. A hyperintensity can be observed close to the 4th ventricle. (C–F) Repeated correlations between LC variability rMSSD (p=0.008, N = 23, 92 observations) or ΔsAA (not significant, p=0.971, N = 17, 56 observations) (C,E). Similar associations are shown for variability in the reference component (D,F) (no significant relationships, p=0.109 and p=0.942, respectively, N = 17, 56 observations). Each marker signifies a participant and distinct shapes show the different task stages.
-
Figure 4—source data 1
Locus coeruleus BOLD and autonomic tone measures across the entire paradigm.
Individual ΔsAA, rMSSD, LC and reference BOLD values per task stage. Individual coherence values between.
- https://cdn.elifesciences.org/articles/52059/elife-52059-fig4-data1-v2.rdata
Figure 4—figure supplement 1
Functional template of the locus coeruleus during resting-state.
Note: LC template made from all LC scans. (A) Hyperintense signal can be seen close to the 4thventricle. (B-E) Group average t-maps from dual-regression of LC component, thresholded at significance level, across preprocessing steps. (B) Spatially normalized (after motion, distortion and slice-timing correction) (C) with FIXed applied (D) with FIXed +Explicit Resp and (E) with FIXed +Explicit Phys.
Figure 4—figure supplement 2
Relationship between LC variability and rMSSD across the task stages for each preprocessing pipeline.
Note: (A) spatially normalized pipeline (not significant, p-value=0.312, CI (95%) = [−0.175, 0.483]). (B) FIXed + explicit Resp pipeline (trend-level negative association, Rrm = −0.29, p=0.07, CI (95%) = [−0.558, 0.033]). (C) FIXed + explicit Phys pipeline (significant, Rrm = −0.411, p=0.009, CI (95%) = [−0.649,–0.1]).
Figure 5 with 3 supplements
Coherence between LC and HRV across the task stages.
(A) LC coherence magnitude squared plotted against frequency for each task stage for a single participant. The high-frequency HRV band is depicted in a lighter shade. (B) Reference ROI-HRV coherence magnitude squared for the same participant. (C) The association between frequency and the predicted magnitude squared coherence between the LC and HRV across the various task stages (based on linear mixed effects models, N = 24 (7920 observations), p-values for all comparisons are provided in Supplementary file 7). Shaded regions in C and D show the 95% confidence intervals. (E) Repeated correlations between LC-HRV coherence and ΔsAA (Rrm(df)=−0.383 (34), p-value=0.021). (F) Reference ROI-HRV coherence and sAA change levels (not significant, Rrm(df)=0.218 (34), p-value=0.201).
-
Figure 5—source data 1
Coherence between locus coeruleus BOLD and HRV.
Individual coherence values between locus coeruleus or reference BOLD and HRV per frequency are provided for each task stage in the paradigm. For panel E and F, the individual ΔsAA and median of maximum coherence between locus coeruleus or reference BOLD and HRV are provided.
- https://cdn.elifesciences.org/articles/52059/elife-52059-fig5-data1-v2.rdata
Figure 5—figure supplement 1
Coherence between LC and HRV across the task stages for the spatially normalized pipeline.
Note: Coherence between LC and HRV across the task stages. (A) LC coherence magnitude squared plotted against frequency for each task stage for a single participant. The high-frequency HRV band is depicted in a lighter shade. (B) Reference ROI-HRV coherence magnitude squared for the same participant. (C) The association between frequency and the predicted magnitude squared coherence between the LC and HRV across the various task stages (based on linear mixed effects models,, N = 24 (7920 observations), p-values for all comparisons are provided in Supplementary file 4). Shaded regions in C and D show the 95% confidence intervals. (E) Repeated correlations between LC-HRV coherence and ΔsAA (Rrm(df)=−0.36 (34), p-value=0.031, CI (95%) = [−0.62,–0.03], N = 17, 56 observations). (F) Reference ROI-HRV coherence and sAA change levels (not significant, Rrm(df)=−0.05 (34), p-value=0.773, CI (95%) = [−0.38,–0.29], N = 17, 56 observations).
Figure 5—figure supplement 2
Coherence between LC and HRV across the task stages for the FIXed pipeline Note: Coherence between LC and HRV across the task stages.
(A) LC coherence magnitude squared plotted against frequency for each task stage for a single participant. The high-frequency HRV band is depicted in a lighter shade. (B) Reference ROI-HRV coherence magnitude squared for the same participant. (C) The association between frequency and the predicted magnitude squared coherence between the LC and HRV across the various task stages (based on linear mixed effects models, N = 24 (7920 observations), p-values for all comparisons are provided in Supplementary file 5). Shaded regions in C and D show the 95% confidence intervals. (E) Repeated correlations between LC-HRV coherence and ΔsAA(Rrm(df)=−0.383 (34), p-value=0.021, CI (95%)=[−0.615,–0.013], N = 17, 56 observations). (F) Repeated correlations between the reference ROI-HRV coherence and ΔsAA (not significant, Rrm(df)=0.208 (34), p-value=0.223, CI (95%)=[−0.14, 0.51], N = 17, 56 observations).
Figure 5—figure supplement 3
Coherence between LC and HRV across the task stages for the FIXed + Explicit Resp pipeline.
Note: Coherence between LC and HRV across the task stages. (A) LC coherence magnitude squared plotted against frequency for each task stage for a single participant. The high-frequency HRV band is depicted in a lighter shade. (B) Reference ROI-HRV coherence magnitude squared for the same participant. (C) The association between frequency and the predicted magnitude squared coherence between the LC and HRV across the various task stages (based on linear mixed effects models,, N = 24 (7920 observations), p-values for all comparisons are provided in Supplementary file 6). Shaded regions in C and D show the 95% confidence intervals. (E) Repeated correlations between LC-HRV coherence and ΔsAA (Rrm(df)=−0.297 (34), p-value=0.078, CI (95%) = [−0.58,0.046], N = 17, 56 observations). (F) Reference ROI-HRV coherence and sAA change levels (not significant, Rrm(df)=0.199 (34), p-value=0.246, CI (95%) = [−0.14,–0.5], N = 17, 56 observations).
Figure 6 with 3 supplements
LC and MTL activity during the task stages (N = 24).
(A) Bilateral hippocampus, bilateral amygdala and entorhinal cortex activation during encoding (Encoding >Not Encoding; thresholded at p<0.05, uncorrected). (B–E) LC activation during encoding (z-values thresholded at p<0.05, uncorrected). (B) LC activation during successfully encoded events with high arousal (HRV minima) (High Arousal successful encoding >Low Arousal successful encoding). (C) LC activation during High Arousal Encoding >Low Arousal Encoding and during D) High Arousal successful Encoding >High Arousal forgotten. (E) Activation of LC for emotionally high arousal encoding events (High Arousal Emotional Encoding >High Arousal Neutral Encoding). (F) Bilateral CA1 activation during recollection of correctly encoded trials (Successful recollection >Forgotten recollection) (z-values thresholded at p<0.05, uncorrected). Boxplots inside the violinplots show median, quartiles (boxes) and range (whiskers). The layer containing white voxels demonstrates the location of the Keren template as reference (Keren et al., 2009). Individual brainstem results for the encoding contrasts are shown in Figure 6—figure supplement 2.
-
Figure 6—source data 1
Locus coeruleus and medial temporal lobe activity during the task stages.
Individual beta’s are provided for the amygdala, hippocampus and locus coeruleus for the examined relevant contrasts during encoding.
- https://cdn.elifesciences.org/articles/52059/elife-52059-fig6-data1-v2.rdata
Figure 6—figure supplement 1
Mask of regions used in the GLM and functional connectivity analyses.
Note: Sagittal, coronal and axial slices of the mask that was used for volume correction for the functional connectivity and general linear model analysis (white; overlaid over the MNI 1 mm template).
Figure 6—figure supplement 2
Individual brainstem results for the GLM contrasts during encoding.
Note: Individual brainstem results for the GLM contrasts during encoding. Each column shows the brainstem activity at p<0.05 (uncorrected) for several contrasts during encoding and for five randomly selected subjects.
Figure 6—figure supplement 3
Encoding-related brain activation for recognition contrasts.
Note: (A) Right middle temporal gyrus activation during encoding (Emotional >Neutral, threshold at p<0.05, uncorrected). (B) Right temporal fusiform cortex activation during encoding (Successful >Not successful recognition, threshold at p<0.05, uncorrected). (C) Right inferior frontal gyrus activation during encoding (Emotional successful recognition >neutral successful recognition, threshold at p<0.05, uncorrected). Significant activation following cluster correction was detected for the (Emotional >Neutral) contrast in the right middle temporal gyrus (peak z-value = 3.77, MNI coordinates [68,-35,-5], cluster volume = 157 mm). For the (Successful recognition >Not successful recognition) contrast significant activation was found in the right temporal fusiform cortex (peak z-value = 4.13, MNI coordinates [39,-35,-21], cluster volume = 101 mm). For the interaction (Emotional successful recognition >neutral successful recognition) a significant cluster was detected in the right inferior frontal gyrus (peak z-value = 4.08, MNI coordinates [Ekman, 1992; Lane et al., 2009; Wagatsuma et al., 2018], cluster volume = 191 mm). No clusters were visible in the amygdala, hippocampus, entorhinal cortex or locus coeruleus for the above contrasts.
Figure 7 with 1 supplement
Integration of LC and MTL timeseries across the task (N = 24).
(A) Functional connectivity (FC) from LC component to MTL structures across task stages. Clusters of decreased FC in hippocampus, amygdala and entorhinal cortex were observed comparing consolidation to baseline (p<0.05, uncorrected). (B-E) Median maximum coherence and phase across task stages between timeseries from the LC and the clusters from the hippocampus, amygdala and entorhinal cortex from the FC analysis (B: baseline resting-state; C: encoding; D: consolidation; E: recollection). Confidence intervals of maximum coherence were calculated per timeseries and the median maximum coherence and phase were extracted. The significant partial coherence (regressing out timeseries from other regions) is visualized. The thickness of the edges indicates the strength of coherence. The absolute phase difference between timeseries is shown in transparencies. Arrows indicate the direction of the phase lag from leading to lagging.
-
Figure 7—source data 1
Coherence and phase between time series of the locus coeruleus and medial temporal lobe clusters.
Individual coherence and phase values between the locus coeruleus and each cluster in the medial temporal lobe (amygdala, hippocampus, entorhinal cortex) are provided per task stage.
- https://cdn.elifesciences.org/articles/52059/elife-52059-fig7-data1-v2.rdata
Figure 7—figure supplement 1
Functional connectivity from LC to MTL during consolidation compared to baseline across preprocessing steps.
Note: Functional connectivity (FC) from LC component to medial temporal lobe regions during consolidation compared to baseline across preprocessing steps (t-values threshold at p<0.05, uncorrected for multiple comparisons). Significant clusters of lower FC between the LC and hippocampus and amygdala during consolidation as compared to baseline were observed across preprocessing steps except for the spatially normalized pipeline.
Figure 8 with 1 supplement
Structural segmentation of the LC and associations with emotional memory performance (N = 24).
(A) 3D visualization of the LC placement in the pons. (B) The LC template, coronal view (hyperintense rod-like shape). (C) Segmentation of the LC in template (green). (D) TFL mean intensity normalized with a pontine reference region of the entire LC correlated with emotional memory recollection rate (r = 0.562; p-value=0.004) E) The LC was divided into three equisized subsegments. (F–H) LC segments intensity correlation with emotional memory recollection rate: (F) caudal (r = 0.523; p-value=0.034) (G) medial (r = 0.523; p-value=0.009) and (H) LC rostral (r = 0.523; p-value=0.074). Shaded regions depict the 95% confidence interval.
-
Figure 8—source data 1
Structural locus coeruleus intensity and memory performance.
Individual locus coeruleus intensity values (entire locus coeruleus, rostral, middle and caudal) and emotional memory scores are provided. The emotional memory scores are calculated as the proportion of adjusted emotional scores versus neutral scores as indicated in the methods.
- https://cdn.elifesciences.org/articles/52059/elife-52059-fig8-data1-v2.rdata
Figure 8—figure supplement 1
Template of the locus coeruleus for structural analyses.
Note: Axial slices of the high-resolution Magnetization-Transfer weighted template (2 mm apart) that was used for structural analysis. Zoomed-in slices at LC level are shown in the middle row. In the last row, a coronal slice of the template is shown at LC level.
Author response image 1
Mid-sagittal, coronal and axial slices in the MNI-space for every T1-weighted (first 3 columns) and fMRI-dataset (last 3 columns).
Every row represents a participant. The first row shows the relevant MNI slices. Also see Author responses image 2-4.
Author response image 2
Author response image 3
Author response image 4
Author response image 5
Overlap ratio of individual ROI (Freesurfer-segmented hippocampus and brainstem) versus the matched ROI across the group.
The distribution for each individual is shown in the x-axis.
Tables
Table 1
Behavioral results of the memory task for emotional and neutral valence.
Note: numbers provided are median and interquartile range (IQR). Differences between emotional and neutral stimuli were tested with a paired Wilcoxon test. All scores, except for the bias express %.
| Emotional (Median (IQR)) | Neutral (Median (IQR)) | Z | p-value | |
|---|---|---|---|---|
| Recognition Hit Rate (Old faces) | 53.33 (45.56, 61.11) | 53.33 (44.44, 67.78) | −1.17 | 0.244 |
| Recognition Miss Rate (Old faces) | 46.67 (38.89, 54.44) | 46.67 (32.22, 55.56) | 1.2 | 0.229 |
| Recognition Correct Rejection Rate (New faces) | 77.27 (65.91, 81.82) | 86.36 (77.27, 90.91) | −3.77 | <0.001 |
| Recognition False Alarm Rate (New faces) | 22.73 (18.18, 34.09) | 13.64 (9.09, 22.73) | 3.81 | <0.001 |
| Recognition Hit Rate - Recognition False Alarm Rate | 23.94 (17.02, 32.83) | 37.68 (28.23, 50.81) | −3.16 | <0.001 |
| Recognition Response Bias | 0.90 (0.81, 0.92) | 0.84 (0.74, 0.9) | 2.29 | 0.02 |
| Recollection Hit Rate Raw | 24.44 (18.89, 31.11) | 26.67 (18.89, 33.33) | 0.35 | 0.726 |
| Recollection Hit Rate (correct recognition) | 50 (43.43, 54.01) | 44.83 (37.97, 58.77) | 0.06 | 0.953 |
| Recollection Hit Rate (Old faces) - Recollection False Alarm Rate (Old faces) | 0.00 (−13.14–8.01) | −10.34 (−24.07–17.54) | 0.06 | 0.953 |
| Bias-corrected Recollection Hit Rate | 38.68 (36.68, 39.46) | 29.97 (28.01, 33.39) | 4.41 | <0.001 |
Additional files
-
Supplementary file 1
Demographics and neuropsychological performance of the group.
- https://cdn.elifesciences.org/articles/52059/elife-52059-supp1-v2.docx
-
Supplementary file 2
Available data per modality and time point.
Note: numbers provided are N and % compared to the total sample size (N = 27); HRV = heart rate variability
- https://cdn.elifesciences.org/articles/52059/elife-52059-supp2-v2.docx
-
Supplementary file 3
Task stages differences in sAA, rMSSD or LC BOLD variance.
Note: Linear mixed effects models with random intercept for each person and task stage as fixed effect. Estimates indicate the unstandardized beta-coefficients. P-values are adjusted for multiple comparisons using the False Discovery rate.
- https://cdn.elifesciences.org/articles/52059/elife-52059-supp3-v2.docx
-
Supplementary file 4
Relationship between frequency and coherence between LC and heart rate variability across the task stages (Spatially normalized pipeline).
Note: Linear mixed effects models with random intercept for each person, task stage, frequency and their interaction as fixed effect. Estimates indicate the unstandardized beta-coefficients. P-values are adjusted for multiple comparisons using the False Discovery rate.
- https://cdn.elifesciences.org/articles/52059/elife-52059-supp4-v2.docx
-
Supplementary file 5
Relationship between frequency and coherence between LC and heart rate variability across the task stages.
Note: Linear mixed effects models with random intercept for each person, task stage, frequency and their interaction as fixed effect. Estimates indicate the unstandardized beta-coefficients. P-values are adjusted for multiple comparisons using the False Discovery rate.
- https://cdn.elifesciences.org/articles/52059/elife-52059-supp5-v2.docx
-
Supplementary file 6
Relationship between frequency and coherence between LC and heart rate variability across the task stages (Fixed + Explicit Resp pipeline).
Note: Linear mixed effects models with random intercept for each person, task stage, frequency and their interaction as fixed effect. Estimates indicate the unstandardized beta-coefficients. P-values are adjusted for multiple comparisons using the False Discovery rate.
- https://cdn.elifesciences.org/articles/52059/elife-52059-supp6-v2.docx
-
Supplementary file 7
Relationship between frequency and coherence between LC and heart rate variability across the task stages (Fixed + Explicit Phys pipeline).
Note: Linear mixed effects models with random intercept for each person, task stage, frequency and their interaction as fixed effect. Estimates indicate the unstandardized beta-coefficients. P-values are adjusted for multiple comparisons using the False Discovery rate.
- https://cdn.elifesciences.org/articles/52059/elife-52059-supp7-v2.docx
-
Supplementary file 8
Task-related activation patterns during encoding and recollection and functional connectivity (FC) alterations across the task stages.
Note: Brain regions activated during encoding, recollection or functional connectivity differences across the conditions (Family-wise error corrected for multiple comparisons using TFCE at p<0.05). Coordinates are provided in 1 mm MNI-space.
- https://cdn.elifesciences.org/articles/52059/elife-52059-supp8-v2.docx
-
Transparent reporting form
- https://cdn.elifesciences.org/articles/52059/elife-52059-transrepform-v2.docx
Download links
A two-part list of links to download the article, or parts of the article, in various formats.
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Dynamic behavior of the locus coeruleus during arousal-related memory processing in a multi-modal 7T fMRI paradigm
eLife 9:e52059.
https://doi.org/10.7554/eLife.52059
