1. Neuroscience
Download icon

Dynamic behavior of the locus coeruleus during arousal-related memory processing in a multi-modal 7T fMRI paradigm

  1. Heidi IL Jacobs  Is a corresponding author
  2. Nikos Priovoulos
  3. Benedikt A Poser
  4. Linda HG Pagen
  5. Dimo Ivanov
  6. Frans RJ Verhey
  7. Kâmil Uludağ
  1. Gordon Center for Medical Imaging, Department of Radiology, Massachusetts General Hospital/Harvard Medical School, United States
  2. Faculty of Health, Medicine and Life Sciences, School for Mental Health and Neuroscience, Alzheimer Centre Limburg, Maastricht University, Netherlands
  3. Faculty of Psychology and Neuroscience, Department of Cognitive Neuroscience, Maastricht University, Netherlands
  4. Center for Neuroscience Imaging Research, Institute for Basic Science and Department of Biomedical Engineering, Sungkyunkwan University, Republic of Korea
  5. Techna Institute & Koerner Scientist in MR Imaging, University Health Network, Canada
Research Article
  • Cited 0
  • Views 817
  • Annotations
Cite this article as: eLife 2020;9:e52059 doi: 10.7554/eLife.52059

Abstract

A body of animal and human evidence points to the norepinephrine (NE) locus coeruleus (LC) system in modulating memory for arousing experiences, but whether the LC would recast its role along memory stages remains unknown. Sedation precluded examination of LC dynamics during memory processing in animals. Here, we addressed the contribution of the LC during arousal-associated memory processing through a unique combination of dedicated ultra-high-field LC-imaging methods, a well-established emotional memory task, online physiological and saliva alpha-amylase measurements in young adults. Arousal-related LC activation followed amygdala engagement during encoding. During consolidation and recollection, activation transitioned to hippocampal involvement, reflecting learning and model updating. NE-LC activation is dynamic, plays an arousal-controlling role, and is not sufficient but requires interactions with the amygdala to form adaptive memories of emotional experiences. These findings have implications for understanding contributions of LC dysregulation to disruptions in emotional memory formation, observed in psychiatric and neurocognitive disorders.

Introduction

Multiple decades of neuroscience and psychology research uncovered that arousal contributes to enhanced memory performance and that norepinephrine (NE) transmission plays a pivotal role in this phenomenon (van Stegeren, 2008; Roozendaal and McGaugh, 2011). The major source of NE to the brain is the locus coeruleus (LC), a small structure in the brainstem. The NE-LC system has widespread efferents to almost the entire brain and can modulate cognition, behavior and autonomic tone via its effects on adrenoreceptors in target neurons (Sara, 2009; Sara and Bouret, 2012; Samuels and Szabadi, 2008a).

Research into NE-modulation of emotional memory enhancement has focused largely on the amygdala (Hermans et al., 2014; Phelps and LeDoux, 2005; McIntyre et al., 2003), an important target region of the LC and a critical nexus in memory for arousing experiences. Rodent work consistently demonstrated that blocking β-adrenergic receptors in the basolateral amygdala (BLA) immediately after learning in arousing conditions impairs memory, while infusion of NE or β-adrenoreceptor agonists enhances consolidation and rescinds the impairment, evaluated one to two days later (Roozendaal et al., 2008; Power et al., 2002). Correspondingly, stimulation of the LC during the learning phase modulates activity in the BLA and synaptic plasticity in the CA1, CA3 and dentate gyrus (Lemon et al., 2009; Wagatsuma et al., 2018). While these studies highlight the importance of adrenergic signaling for plasticity and hippocampal-dependent learning, blocking β-adrenoreceptors prior to learning had no effect on subsequent retrieval 24 hr later (Wagatsuma et al., 2018; Khakpour-Taleghani et al., 2008). It has been suggested that in the context of strong LC stimulation tonic activity can be shifted into phasic activity and other neuromodulators may then elicit similar but possibly weaker plastic processes (Roozendaal and Hermans, 2017; Vazey et al., 2018; Hansen and Manahan-Vaughan, 2015).

In humans, administration of propranolol 90 min before encoding attenuated expected amygdala responses during encoding of emotional stimuli, and even though the propranolol was eliminated at retrieval, hippocampal responses during retrieval of emotional stimuli were absent (Strange and Dolan, 2004). In addition, propranolol impaired memory accuracy for emotional events and decreased saliva measures of NE (Hermans et al., 2014; McIntyre et al., 2003; Strange and Dolan, 2004; Cahill et al., 1994). These findings suggest, consistent with animal studies, that emotional memory is dependent upon β-adrenergic-related amygdala-hippocampus interactions.

These network interactions can vary over time or be task-dependent. For example, the LC is involved in consolidation, but within a critical time period after learning, the time required for the hippocampus to facilitate the reorganization and stabilization of storing information in the neocortex (Khakpour-Taleghani et al., 2008). In humans, increases in NE-tone during encoding, but not during retrieval, modulated memory performance (Rimmele et al., 2016). This dynamic behavior stems from the idea that release of NE due to LC-activation modulates, gates and tunes neural activity in a way that optimizes the signal-to-noise ratio (Sara, 2009; Sara and Bouret, 2012). Thus, NE selectively reinforces brain activity and memory representations only if arousal-related LC activation occurs at the right time and magnitude (Mather et al., 2016). When increases in NE are coupled to arousal, they can lead to hippocampal synaptic plasticity (Lemon et al., 2009; Hansen and Manahan-Vaughan, 2015), and facilitate the dynamic reorganization of neural networks (Bouret and Sara, 2005; Zerbi et al., 2019). So far, arousal-induced dynamic changes in LC activity and connectivity with its key target regions during task performance has not yet been examined in animals (Zerbi et al., 2019).

Motivated by the previous findings, we sought to examine the contribution of the LC directly in vivo in humans to arousal-related memory enhancement and hypothesized that interactions between the LC and MTL regions vary across memory stages, dependent on the level of arousal.

Examining the LC in vivo functionally and obtaining a robust signal in humans is exceedingly difficult given its small size, proneness to noise (pulsatility of surrounding vessels and motion) and its location adjacent to the fourth ventricle. Our recent efforts in ultra-high-field imaging, providing better spatial resolution and signal-to-noise ratio, open up exciting possibilities. Here, we combined 7T-fMRI scanning during an established emotional memory paradigm, our novel 7T-MRI sequence which allows to image the LC at high resolution (Priovoulos et al., 2018), with objective measures of arousal, viz. online physiological measures and serial saliva measures, a proxy for NE, in young adults.

Results

Twenty-seven young right-handed adults (13 female, 20–30 years, mean ± SD = 22.95±1.96) underwent 7T MR imaging which included structural scans (T1-weighted scan, dedicated LC imaging) and ultra-high-resolution blood oxygenation level-dependent (BOLD) signal fMRI. The functional imaging part consisted of a baseline resting-state scan and an emotional memory task, in which participants were required to memorize face-name associations (Figure 1). Faces had an emotional (negative) or neutral expression as determined by independent rating. After the encoding phase, a second resting-state scan (termed ‘consolidation’) was collected. This was followed with a recollection fMRI-scan, during which participants had to indicate whether they recognized the face and, upon endorsement, had to select the name of the face out of three options (all options were seen previously to stimulate recollection instead of recognition processes). During the functional imaging, we collected breathing (respiratory bellows), cardiac information (pulse oximeter) and serial saliva samples for alpha-amylase (sAA) measures, a proxy for catecholaminergic activity. Three fMRI-datasets were excluded due to technical reasons, one due to spiking because of coil failure one due to a reconstruction error and one due to temporary failure of the response-recording device. Supplementary file 1 summarizes the neuropsychological test performance of the participants.

Figure 1 with 2 supplements see all
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.

Emotional memory performance is linked to sAA and heart-rate variability changes

We first evaluated the effect of emotional valence on memory performance, as it is well established that emotional information is better remembered than neutral information. All participants performed the task above chance level in recollection (33.3%), with a median number of hits (correctly recollected) of 48.33% (IQR = 40.78, 54.89) and median number of miss rate (incorrectly recollected) of 51.67% (IQR = 45.11, 59.22). There was no significant difference (Wilcoxon paired test: Z = −1.17, p-value=0.244) between the recognition hit rate for emotional (Median = 53.33% (IQR = 45.56, 61.11)) compared to neutral faces (Median = 53.33% (IQR = 44.44, 67.79, Figure 2A). However, more false alarms were detected (Wilcoxon paired test: Z = 3.81, p-value<o0.001) for emotional (Median = 22.73% (IQR = 18.19, 34.09) compared to neutral faces (Median = 13.63% (IQR = 9.09, 22.78); Figure 2B). We observed no recollection difference (Wilcox test: Z = 0.06, p-value=0.95, Figure 2D) between emotional (median = 0.5 (IQR = 0.43, 0.54)) and neutral valence trials (median = 0.45 (IQR = 0.38,0.59), Table 1). However, given that performance in the recollection phase may be affected by a response bias in the recognition phase (Figure 2E–F), we calculated the false alarm rate and the response bias using the likelihood ratio beta from signal detection theory. A paired t-test revealed a bias in misclassifying emotional faces during recognition (t(27)=5.34, p-value<0.001, Figure 2B) and a higher response bias for the emotional recognition condition (t(27)=2.49, p-value=0.019, Figure 2C), towards responding that they had seen the face before. We decided to focus on recollection because of this response bias, our focus on the medial temporal lobe regions and the notion that recollection is more likely to elicit hippocampal activity than recognition (Rugg and Vilberg, 2013).

Figure 2 with 1 supplement see all
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.

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))
Zp-value
Recognition Hit Rate
(Old faces)
53.33
(45.56, 61.11)
53.33
(44.44, 67.78)
−1.170.244
Recognition Miss Rate (Old faces)46.67
(38.89, 54.44)
46.67
(32.22, 55.56)
1.20.229
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 Rate23.94
(17.02, 32.83)
37.68
(28.23, 50.81)
−3.16<0.001
Recognition Response Bias0.90
(0.81, 0.92)
0.84
(0.74, 0.9)
2.290.02
Recollection Hit Rate Raw24.44
(18.89, 31.11)
26.67
(18.89, 33.33)
0.350.726
Recollection Hit Rate (correct recognition)50
(43.43, 54.01)
44.83
(37.97, 58.77)
0.060.953
Recollection Hit Rate (Old faces) - Recollection False Alarm Rate (Old faces)0.00
(−13.14–8.01)
−10.34
(−24.07–17.54)
0.060.953
Bias-corrected Recollection Hit Rate38.68
(36.68, 39.46)
29.97
(28.01, 33.39)
4.41<0.001

Repeating the analysis after correcting for the bias (partial residuals), showed that participants recollected on average more names with an emotional (median = 38.68, IQR = 36.68, 39.46) relative to those with a neutral valence (median = 29.97, IQR = 28.01, 33.39; paired Wilcoxon test: Z = 4.41, p-value<0.001, Figure 2G), in accordance with an expected memory advantage for emotional trials. No reaction time difference between emotional (median = 1.90 (IQR = 1.68,2.10)) and neutral conditions (median = 1.88 (IQR = 1.71, 2.02)) was detected in recollection (Z = 0.09, p-value=0.93) or recognition (Z = −0.43, p-value=0.68; emotional (median = 1.50 (IQR = 1.32, 1.65)); neutral (median = 1.52 (IQR = 1.32, 1.61))). In the remaining analyses, we used this bias-corrected emotional recollection score (Figure 2—figure supplement 1; Table 1 for an overview of all scores). Males and females did no differ in the corrected recollected rate (β = 1.49, z = 1.18, p=0.24), nor did we observe a possible modulation by valence (interaction sex by valence: β = 0.20, z = 0.91, p=0.91).

For task clarification, see Figure 1. The ‘Recollection Hit Rate Raw’ expresses the number of correct recollection of old names divided by the total number of old faces. The ‘Recollection Hit Rate (correct recognition)’ expresses the recollection success normalized for recognition, that is the number of correct recollection of old names divided by the number of correct recognition of old faces (%Hits). The %False Alarm is 1- %Hits (or Recollection Hit Rate (correct recognition)”). The ‘Bias-corrected Recollection Hit Rate’ is the bias-corrected recognition rate, as described in Figure 2—figure supplement 1. Note that the recollection hit rate for new faces is by definition 0, since they have not been shown in association with a name during encoding.

While there is general agreement in the valence of these images, individual differences in the elicited arousal are common. Therefore, we collected autonomic tone measures. We recorded heart rate variability (HRV), a proxy of heart rate variability and an index of parasympathetic modulation, particularly in the high (0.15–0.4 Hz) and low-frequency band (0.03–0.15 Hz) (Hedman et al., 1995). Root-mean square differences of successive peak-to-peak intervals (rMSSD) were calculated over each time period. RMSSD correlates tightly to parasympathetic activity (Kleiger et al., 2005) and is relatively free of respiratory influences. In addition, we measured sAA, a sensitive marker for the sympathetic nervous system, in particular NE.

To facilitate analyzing rMSSD and sAA at similar time points along the paradigm, we calculated the difference in sAA (ΔsAA) levels before and after each of the stages of the fMRI task (resulting in four values per participants, except for missing values due to insufficient saliva; Supplementary file 2). Linear mixed effects models with ΔsAA across the task stages revealed an increase in ΔsAA during consolidation relative to encoding (β = −65.95, t(47.92) = −2.91, p=0.033, Figure 3A). No differences in ΔsAA were found when comparing the other task stages (Supplementary file 3A). These findings did not change when covarying for the baseline sAA measurement outside the scanner or when covarying for sex. To further probe the increase in ΔsAA during consolidation, correlation analyses showed a positive, at trend level, association with emotional memory performance (r = 0.452, N = 18, p-value=0.059, Figure 3B).

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.

For the rMSSD, the linear mixed effects models showed no significant change in rMSSD across the task stages (Figure 3C, Supplementary file 3B). Given that both measures relate to the autonomic system, we then correlated ΔsAA to RMSSD across the stages using bootstrapped (n = 5000) repeated measures correlations and observed a negative relationship (Rrm(df)=−0.387,(35), p=0.018, CI (95%)=[−0.638,–0.061], Figure 3D). These observations suggest that perturbations of arousal with a NE-dominant response, reflected in increases in sympathetic and decreases in parasympathetic tone, can enhance memory performance in healthy adults.

LC activity couples with autonomic tone across the task stages

The LC is the centerpiece of the brain’s sympathetic noradrenergic system, and is known to control autonomic function. Therefore, we examined the relationship between LC fMRI and HRV timeseries across the task stages.

Using brainstem-specific group-level independent-component-analyses (ICA) on the baseline resting-state fMRI in combination with a custom anatomical LC template from our MT-TFL sequence, we identified a bilateral LC component (Figure 4A and B). To assess the specificity of our findings to the LC, we also investigated a control component in the pons (reference ROI). In addition, sensitivity analyses were done by evaluating the outcomes across four different preprocessing pipelines, each time adding in more physiological noise correction (Figure 4—figure supplement 1).

Figure 4 with 2 supplements see all
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.

Linear mixed effects models showed no differences across task stages for LC BOLD variability (Supplementary file 3C). Consistent with our previous analyses, we applied repeated measures correlation and observed a negative correlation between LC variance and rMSSD across task stages (Rrm(df)=−0.326,(63), p-value=0.008, CI (95%)=[−0.531,–0.084], Figure 4C). We found no correlation between rMSSD and variance in a control region, a component close to the 4th ventricle that explained a similar amount of variance in the ICA (Rrm(df)=−0.201(63), p=0.11, CI (95%)=[−0.494, 0.427], Figure 4D). We also observed no correlation between LC variance and ΔsAA (Rrm(df)=−0.006,(38), p-value=0.971, CI (95%)=[−0.325, 0.314], Figure 4E) or between the reference component variance and ΔsAA (Rrm(df)=−0.012,(38), p-value=0.943, CI (95%)=[−0.309, 0.330], Figure 4F). These findings were reproduced across preprocessing pipelines indicating the robustness of our results (Figure 4—figure supplement 2).

In the previous analyses, we demonstrated a negative correlation between LC variability and rMSSD or ΔsAA across the entire paradigm. These results echo previous work demonstrating that blocking β-adrenoreceptor increases HRV (Bittiner and Smith, 1986). However, rMSSD reflects beat-to-beat variance in HR in the time-domain and is relatively non-specific to the frequency-domain. Autonomic fluctuations due to emotionally salient stimuli can alter the power of frequencies in the HRV and this can occur in concert with changes in LC BOLD responses (Bittiner and Smith, 1986; Akselrod et al., 1981; Lane et al., 2009). Therefore, we applied spectral analyses to assess the nature of autonomic modulations on LC activity at each task stage by calculating the coherence magnitude squared between the timeseries of the HRV and LC for each task stage (see example in Figure 5A). By using a within-subject design with a baseline measure we can eliminate interindividual differences in blood pressure, respiration and, by combining HRV fluctuations with other measures related to the autonomic system, we are well-positioned to differentiate autonomic contributions to LC activity due to emotion or arousal across the different memory stages. Here, we will use the FIXed+explicit Phys pipeline to remove respiratory confounding.

Figure 5 with 3 supplements see all
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).

We ran linear mixed effect models to examine whether the relationship between frequency and coherence varied across task stages (Supplementary file 7A), followed by simple slopes analyses to determine the region of significant (RoS) moderation at α <0.05. Greater LC-HRV coherence levels were observed during baseline compared to consolidation (β = −0.20, t(7889) = −6.65, p<0.001, RoS:<0.27 Hz), during encoding compared to consolidation (β = 0.13, t(7889) = 4.31, p<0.001, RoS:<0.29 Hz), during recollection compared to baseline (β = −0.20, t(7889) = −6.55, p<0.001, RoS:>0.25 Hz) and during recollection compared to encoding (β = −0.12, t(7889) = −4.17, p<0.001, RoS:>0.27 Hz; Figure 5C).

For the control region (Figure 5B and D, Supplementary file 7B), coherence between the reference region and HRV coherence was greater during recollection compared to baseline (β = −0.13, t(7889) = −4.15, p=0.001, RoS:>0.28 Hz), or at trend-level compared to consolidation (β = −0.09, t(7889) = −2.95, p=0.050, RoS:>0.33 Hz). These findings were replicated across preprocessing pipelines, except for the spatially normalized pipeline where associations between frequency and coherence were noisier (Figure 5—figure supplements 12, Supplementary file 45). These patterns indicate greater LC activity coupled to greater parasympathetic inhibition during baseline and recollection, which has been associated with better discrimination between emotional versus neutral faces, attentional resources and better retrieval of context information from stored representation to perform memory task successfully (Lane et al., 2009; Thayer et al., 2012). On the other hand, consolidation demonstrated lower coherence compared to encoding or baseline, which is consistent with the premise that lower HF-HRV is associated with higher arousal. The latter is consistent with the increase in ΔsAA during consolidation (Figure 3B).

To further probe the importance of coupling between autonomic tone and LC variability, we defined the frequency band of maximum coherence as the confidence interval of the median of the maximum-individual coherence between LC and HRV (median = 0.217, 95% CI=[0.189, 0.317]. Interestingly, the width of this CI also nicely fits with the standard definition of high-frequency band of parasympathetic influence on HRV [0.15, 0.4 Hz]). The median coherence within this band was extracted for each individual across conditions to obtain a single point-estimate for our repeated measures correlation analyses with ΔsAA. We observed a negative relationship between LC-HRV median coherence and ΔsAA (Rrm(df)=−0.383 (34), p-value=0.021, CI (95%)=[−0.64,–0.05], Figure 5E). We replicated this analysis across preprocessing stages. Additionally, we repeated this analysis for the reference component, but observed no significant relationship with ΔsAA (Rrm(df)=0.218 (34), p-value=0.201, CI (95%)=[−0.13, 0.52], Figure 5F). Finally, no significant correlation was detected between LC-HRV coherence during any of the stages and emotional recollection rate (all ps > 0.05). The covariation between LC activity and HRV demonstrate that even though both emotion and cognition operate simultaneously, encoding and recollection are coupled to moment-to-moment autonomic regulation supporting cognitive functions required for the task at hand. However, this LC-parasympathetic coupling, appertaining to the central autonomic network, is not sufficient to predict accurate emotional memory recollection. Conversely, LC BOLD signal during consolidation may be specifically tied to arousal-related sympathetic tone, an inference supported by our ΔsAA analyses and the at-trend correlation between ΔsAA during consolidation and memory performance. We should note that the physiological relevance of high-frequency fMRI signal remains unclear, but it has indeed been linked to rapid modulation of the brainstem driving information exchange (Billings et al., 2018).

Arousal-related successful encoding elicits LC activation

We hypothesized that the LC and amygdala would be activated during encoding of emotional compared to neutral faces. The GLM within our predefined mask (Figure 6—figure supplement 1) largely confirmed this, by showing activation in the CA1 and 3, bilateral BLA (extending into the centromedial amygdala) and right entorhinal cortex during successful encoding versus no successful encoding (Supplementary file 8A, Figure 6A). However, no significant clusters were detected for emotional >neutral or for emotional successful encoding >neutral successful encoding.

Figure 6 with 3 supplements see all
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.

Given our evidence of the LC’s role in modulation of the autonomic system, we examined contrasts from HRV-convolved clusters. No significant clusters were observed after cluster correction in the MTL or pons. But given the limited signal-to-noise ratio typical for brainstem responses and the small-elongated shape of the LC (10–12 voxels in our high-resolution images), we performed our analyses at an uncorrected p<0.05 within a small-volume search space. The high arousal (local HRV minima) events (High Arousal > Low Arousal) showed significant bilateral activation at the LC (peak z-value = 3.3, Figure 6C), consistent with our previous results examining HRV-LC coherence during encoding. In addition, significant voxels of LC activation were observed for the interaction of high compared to low arousal during successful encoding (High Arousal successful Encoding > Low Arousal successful Encoding; peak z-value = 5.2; Figure 6B), successfully encoding versus forgotten trials during high arousal (High Arousal successful Encoding > High Arousal forgotten, peak z-value = 2.9, Figure 6D) and bilaterally for encoding of emotional stimuli relative to neutral during high arousal (High Arousal Emotional Encoding > High Arousal Neutral Encoding, peak z-value = 5.8, Figure 6E).

The extracted beta-coefficients of this contrast correlated with emotional memory performance (r = 0.506, N = 24, p=0.019), indicating that higher LC activity in emotionally encoded trials under high arousal as compared to neutral trials was related to better memory performance. These results suggest that the MTL and BLA is involved in the accuracy of learning, while LC clusters are involved in encoding of emotional stimuli, a role that is particularly modulated by arousal.

The hippocampus predominates successful recollection

We expected substantial involvement of the hippocampus during successful recollection, but potentially also involvement of the LC and amygdala during emotional recollection. The GLM revealed activation in the left CA1 and subiculum for successful recollection (Supplementary file 8B, Figure 6F). We found no significant clusters for the emotional versus neutral contrast or for the interaction between recollection success and valence (emotional successful recollection >neutral successful recollection). While we did see a cluster encompassing voxels in the LC region for the emotional vs. neutral and emotional successful recollection >neutral successful recollection contrasts, this cluster contained voxels in the 4th ventricle and the peaks of the clusters were not located in the LC and therefore should be interpreted with caution. No clusters were observed in the MTL or pons related to the HRV local minima or when convolving arousal with the memory regressors. Therefore, involvement of the LC during recollection could not be confirmed in our data.

Interactions between the LC and MTL transition across task stages

To compare interactions between regions across the task stages we performed seed-to-voxel analyses using cross-correlation at lag = 0 s (denoted as functional connectivity FC) between the LC using the component of the baseline resting-state (Figure 4—figure supplement 1) and the predefined MTL regions (Figure 6—figure supplement 1): hippocampus (HIPP), amygdala (AMY) and entorhinal cortex (EC). We observed a decrease in FC during consolidation compared to baseline between the LC and left BLA, (para)subiculum and EC (Supplementary file 8C, Figure 7A). No effects were observed when comparing baseline with encoding or recollection and these findings were reproduced across preprocessing pipelines (Figure 7—figure supplement 1).

Figure 7 with 1 supplement see all
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.

Areas related to arousal and salience processing, such as the LC, have particularly variable behavior that can result in a negative FC value, complicating their interpretation (Chang and Glover, 2010). As our FC analyses assume stationarity within the fMRI timeseries, we aimed to understand these negative FC changes by exploring the frequency-specific magnitude and phase correlation changes in the follow-up spectral coherence analyses. Phase or magnitude changes in the frequency plane can result in reduced cross-correlation values.

To understand whether timeseries from specific regions were leading or lagging during the task stages, we estimated the median phase lag and coherence across the frequency band of maximum coherence between the timeseries of the LC and MTL regions. Similar to previously determined LC-HRV median maximum coherence (Figure 5), we calculated the confidence interval of the median of the maximum-individual coherence between timeseries. Linear mixed effects models were implemented with either median coherence or phase at maximum coherence between the task changes as outcome and task stage, timeseries combination and their interaction as fixed effects. We observed an increase in HIP-AMY coherence for encoding compared to baseline (β = −0.13, t(821)=-4.39, p<0.001, 95% CI[−0.20,–0.052]), a decrease in recollection compared to encoding (β = 0.10, t(821)=3.37, p=0.004, 95% CI[0.023, 0.171]) and a trend for consolidation decrease compared to baseline (β = −0.07, t(821)=-2.35, p=0.088 95% CI[−0.141, 0.007]) and a change in phase from AMY leading the LC in encoding to the LC leading the AMY in recollection (β = 0.91, t(821)=2.73, p=0.033, 95% CI[0.051,1.772] Figure 7B). Coherent behavior between timeseries may be influenced by other brain regions. Therefore, we calculated partial coherences and compared these against a null distribution consisting of partial coherences from randomized time-series with matched auto-regressive coefficients. The surviving significant coherent relationships between timeseries are visualized in Figure 7 (B–E) in terms of coherence magnitude and phase. Accounting for timeseries of the AMY reduced HIPP-LC coherence during encoding, and accounting for HIPP timeseries significantly reduced coherence between AMY and LC during consolidation.

When relating emotional memory to coherence for these ROI pairs using linear regressions (adjusted for baseline coherence), we observed that greater memory performance was associated with greater AMY-EC coherence during encoding (β = 0.28, t(21)=2.62, p=0.038, 95% CI[0.057,0.508]), and EC-HIP coherence during encoding (at trend level: β = 0.27, t(21)=2.38, p=0.054, 95% CI[0.033,0.512]).

LC structure is associated with sAA and memory performance

Previous studies related LC structure (intensity) to cognitive and autonomic tone measures (Hämmerer et al., 2018; Mather et al., 2017). The LC segmentation is depicted in Figure 8A–B (see also Figure 8—figure supplement 1). The LC TFL intensity normalized to a pontine tegmentum ROI showed a positive correlation with emotional recollection scores (r = 0.56, p=0.004, Figure 8D). Increased motion correlated borderline with TFL intensity ratio (r = 0.38, p=0.071) and emotional recollection (r = 0.34, p=0.108), but adjusting for motion did not alter the positive relationship between normalized TFL intensity and emotional recollection (p=0.004). When subdividing the LC into three equisized parts (Figure 8C), we observed a positive correlation with emotional recollection (Figure 8D) for the medial (r = 0.52, p=0.009) and caudal (r = 0.43, p=0.034) parts of the LC, not rostral (r = 0.37, p=0.074).

Figure 8 with 1 supplement see all
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.

We then related the mean TFL intensity ratio to our measure of autonomic tone and observed a positive relation with ΔsAA during consolidation (r = 602, p=0.008, N = 18). When examining the different sections of the LC, this positive relationship was only observed for the caudal part of the LC (r = 0.79, p<0.001, N = 18). No association was observed between TFL intensity and rMSSD or LC BOLD variability.

Discussion

The LC is connected with evolutionary ancient brainstem regions involved in autonomic functions, but also with the associative neocortical regions, allowing it to be ‘the cognitive limb of a globally conceived sympathetic nervous system’ (Aston-Jones et al., 1991). We set out to investigate the contribution of the LC and its interactions with MTL regions during arousal-related memory enhancement by combining dedicated ultra-high-field MRI methods, an established emotional memory task, physiological and saliva measures. Consistent with previous work, we observed that forming adaptive memories in response to arousing events was predominantly associated with NE- activity. Considerably less is known about the LC’s interactions with MTL regions during the different memory processing stages. Our results now demonstrate that LC’s role during these memory processing stages was shaped by arousal-related autonomic changes and its close interactions with the amygdala. During encoding, stimuli eliciting arousal-related autonomic changes were associated with LC activity following amygdala activation during successful learning. During consolidation, the role of the LC gradually changed from being arousal-driven to maintenance or updating internal models during recollection, involving predominantly hippocampal activity. These results highlight the dynamic nature of neuromodulatory brain networks during arousal-related memory processes (Hermans et al., 2011). Understanding the neural mechanisms underlying memory for emotional experiences increases our understanding of diseases in which a dysregulation or decoupling of one of these systems could lead to affective, autonomic or neurocognitive disorders (Samuels and Szabadi, 2008b; Jacobs et al., 2019).

Activation of the LC occurs in parallel with the autonomic system, and modulates cognition: across the entire memory task greater variability in LC BOLD signal was associated with lower parasympathetic activity (rMSSD), which in turn was associated with greater sympathetic modulation (ΔsAA) and better memory performance. Beyond function, we also observed that structural properties of the LC, the middle section, related to better emotional memory performance, and greater caudal LC intensity was related to greater NE. These correlations fit with the well-described topographic projections of the LC, where the rostral and middle part project densely to higher-order associative regions of the brain and the caudal part of the LC to the nucleus ambiguus and dorsal motor nucleus of the vagus (Sara, 2009; Samuels and Szabadi, 2008a), which both modulate the parasympathetic outflow based on sympathetic signals.

We also observed task-stage-specific associations between LC’s connectivity to the MTL regions and autonomic tone measures. During consolidation, we observed a different LC-MTL functional connectivity coupling relative to baseline, possibly reflecting transient processes related to less parasympathetic tone (lower coherence in the high frequency domain). Correspondingly, increases in sAA during consolidation correlated positively with emotional performance. Such increases in sympathetic tone may reflect arousal-related NE-synaptic consolidation processes (Packard et al., 1994).

While systems consolidation requires years, synaptic consolidation represents the initial phase after learning where mnemonic representations change and new memory tracers are formed. Previous reports demonstrated that synaptic consolidation may be captured by the resting-state signal after learning a task (Vilberg and Davachi, 2013; Tambini et al., 2010; Jacobs et al., 2014). This consolidation phase may be a transitional stage, where the role of the LC slowly alters from an arousal-based learning experience to maintenance of memories via long-term potentiation (LTP). LTP depends on the activation of the BLA, highlighting the critical role of the amygdala (Frey et al., 2001). A myriad of animal studies have demonstrated that elevated NE in the amygdala affects memory consolidation for stress-induced events and can facilitate hippocampal synaptic plasticity (Hermans et al., 2014; McIntyre et al., 2003).

In support of this premise, we observed mainly arousal-related LC activity during encoding. During encoding, the LC was activated during (successful) acquisition of information that was associated with elevated arousal. However, amygdala and hippocampal activation was only detected for successful learning, independent of valence of arousal. Arousal-related amygdala activation may have been attenuated by the cognitive task during encoding (Lange et al., 2003). Nonetheless, arousal-related LC-activation with neural input from amygdala-entorhinal co-activation was associated with memory enhancement. This is largely consistent with work reporting that sympathetic arousal can predict memory performance for high-arousing stimuli when this coincides with amygdala activation (de Voogd et al., 2016). During recollection, we observed no arousal-related activity, but successful recollection was associated with CA1 activity. Activation of β-adrenoreceptors in the CA1 during consolidation is required for later retention of the memory task, but not for the aversive component of the task (Mello-Carpes and Izquierdo, 2013; Chen et al., 1992). The markedly higher frequency window of the HRV-LC coherences during recollection, suggests less NE-sympathetic responsiveness at this stage. Increases in LC-hippocampus and LC-amygdala coherence variability during recollection are possibly related to increased parasympathetic inhibitory modulation supporting cognitive operations. Alternatively, it could be speculated that spill-over from NE- sympathetic processes during consolidation drive retention processes.

In addition to different activation patterns across stages, we also observed a switch in the role of LC’s connectivity with MTL regions. During encoding, the LC is on the receiving end, in particular from the amygdala, whereas this changes to a leading role in recollection. This phase switch between the LC and amygdala conforms with functional and anatomical descriptions of projections between the LC and central amygdala or BLA. The LC receives afferents from the central amygdala mainly at the periLC-region (Bergado et al., 2007). The central amygdala controls stress responses and is involved in salience processing and attention processes, important for efficient encoding (Phelps and LeDoux, 2005; Roozendaal et al., 1996). While the central amygdala has no effect on hippocampal LTP, the BLA has memory-enhancing effects on the subiculum and CA1 via β-adrenergic mechanisms (Phelps and LeDoux, 2005; Roozendaal et al., 1996). The NE-input to the BLA arrives mainly from the LC (Chen and Sara, 2007). The BLA does not project directly to the LC, but via a small direct projection to the central amygdala (Bouret et al., 2003).

Recent general theories of brain function, such as the active inference theory (Friston et al., 2015), posit that NE responses from the LC can be considered as a prediction error during learning originating from information from the amygdala, whereas signals from the LC to the amygdala and cortex guide transfer of information and model updating (Sales et al., 2019). In these models, the amygdala is critical to maintain allostasis. The central amygdala network would control the arousing physiological responses through its descending predicting signals to the brainstem, whereas the BLA would be part of a neutral signal uncertainty network as it receives prediction errors from the LC. It has been suggested that NE increases accuracy of episodic memories (Barsegyan et al., 2014). While our study was not designed to test the active inference theory, the proposed models are consistent with our phase-differences across stages. Future in-vivo recordings, computational and carefully designed studies are warranted to test these models in detail.

No study is without limitations. First, as our focus was on the LC and MTL, we specifically opted to acquire fMR images with a high isotropic spatial resolution. This limited our field of view and therefore, we did not investigate distant modulatory effects. Second, the LC is challenging to investigate in humans due to limits in spatiotemporal resolution and lower signal-to-noise ratio in the brainstem compared to neocortical signal. We had to eschew cluster-based correction. The small in-plane diameter of the LC means that only low-levels of smoothing can be applied, which would reduce the signal-to-noise, and the probability of discovering a contiguous cluster in the highly-elongated axial plane of the structure. Ultimately, this would result in an overly stringent multiple comparison adjustment. Nevertheless, we are confident that our GLM findings do not represent false positives due to their reproducibility across preprocessing pipelines, correlation to behavior and consistent anatomical co-localization with an ex-vivo validated template and anatomical atlases.

To conclude, interactions between the NE-LC, amygdala and hippocampus, drive arousal-related memory processing, but transitions in interactions occur along the memory processing stages. During successful encoding, we observed high-arousal-related LC activation, possibly reflecting attentional processes, in conjunction with amygdala activation. During consolidation, LC-MTL connectivity is dynamic and associated with NE increases possibly reflecting LTP. This gradually transitions to learning and updating internal models during recollection, which mainly involves hippocampal activity. These findings contribute to our understanding of LC activation for forming adaptive memories of emotional experiences and can serve as a basis for understanding how LC dysregulation may potentially disrupt this adaptive behavior.

Materials and methods

Participants

In total, 27 healthy young right-handed adults (13 female, age 20–30 years old, mean ± SD = 22.95±1.96) were recruited via local advertisements. All participants were screened to exclude a history of major psychiatric or neurological disorders, having a history of brain injury of brain surgery, taking medications that may influence cognitive functioning or being not eligible for MRI-scanning. All participants had normal or corrected-to-normal visual acuity and reported normal hearing abilities. Since depression has been linked to LC atrophy, all participants were screened on the Hamilton Rating Scale for Depression (all within normal range = 0–10; mean ± SD = 3.1±3.3). All participants also endorsed a degree of appraisal of stress-full events that was within the normative range on the Perceived Stress Scale-10. The Perceived Stress Scale-10 is a questionnaire tapping into global non-specific appraisal of stress that also relates to glucocorticoid levels and the number of stressful life events. All participants received monetary compensation for their participation and provided written informed consent. Approval of the experimental protocol was obtained from the local ethical committee of the Faculty of Psychology and Neuroscience at Maastricht University.

Neuropsychological assessment

Request a detailed protocol

All participants underwent a comprehensive battery of cognitive tests covering episodic memory (15-word learning test (WLT): learning and delayed recall), working memory (digit span forward and backward), language (semantic and category verbal fluency), attention (concept shifting task (CST)), information processing speed (Letter Digit Substitution Test (LDST)) and executive functions (Stroop Color Word Task (SCWT)). All cognitive tests were presented in the same order to each participant.

Experimental paradigm

Request a detailed protocol

To investigate emotional memory a well-established face-name association task (Sperling et al., 2001) was modified by including faces with negative or neutral valence.

Stimuli selection

Request a detailed protocol

Several databases consisting of emotional and neutral faces (Ekman, PERT96, ER40, MMI facial expressions, Radboud faces, Utrecht ECVP) (Ekman, 1992; Gur et al., 2002; Langner et al., 2010) were concatenated. All faces were transformed to gray-scale and resampled at the same resolution (750 × 830 pixels2) using Matlab and were manually cropped to the same template and resolution. Distinctive details beyond the face were darkened manually. Pictures that still contained potential distinctive features, such as facial hair, were excluded. All images had the same mean luminance and were presented on a black background. An independent sample of 10 individuals between 20–30 years old (five women) were shown the faces and asked to rate the emotional valence of each face on a 5-point Likert Scale (0: completely neutral, 5: very emotional). These ratings were ranked and 135 faces were selected (68 faces (50% male) with the highest emotional rating, 68 faces (50% male) with the lowest emotional rating (neutral)). Two Dutch native speakers (NV and HILJ) matched these 135 faces to an age-appropriate name from a list of the 400 most popular Dutch names (http://www.behindthename.com/top/lists/netherlands/).

Task description

Request a detailed protocol

The face-name association task (event-related design) consisted of encoding, consolidation, recognition and recollection stages. During encoding the participants were shown 90 faces (45 emotional, 45 neutral, gender-matched across conditions) along with their name for 4 s. The participants were asked to memorize the face-name association. To keep their attention to the task they were also asked to make a subjective judgment as to whether the face matched the name or not. A jitter of average length of 2 s (0.5–6 s, created with optseq [Dale, 1999]) was entered between trials, during which a fixation cross was shown. During the consolidation phase, which lasted approximately 8 min, the participants were asked to stare at a fixation cross. During the recollection stage, the participants were shown the 90 faces shown in encoding along with 45 new faces, matched for sex and emotion intensity to the initial 90 ones in a random order. The participants had 3 s to decide whether or not they had seen the face during encoding (recognition phase). If participants indicated they had seen the face before, they subsequently were asked to choose the name that was previously associated with that face (participants received three options, recollection phase). The three names shown were chosen randomly from the list of sex-specific names that were already shown in encoding, to avoid easy name recognition. Recognition or familiarity represents a fast process, the feeling that the face was seen previously without recollecting it. In contrast, recollection is a slow process that involves the active retrieval of details of the face-name pair (Yonelinas et al., 2010).

A jitter of average length of 1.5 s (0.5–6 s) was employed between trials, during which a fixation cross was shown. No more than two consecutive faces of emotional valence were shown during any of the task stages. Stimuli were viewed on a back-projection display via a mirror mounted onto the head coil. Behavioral responses were collected through a MR-compatible button box (Current Designs, eight-button response device, HHSC-2 × 4 C). All stimuli, event identities and timings were presented and logged using E-Prime 2.0 (Psychology Software Tools).

We calculated hits and false alarms for both the emotional and neutral condition during encoding. For the recollection (identification of the name), we calculated hits and false alarms for both the neutral and emotional condition. Reaction times were also analyzed. Performance in the recollection condition may be affected by the response bias in the recognition phase between emotional and neutral stimuli. Therefore, we calculated the likelihood ratio beta, as a measure of the response bias (beta = exp (d' x c)) (d' index calculated as by Stanislaw and Todorov, but corrected for ceiling effects [Jacobs et al., 2015]). This response bias was derived from the signal detection theory, a method that allows us to differentiate the signal from the noise (Stanislaw and Todorov, 1999). The bias measure indicates the extent to which a certain type of response is more probable than another, such as for example responding in a more conservative manner. We performed a paired t-test to compare the bias between emotional and neutral during the recognition phase. This bias was then regressed our from the emotional and neutral recollected hits in two separate regressions (Figure 2). The residuals of each regression represented the bias-corrected emotional and neutral recollection rate. Finally, to obtain an individual emotional recollection rate, we calculated an adjusted emotional memory performance score =Emotional retrieval rateEmotional retrieval rate+Neutral retrieval rate (Figure 2—figure supplement 1). This measure was used throughout the manuscript.

Imaging parameters

Request a detailed protocol

MR scans were performed in a 7T Magnetom Siemens (Siemens Healthineers, Erlangen, Germany) with a 32-channel head coil (Nova Medical, Wilmington, MA, USA). First, we acquired a Magnetization Prepared 2 Rapid Acquisition Gradient Echoes (MP2RAGE) sequence (Marques et al., 2010) for whole brain imaging (TR = 5000 s, TE = 2.47 ms, flip angle = 5o/3o, voxel size = 0.7×0.7 × 0.7 mm3, number of slices = 240). An in-house developed magnetization transfer-weighted turbo flash (MT-TFL) sequence sensitive to LC contrast (Priovoulos et al., 2018) was performed to image the LC at high resolution. The sequence consisted of a multi-shot 3D readout (TR = 538 ms, TE = 4.08, flip angle = 8o, voxel size = 0.4×0.4 × 0.5 mm3, number of slices = 60) with center-out k-space sampling, preceded by 20 long off-resonant Gaussian sinc pulses (pulse length = 5.12 ms, bandwidth = 250 Hz, B1 = 0.25μT). For the MT-TFL sequence, the field-of-view (FOV) was placed approximately perpendicular to the pons and covered the area between the inferior colliculus and the inferior border of the pons. A matched TFL sequence but with the MT pulses turned off was similarly acquired. For high resolution BOLD fMRI imaging we acquired multiband EPI sequences (Moeller et al., 2010; Setsompop et al., 2012) (TR = 2000 ms, TE = 19 ms, isotropic voxel size = 1.25×1.25 x.125 mm3, number of slices = 50, M-factor = 2, GRAPPA R = 3). The field of view was placed at an angle of 45o to the brainstem, to reduce the effect of physiological movement at the pons level and to be approximately perpendicular to the hippocampus to optimize subfield differentiation. After each BOLD fMRI acquisition, five more volumes were acquired with reversed phase encoding direction to facilitate distortion correction. Four functional runs were collected: initial resting-state, an encoding phase, second resting-state and a recognition phase.

Saliva samples collection

Request a detailed protocol

Salivary alpha-amylase (sAA) is a sensitive marker for sympathetic nervous system reactivity that correlates with plasma norepinephrine and shows similar dynamics (Rohleder and Nater, 2009). Samples of sAA were collected at various time points throughout the scan session: before entering the bore, before the first resting-state scan, before encoding, after encoding, before recollection, after recollection and 30 min after the scan session. This sampling scheme allows us to measure acute responses. Saliva samples were obtained with cotton swabs (Sarstedt, Nümbrecht, Germany) that were exchanged with the participants via a MRI-compatible arm extension. Each test session was scheduled at the same part of the day. Participants were instructed to refrain from food consumption or brushing their teeth one hour before scanning and to abstain from caffeinated drinks for four hours prior to the scanning. Compliance was checked with a questionnaire and all participants followed the instructions. Before the first saliva sample the participants were asked to rinse their mouth with water. The participants were instructed to place the cotton roll between their cheek and their teeth or beneath their tongue and to not chew on it; after one minute the cotton rolls were retrieved. The saliva samples were stored at −20°C until their processing at the Dresden University of Technology. After thawing, saliva samples were centrifuged at 3000 rpm for 5 min. Alpha-amylase was measured by a quantitative enzyme kinetic method Nater and Rohleder, 2009). When inadequate amount of saliva was collected at specific time points, these time points were excluded from the saliva analysis. Missing data for the sAA time point are provided in Supplementary file 2.

Pulse rate preprocessing

Request a detailed protocol

Heart rate and chest movements were measured concurrently during the fMRI acquisitions with a pulse oximeter and respiratory bellows respectively. The pulse rate signal was preprocessed in R with the package RHRV. Automatic removal of outliers was performed by adaptive thresholding (beats whose value exceeded the cumulative mean threshold were rejected) and by rejecting beats with physiologically improbable values (less than 25 and more than 200 beats-per-minute) (Vila et al., 1997). To allow power spectral analysis, we interpolated to an evenly spaced pulse rate series with a cubic space interpolation and pulse (heart) rate variability (HRV) timeseries (Hill and Siebenbrock, 2009; Pinheiro et al., 2016) (differences of successive R-R intervals, sampled equidistantly at the TR) were extracted. Root-mean square differences of successive R-R intervals (rMSSD) were calculated as a measure of high-frequency (0.15–0.40 Hz) arousal over each time period. rMSSD has been shown to tightly correlate with parasympathetic activity and be relatively free of respiratory influences (Hill and Siebenbrock, 2009). Lower rMMSD is associated with poor vagus-mediated HRV and indicative of higher stress of arousal. For the analyses focused on the time-domain, we used the rMSSD metric and for the frequency domain, we used the HRV terminology. Datasets with missing cardiac physiological signal for more than 10% of the time points were excluded (Supplementary file 2).

Preprocessing of MRI data

Preprocessing of structural data

Request a detailed protocol

The T1-weighted MP2RAGE images were processed using FreeSurfer (FS) version 6.0.0 (https://surfer.nmr.mgh.harvard.edu/) using the software package's default, automated reconstruction protocol as described previously (Fischl et al., 1999). Briefly, each T1-weighted image was subjected to an automated segmentation process involving intensity normalization, skull stripping, segregating left and right hemispheres, removing brainstem and cerebellum, correcting topology defects, defining the borders between grey/white matter and grey/cerebrospinal fluid and parcellating cortical and subcortical areas. The hippocampal-amygdala segmentation algorithm in FS 6.0 predicts the location of subregions by using a probabilistic atlas built from a combination of manual delineations of the hippocampal formation from ultra-high resolution ex-vivo MRI scans showing definitive borders and manual annotations of the surrounding subcortical structures (e.g. amygdala, cortex) from an independent dataset (Iglesias et al., 2015a). Using FS’s visualization toolbox, freeview, we visually inspected and, if necessary, edited each image for over- or under-estimation of the gray/white matter boundaries and to identify brain areas erroneously excluded during skull stripping. In addition, we checked that the hippocampal/amygdala subregion mask was well positioned and that the ranking of subfield-specific volumes was consistent with the literature. An automated segmentation of the pons of the brainstem with Bayesian inference was applied (Iglesias et al., 2015b). The T1 to MNI template registrations were calculated using a diffeomorphic transform with the antsRegistrationSyN.sh function (ANTS 2.1 (http://stnava.github.io/ANTs/) [Avants et al., 2011]). This algorithm has been shown to provide superior performance compared to most other registration algorithms used in neuroimaging (Avants et al., 2011; Klein et al., 2009).

The partial field-of-view high resolution MT-TFL scans were registered to the T1 data using the boundary-based registration. A study-specific template of the LC scans was created with an iterative diffeomorphic warp estimate using the buildtemplateparallel.sh script of the ANTS package. The LC was segmented on the template (by NP). This segmentation was repeated 4 weeks later and only voxels that were identified during both sessions were used in the final segmentation. The left LC was 12 mm in length (108 voxels, 9.918 mm3) while the right LC measured 10 mm in length (103 voxels, 9.459 mm3). The LC was divided in three equisized segments (rostral, medial and caudal) for further analyses. The segmented LC was projected to the high-resolution LC specific sequences, including MTR, MT-TFL and TFL to extract the mean intensity. MTR was calculated as

MTR=MTweighted-TFLTFL

The reference region for normalization was defined as a 10 × 10 voxels ROI in the pontine tegmentum at the same level as the LC). Normalization was done by dividing the LC intensity values with the reference region intensity. Given that motion affects MR images, a retrospective motion metric (AES) was also calculated with the homonymous Matlab toolbox.

Preprocessing of BOLD fMRI data

Request a detailed protocol

Image preprocessing of the fMRI data was performed with FSL 5.0.9 (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/). The first five volumes were removed to ensure that data within steady-state were examined. The functional images were motion-corrected to the last image with a 6-dof transform using MCFLIRT (Jenkinson et al., 2002). To reduce EPI-distortions a displacement field was calculated and applied with FSL-TOPUP by using the last 5 frames of each fMRI scan and the following reversed phase encoding direction scans. As the brainstem is known to be sensitive to physiological artefacts, we prepared four separate denoising pipelines, to allow us to check the reproducibility of our findings (Figure 1—figure supplement 2). The pipelines consisted of 1) minimally processed data (distortion-, motion- and slice-timing corrected; denoted as ‘Spatially normalized’ data), 2) spatio-temporally denoised data (distortion-, motion-, slice-timing corrected and with the ICA-FIX algorithm applied (Griffanti et al., 2014) and smoothed; denoted as ‘FIXed’ data), 3) spatio-temporally denoised data along with respiration temporal regressors (distortion-, motion-, slice-timing, FIXed corrected and with respiration regressors regressed out with PNM and smoothed; denoted as ‘FIXed + explicit Resp’ data) 4) spatio-temporally denoised data along with respiration and pulse temporal regressors (distortion-, motion-, slice-timing, FIXed corrected and with both respiration and pulse regressors regressed out and smoothed; denoted as ‘FIXed + explicit Phys’ data). To apply the FIX algorithm, the already available CMRR-based 7 T HCP dataset with a similar multiband factor and resolution was used as a trained-weight file; the noise component classification was checked against manual classification performed by NP and LP. The interrater difference between FIX and the manual classification was smaller compared to the interrater difference between the two manual raters and therefore, FIX was preferred. FIX removed a similar percentage of noise components across the task stages (median baseline: 48.52% [IQR = 41.76–52.45%]; median encoding: 54.47% [IQR = 35.42–64.04%]; median consolidation: 48.93% [IQR = 33.99–59.68%]; median recollection: 49.67% [IQR = 43.74–64.3%]).

The cardiac and respiratory waveforms recorded during the scan session were entered as slice-specific regressors; three orders of cardiac, fours orders of respiratory and a single set of interaction terms were used as suggested by Harvey and colleagues (Harvey et al., 2008). Because HRV is known to vary depending on the respiration phase (respiratory sinus arrhythmia) (Yasuma and Hayano, 2004), an additional binary slice-specific regressor was added that represented inhalation and exhalation to model BOLD variance related to one of the two phases. The regressors were then used to model the physiological noise in the fMRI signal; the noise signal was adjusted for interactions with the ICA-FIX denoising regressors, so that artefacts were not re-introduced due to the successive regressions. Finally, slice-timing correction was performed based on the multiband acquisition matrix and a 3D spatial smoothing kernel of 1.5 mm FWHM was applied (size of the kernel estimated in accordance with the minimum LC width). Throughout the main manuscript, the FIXed data were employed (since this ensured the maximum possible sample size, due to no missing pulse and respiration data). Results of the sensitivity analyses (reproducibility) by running the other pipelines are indicated were applicable. A boundary-based registration was calculated for the linear registration between the fMRI and T1data using FSL's epi_reg based on the FreeSurfer segmentations. The T1 to MNI template registrations were done using ANTS as described in the preprocessing of the structural data. The linear transformations were transformed to the ITK-SNAP reference using c3d tools and combined with the warps, so that when transforming the data from native fMRI to MNI there was a single interpolation. The accuracy of the transforms was individually checked by NP (Figure 1—figure supplement 1).

Initial resting-state (‘baseline’)

Request a detailed protocol

To robustly estimate the LC signal, we performed group-level ICA in the baseline resting-state data, which included projection of the fMRI data to the MNI 1 mm template, masking for the brainstem and temporal concatenation, followed by spatial ICA by maximizing non-Gaussian sources. The output group-level component map was thresholded at p<0.01 FWE-corrected and binarized. One component showed strong spatial correlation with our LC atlas (2.47% explained variance). We then applied dual regression to estimate the individual LC timelines; statistical significance was set at p<0.05 family wise error (FWE)-corrected for multiple comparisons with TFCE (Threshold-Free Cluster Enhancement). A component that explained a similar amount of signal variance and was at a similar location yet spatially unrelated to the LC was picked as an additional control component and the individual timelines were also extracted. Finally, the dual regression (and the extraction of timelines) was repeated with the same component map for all preprocessing stages (spatially normalized (after motion, distortion and slice-timing correction), with FIX applied, with FIXed + explicit Resp applied and with FIXed + explicit Phys) to ensure that the LC response was consistent across different steps of denoising and therefore not the result of physiological noise or the steps taken to remove it.

Task-related fMRI analyses

Request a detailed protocol

At individual level, a small-volume corrected general linear model was fit at each voxel with FMRI Expert Analysis Tool (FEAT), using generalized least squares with a voxel-wise, temporally and spatially regularized autocorrelation model. A high-pass filter (cutoff at 100 s) and minimal spatial smoothing (kernel = 1.5 mm) were applied before the fit. The area of interest (ROI) was defined in advance and consisted of the LC and surrounding region projected to the native fMRI space (LC, HIPP, AMY and EC being our a-priori ROIs; the 4th ventricle and pons were also included to ensure the spatial specificity of the effect and that it was not due to uncorrected motion due to proximity to a CSF space (total mask size in native space approximately = 18000 voxels). The regressors consisted of the emotional trials that were later successfully recollected, the neutral trials that were later successfully recollected, the emotional trials that were not recollected, the neutral trials that were not recollected and their respective convolutions with the HF-HRV local minima. Contrasts of interests were emotional trials vs neutral trials (Emotional >Neutral), successfully recollected trials vs. not successfully recollected (Successful encoding >Not successful encoding) and the interaction (Emotional successful encoding >neutral successful encoding).

The HRV local minima calculated over a rolling window of 15 s was also added as a regressor to account for the main effect of arousal on LC. Additionally, we examined (emotional) encoding during the local HRV over a sliding window of 15 s (high arousal). These regressors were convolved with the high arousal regressor and we contrasted high arousal events with the rest of the timeline (High Arousal vs. Low Arousal), as well their interactions with encoding (High Arousal Successful encoding >Low Arousal Successful Encoding, High Arousal Successful Encoding >High Arousal Not Successful Encoding). To examine the effect of emotional valence on encoding events under high arousal we calculated (High Arousal Emotional Encoding vs. High Arousal Neutral Encoding). These contrasts were examined both in the encoding and recollection task stages.

Contrasts related to recognition were not our primary interest, however, we fit encoding regressors for emotional trials that were later successfully recognized, the neutral trials that were later successfully recognized, the emotional trials that were not recognized and the neutral trials that were not recognized. Contrasts of interests were emotional trials vs. neutral trials (Emotional >Neutral), successfully recognized trials vs. not successfully recognized (Successful recognition >Not successful recognition) and the interaction (Emotional successful recognition >neutral successful recognition). Results are presented in Figure 6—figure supplement 3.

All regressors were convolved with a double gamma haemodynamic response function before entered in the model. The contrasts of the coefficients and their variance estimates were subsequently warped to the MNI 1 mm space and a group-level linear mixed effects regression was fit at each voxel, using generalized least squares with a local estimate of random effects variance and outlier de-weighting. The Z-maps were cluster-wise corrected with a cluster significance threshold of p<0.05 after being thresholded at Z > 2.3. Cluster peaks are reported with respect to the Freesurfer atlas.

To extract beta estimates from the LC, we first dilated the LC-segmentation by 2.3 mm and excluding voxels that overlapped with the 4th ventricle mask. The dilation was performed to match the physiological point spread function (FWHM) of the BOLD effect at 7T for gradient-echo acquisitions (Shmuel et al., 2007). Then, the z-map was thresholded at p<0.05 and extracted the mean subject-specific beta values using the dilated LC mask.

LC Functional connectivity across all memory stages

Request a detailed protocol

Seed-to-voxel functional connectivity analyses were performed between the LC and regions of interest in the MTL (hippocamus, amygdala and entorhinal cortices) across all task stages. The LC ROI was defined based on ICA component in the baseline resting-state scan (see ‘initial resting-state’). Correlation maps were generated by correlating the average BOLD time course from the LC-seed with the BOLD timeseries of each voxel within our a-priori defined ROIs (hippocampus (HIPP), amygdala (AMY) and entorhinal cortices (EC)). These ROIs were dilated by 2 mm to include surrounding voxels in the ventricles to scrutinize for possible partial-volume or motion-related false-positives in the CSF. The correlation coefficients were converted to normally distributed scores using Fisher’s r-to-z transformation. Individual maps were warped to the 1 mm MNI template and fed into the second-level analyses with a voxelwise t-test to compare functional connectivity between the LC and the ROI’s comparing encoding, consolidation, recollection with baseline. The second-level statistical analysis consisted of a non-parametric, permutation test with threshold free cluster enhancement; results were family-wise error corrected (p-value<0.05).

Coherence analysis

Request a detailed protocol

Given that autonomic influences in the HRV are known to be most prominent in specific frequency bands (Shaffer and Ginsberg, 2017), it is of interest to examine the correlation between fMRI and HRV measures per frequency. The magnitude squared coherence between timeseries can then be expressed as:

CA,B(f)=|PA,B(f)|2PA,A(f)2PB,B(f)2,

with PA,B(f)|2 being the cross power spectral density and Px,x(f)2 being the power spectral density for timeseries A and B. We estimated the coherence magnitude squared over 90 frequency bins with Hamming windows (20 % overlap) within the 0 - 0.5 Hz range for the LC and HRV, as well as LC and FC clusters. Similarly, the phase lag between the timeseries per frequency bin was estimated as:

θfMRI,HRV=tan-1(Re(PfMRI,HRV(f))Im(PfMRI,HRV(f))).

To obtain a robust single-point coherence metric, we estimated the median across the frequency band of maximum coherence. This was defined as the confidence interval (CI) around the median of the distribution of individual maximum-coherence frequencies between LC (or reference) and HRV at baseline (given that the frequency of maximum coherence between LC and HRV varies between individuals). These CIs were obtained for all timeseries and we extracted the median coherence and phase lags within the CIs. A coherence-magnitude null distribution was created by fitting an AR1 model for each timeseries, creating a randomized timeseries with a matched AR coefficient and calculating the respective coherences. These null-distributions are similar to the bootstrap and hence, the resampled timeseries have similar properties as the original data (distribution, autocorrelation) and have less bias in the rejection rates of the null-hypotheses.

Statistical analyses

Request a detailed protocol

Statistical analyses were performed using statistical software (R version 3.4.2). Group characteristics are presented in median and interquartile range. Differences between emotional and neutral trials in the behavioral data were tested with Wilcox signed-rank test or robust linear regression using the Huber-M estimator. Robust regression is a more conservative test compared to linear least-square regression methods, as the resulting models are stout against outliers. Changes in the autonomic measures (sAA and HRV) across the task stages (baseline, encoding, consolidation, recollection) were assessed with mixed effects linear regression using the maximum likelihood estimation with the autonomic measure as outcome measure, stage as predictor and a random intercept for each subject. The linear mixed effects model can account for missing data in longitudinal datasets. The most complex linear mixed effects models in this manuscript follows this formula:

Outcomeij=β1+β2PredictorAi+β3Taskstageij+[β4predictorAi×Taskstageij]+b1i+ϵi,
  • Outcome = outcome variable measured over time

  • Predictor A: variable of interest depending on the investigated model

  • Task stageij = Task stage (time)

  • b1i=random intercept for each subject

To obtain sAA measures over the same time frame as the HRV (rMSSD) measure, we calculated ΔsAA as: sAA end - sAA beginning for each task stage. To understand relationships between ΔsAA, HRV and LC activity across the task stages, we performed bootstrapped (n = 5000) repeated measures correlation.

For the frequency-specific effects of task stage on LC-HRV coherence, we entered the coherence magnitude squared into a linear mixed effects models with task stage, frequency and the interaction between task stage and frequency as fixed factors and participants as random intercept effect, using the maximum likelihood estimation. We performed simple slopes analyses to determine the region of significance of frequency for the fitted relationship between coherence and the task stages (where the 95% confidence interval did not include zero). The extracted point of maximum coherence (based on the defined confidence intervals) across task stages was correlated with ΔsAA using bootstrapped (n = 5000) repeated measures correlation.

For the fMRI analyses, we additionally correlated the extracted subject-specific mean LC-related beta-coefficients with emotional memory performance.

To determine which areas were lagging or leading in the spectral coherence analyses, we used the point of maximum coherence and extracted within the CIs the median coherence and phase lag per person. Two linear mixed models were fitted with median coherence or phase at maximum coherence as dependent variables and task stage, timeseries combination (LC, AMY, HIPP or EC) and the task stage by timeseries interaction as fixed effects and participants as random intercept factor. To examine that significant coherence between timeseries were independent of other timeseries of the network, partial coherences were calculated. For example, for three coherences between timeseries, we calculated the partial coherence in cases where significant coherence magnitude existed between region A – region B, region B and region C, and region A and region C and where the coherence phase of A < phase of B < phase of C. The spectral matrix between these timeseries was calculated and partial coherences within the CIs of median maximum coherence were extracted. The results were compared with Wilcoxon tests against a null distribution determined from the partial coherences of the matched-AR randomized timeseries. In addition, linear regressions were used to test associations between emotional memory performance and coherence at specific task stages per ROI pair. Coherence at baseline was entered as covariate.

Finally, within the structural LC analyses, we correlated the mean intensity of the LC from the MTR, MT-TFL, TFL and normalized TFL images to emotional memory performance. Similarly, Pearson correlations were also used to assess the relationship between the AES motion metric, normalized TFL mean intensity and emotional memory performance; and between LC intensity, ΔsAA, rMSSD and LC variance for each task stage.

All p-values were two-sided. Contrasts for interactions within the linear mixed effects models were computed using estimated marginal mean trends with reference grids and smooth functions and these results were corrected with the Tukey’s test. Separate models, including correlations and regressions, were corrected for multiple comparisons using the False Discovery rate (FDR)-approach.

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
    Facial expressions of emotion: an old controversy and new findings
    1. P Ekman
    (1992)
    Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences 335:63–69.
    https://doi.org/10.1098/rstb.1992.0008
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
    Are all measures created equal? heart rate variability and respiration - biomed 2009
    1. LK Hill
    2. A Siebenbrock
    (2009)
    Biomedical Sciences Instrumentation 45:71–76.
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
    Can PPG be used for HRV analysis?
    1. N Pinheiro
    2. R Couceiro
    3. J Henriques
    4. J Muehlsteff
    5. I Quintal
    6. L Goncalves
    7. P Carvalho
    (2016)
    Conference Proceedings : . Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual Conference. pp. 2945–2949.
    https://doi.org/10.1109/EMBC.2016.7591347
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
  61. 61
  62. 62
  63. 63
  64. 64
  65. 65
  66. 66
  67. 67
  68. 68
  69. 69
  70. 70
  71. 71
    Calculation of signal detection theory measures
    1. H Stanislaw
    2. N Todorov
    (1999)
    Behavior Research Methods, Instruments, & Computers 31:137–149.
    https://doi.org/10.3758/BF03207704
  72. 72
  73. 73
  74. 74
  75. 75
  76. 76
  77. 77
  78. 78
  79. 79
  80. 80
  81. 81
  82. 82

Decision letter

  1. Timothy E Behrens
    Senior Editor; University of Oxford, United Kingdom
  2. Dorothea Hämmerer
    Reviewing Editor; University College London, United Kingdom
  3. Mara Mather
    Reviewer; University of Southern California, United States

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

This paper presents a rich study investigating Locus Coeruleus (LC) function with a variety of MR and physiological measures. The study of LC function has recently gained more interest due to the LCs early decline in dementia. The authors show that structural and functional measures of the LC can be used to gain insight in its role in memory encoding and retrieval as well as related physiological processes.

Decision letter after peer review:

Thank you for submitting your article "Dynamic behavior of the locus coeruleus during arousal-related memory processing: a multi-modal 7T fMRI study" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Timothy Behrens as the Senior Editor. The following individuals involved in review of your submission have agreed to reveal their identity: Mara Mather (Reviewer #3).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

This paper reports on an impressive study which presents the state of the art in the type and quality of measures needed for investigating Locus Coeruleus (LC) function, also including MR measures developed by the authors themselves. A shortcoming of the paper is that the analysis methods used are not always sufficiently documented to enable an assessment of their precision, usefulness and validity. Especially with a structure so rarely investigated in fMRI, a study with such a nice state-of-the-art set-up should help us understand what aspects of the LC are possible to investigate with fMRI and which aren't. In this light, I would also think that already a convincing report on basic task-related fMRI responses in the LC can present a sufficient scientific advance. More advanced additional analyses might not be needed if their physiological plausibility proves unsatisfying.

Essential revisions:

Specifically, the following revision requirements are suggested, also based on the comments by the external reviewers:

Introduction:

1) The authors should be clearer in the Introduction as to how LC/NE effects might differ between different stages of memory processing.

Behavioral analyses on memory measures:

2) The bias measure hasn't been sufficiently described to be understandable. Also given difficulties in the task design for the name recognition memory part, it is suggested to just focus on Hits-FAs of the emotional face recognition task (not the name recognition task).

Preprocessing of fMRI data:

3) The authors mention comparing different preprocessing pipelines but (to my knowledge) don't show statistics on how these result in differences in the GLM results or SNR measures in their regions of interest. These should be added and are in particular critical as the authors chose to proceed with a denoising approach that does not include respiration and pulse regressors (maybe outline to what extent these might be captured even without explicit regressors in the chosen ICA denoising approach).

4) Precision in the spatial coregistration/normalisation is paramount for group analyses on such a small structure. The authors should provide supplementary information that allows to evaluate the precision of the spatial transformations for every individual fMRI dataset.

Physiological/functional measures:

5) it isn't clear why a relationship between a change in sAA across subtasks and task-specific measures of rMSSD is more relevant than a relationship between task-specific sAA levels and rMSSD measures.

6) the significance of the coherence analyses linking LC BOLD variability and HRV/sAA are unclear (what does a higher coherence between HRV and BOLD variability mean? Why does this have to be related to a change in sAA?). Please only leave this in /a subset of analyses in) if the analyses can be reasonably physiologically motivated. Also, for these analyses, exclusively pulse-corrected fMRI data should be used.

7) Please scrutinize the GLM results for circularity in the ROI selection.

8) It is not clear what variability in the coherence measures between e.g. LC and MTL indicates functionally. It might be cautioned against interpreting these for areas/task conditions where the task-related GLMs did not show significant activations?

Discussion:

9) The Discussion should be scrutinized for claims that exceed the physiological significance of the results.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Dynamic behavior of the locus coeruleus during arousal-related memory processing in a multi-modal 7T fMRI paradigm" for further consideration by eLife. Your revised article has been evaluated by Timothy Behrens (Senior Editor) and a Reviewing Editor.

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below:

We thank the authors for their work in revising this manuscript which has clarified several issues. There are still some concerns with the paper which we ask the authors to address:

1) As a general rule, we would like to ask the authors to go through the methods again and make sure that all analyses can be retraced by the reader. This concerns in particular the behavioral analyses. It is still not clear how the bias memory score was calculated. Please indicate which data were used for this (Hits-FAs of names related to which faces, old or new?) and give the precise formulas of all the calculations applied to these data.

2) Related to this, to add a more accessible measure of emotional memory effects (of names associated with faces), please add a table giving Hits, FAs, and Hits-FAs for names separately for old emotional and old neutral faces and separately for new emotional and new neutral new faces (incorrectly indicated as old (FA)), as well as statistics on the difference in Hit-FAs measures for names only on correct as old identified emotional versus neutral faces.

3) We ask the authors to consider limiting the analyses exploring links between LC resting state variability and (indirect) physiological indicators of NA (sAA and HRV) to correlations across task stages and analyses with HRF-convolved low and high HRV events, which are very interesting. By contrast, the coherence analyses in particular frequencies of resting state fMRI data with HRV seem not well motivated by the cited literature (which physiological processes do these high-frequency-specific fMRI data likely represent?). Alternatively, to taking these results out, the authors can revise their motivation for these analyses and make the interpretation of the results clearer. The reviewers found these hard to understand in an already complicated paper.

4) The significance of the exploratory coherence analyses, which report changes in the variability in coherence as their outcome measure seems to be also less clear in its physiological meaning (which processes that we can measure in fMRI are changing in coherence here and why should this be related to sAA?). Again, we invite the authors to remove these analyses. However, If the authors feel that this is adding valuable information, they are invited to leave these analyses in and make the interpretation of the results more accessible to the reader.

5) The results from the analyses on changes in coherence across task stages are not yet clear. I am wondering whether such analyses are prone to spurious results: if HIP-AMY coherence is higher in consolidation > baseline and in baseline > encoding, shouldn't it also be higher in consolidation > encoding? What does it mean if this effect is not picked up by the analyses? If the authors have easy access to the data, they may consider adding an analysis in a control area that shouldn't be implemented in consolidation or retrieval to provide cut-offs for avoiding spurious results or suggest alternative approaches to control for spurious results. Again, however, an alternative is to discuss this issue.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Dynamic behavior of the locus coeruleus during arousal-related memory processing in a multi-modal 7T fMRI paradigm" for further consideration by eLife. Your revised article has been evaluated by Timothy Behrens (Senior Editor) and a Reviewing Editor.

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below:

We thank the authors for implementing most of the requested changes. However, in their responses to Comment 1 and 2, they misunderstood or overlooked important aspects of the question. We would like to ask them to adjust their responses and changes in the paper as follows:

Re Comment 1: Please also indicate whether the names used for the calculation of the bias score were paired with new or old faces or whether this was collapsed across both old and new faces.

Re Comment 2: There was a misunderstanding in the response to comment 2. The statistics required in the table were the hit and false alarm rate for names, not for faces. Please report the Hit rate, FA rate and Hit-FA rate separately for the names paired in the memory tests with old emotional and old neutral faces and for those paired with new emotional and new neutral new faces.

https://doi.org/10.7554/eLife.52059.sa1

Author response

Essential revisions:

Specifically, the following revision requirements are suggested, also based on the comments by the external reviewers:

Introduction:

1) The authors should be clearer in the Introduction as to how LC/NE effects might differ between different stages of memory processing.

We thank the reviewer for this excellent remark. Several animal studies have exactly tried to answer the question to what extent the LC plays a role in the different stages of memory processing. The animal literature is not consistent as different methods have been used (pharmacological manipulations, stimulation or lesion studies), and measurements were done at different time points along the task at hand.

With respect to encoding, studies injecting propranolol, a β-adrenoreceptor antagonist, prior to inhibition avoidance learning observed no effect on subsequent recognition tests [1-3]. While this may suggest that β-adrenoreceptors are not involved in learning, others have demonstrated that blocking β-adrenoreceptors prior to learning does affect long-term potentiation [4]. Possible explanations for these inconsistencies may be that activation of the LC induced release of dopamine [2], leading to (less strong) long-term potentiation. The timing and intensity of locus coeruleus activation may be critical here. The amplitude as well as the duration of norepinephrine release in the amygdala of rats after inhibitory avoidance training determined their subsequent memory enhancement effect [5].

It has been suggested that encoding of arousing information shifts norepinephrine neurons from tonic into phasic activity [6, 7]. This phasic signaling would sustain into and support consolidation. Studies infusing propranolol immediately after training in the basolateral amygdala demonstrated impaired recognition performance [1, 7-9], indicating that activation of the adrenergic system plays an important role in consolidation and neuronal plasticity. Interestingly, Khakpour-Taleghani and colleagues demonstrated that these latter effects may be time-dependent, as inactivation of the LC 90 or 360 minutes after learning had no effect on subsequent memory retrieval and suggests that later phases of consolidation may be dependent on distinct molecular processes.

Adrenergic signaling effects may also play a time-limited role in hippocampus-dependent memory retrieval. Stimulating the LC prior to the retrieval test enhanced [10] performance and inactivation of the LC 5 minutes before the retention test impaired memory performance. Retrieval could become independent of norepinephrine when the hippocampus is no longer playing a critical role in the consolidation process. This may explain why animals with very little norepinephrine in the forebrain are still able to perform well on retention tasks [11].

While outside the purpose of the current manuscript, it is important to note that other neuromodulators may affect memory formation by influencing norepinephrine neurons, such as for example glutamate, glucocorticoid, GABA, acetylcholine and estrogens.

In the Introduction:

“Research into NE-modulation of emotional memory enhancement has focused largely on the amygdala[6-8], an important target region of the LC and a critical nexus in memory for arousing experiences. Rodent work consistently demonstrated that blocking β-adrenergic receptors in the basolateral amygdala (BLA) immediately after learning in arousing conditions impairs memory, while infusion of NE or β-adrenoreceptor agonists enhances consolidation and rescinds the impairment, evaluated one to two days later [9, 10]. Correspondingly, stimulation of the LC during the learning phase modulates activity in the BLA and synaptic plasticity in the CA1, CA3 and dentate gyrus [11, 12]. While these studies highlight the importance of adrenergic signaling for plasticity and hippocampal-dependent learning, blocking β-adrenoreceptor prior to learning had no effect on subsequent retrieval 24 hours later [1, 2]. It has been suggested that in the context of strong LC stimulation tonic activity can be shifted into phasic activity and other neuromodulators may then elicit similar but possibly weaker plastic processes [4, 6, 7].”

In the Introduction:

“These network interactions can vary over time or be task-dependent. For example, the LC is involved in consolidation, but within a critical time period after learning, the time required for the hippocampus to facilitate the reorganization and stabilization of storing information in the neocortex [15].”

Behavioral analyses on memory measures:

2) The bias measure hasn't been sufficiently described to be understandable. Also given difficulties in the task design for the name recognition memory part, it is suggested to just focus on Hits-FAs of the emotional face recognition task (not the name recognition task).

We apologize for not being clear in our methodology and explaining our setup. We will clarify this here and in the manuscript.

Concerning the bias measure: The bias measure originates from signal detection theory, often used in psychology [12]. The bias provides a metric that indicates the strategy of the participant, i.e. whether there is a bias towards a liberal answer (for example indicating that he or she saw the face previously) or a more conservative approach (for example answering that the face had not been seen previously). The advantage of using the likelihood ratio of the Beta is that it allows to interpret the values within a statistical framework. As Figure 2A-B (t=2.493, p<0.019) shows, there was a positive bias for the emotional faces compared to the neutral faces, hence subjects were more likely to indicate that they had seen the face earlier for the emotional condition than for the neutral condition. This bias will impact the retrieval scores, as participants will then receive more emotional trials during the retrieval condition relative to neutral trials. This is exactly what our data shows: the bias reduced retrieval performance (the participants falsely identified the faces as seen before) indicating that this was indeed a bias and not simply better recognition. This bias makes it hard to disentangle true recognition from false recognition, but the setup of our two-stage memory task and the application of signal detection theory, allow us to correct for this bias in the retrieval stage. Correcting for this bias cleaned up the retrieval scores (see Figure 2F in the manuscript). We believe that this will also be valuable to other studies with a similar approach.

Concerning our task design: We also would like to clarify the set-up of the task and the terminology that we used for differentiating between the memory processing stages. During encoding, participants view emotional and neutral faces coupled to a Dutch name. The encoding phase is followed by a resting-state period, which we term “consolidation”. Then, we conceptually distinguish recognition from retrieval. During recognition, we ask participants whether they believe they have seen the face previously (the face is shown without any names). If they endorse this question, they will be asked which name was associated to the face during the encoding phase. The latter is in our manuscript referred to as retrieval. Recognition or familiarity is a fast process that does not necessarily involve active retrieval of details surrounding the face [13]. We do recognize that the word “retrieval” may have been confusing, as in our task there is still no involvement of free or cued recall. Therefore, we have clarified in the manuscript that this process is more similar to recollection, the retrieval of qualitative information about a face-name pair. We should note that the participants had already been exposed to all names during encoding, so that this recollection step does not consist of simple name recognition. To avoid a bias in name recollection, all the incorrect (false positive) names during recollection were mixed in terms of valence.

Concerning the choice of recognition versus recollection: Recognition (familiarity) and recollection are two functionally independent processes and are also guided by two distinct brain systems. Familiarity is associated with lateral prefrontal and parietal regions, supporting processing associated with identifying the stimuli. On the other hand, recollection involves medial prefrontal, medio-lateral parietal and hippocampal regions. The hippocampus has been related to familiarity processes under high confidence ratings, such as recollection, given that recollection relies on mechanisms related to pattern separation and pattern completion [13-16]. See also the below additional fMRI GLM analyses for recognition. As expected, we do see activity in the fusiform gyrus (successful recognition) and lateral prefrontal regions (successful emotional recognition), but not for the hippocampus.

Data of the recognition part of the task: Following the advice of the reviewer, we examined the hit and false alarm rates of the face recognition part of the task. There was no significant difference (Wilcoxon paired test: Z = -1.24, p-value=0.222) between the recognition hit rate for emotional (Median = 53.33% (IQR=45.56 – 61.11)) vs neutral (Median = 53.33% (IQR = 44.44-67.79); Figure 1A). In accordance with our report of a bias during face-recognition, a significant increase of false alarms was detected (Wilcoxon paired test: Z=3.86, p-value<0.001) for emotional (Median = 22.73% (IQR = 18.19-34.09) compared to neutral faces (Median = 13.63% (IQR = 9.09- 22.78); Figure 2B).

Brain activation patterns related to recognition: A general linear model was fit at each voxel with FMRI Expert Analysis Tool (FEAT), using generalized least squares with a voxel-wise, temporally and spatially regularized autocorrelation model over the whole brain. A high-pass filter (cutoff at 100 s) and minimal spatial smoothing (kernel = 1.5 mm) were applied before the fit. The regressors consisted of the emotional trials that were later successfully recognized, the neutral trials that were later successfully recognized, the emotional trials that were not recognized and the neutral trials that were not recognized. Contrasts of interests were emotional trials vs. neutral trials (Emotional > Neutral), successfully recognized trials vs. not successfully recognized (Successful recognition > Not successful recognition) and the interaction (Emotional successful recognition > neutral successful recognition).

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; Figure 6—figure supplement 3). 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 [57,29,12], cluster volume = 191 mm). No clusters were visible in the amygdala, hippocampus, entorhinal cortex or locus coeruleus for the above contrasts.

In the Results:

“This was followed with a recollection fMRI-scan, during which participants had to indicate whether they recognized the face and, upon endorsement, had to select the name of the face out of three options (all options were seen previously to stimulate recollection instead of recognition processes).”

In the Results:

“There was no significant difference (Wilcoxon paired test: Z = -1.24, p-value=0.222) between the recognition hit rate for emotional (Median = 53.33% (IQR=45.56 – 61.11)) compared to neutral faces (Median = 53.33% (IQR = 44.44-67.79). However, more false alarms were detected (Wilcoxon paired test: Z=3.86, p-value<0.001) for emotional (Median = 22.73% (IQR = 18.19-34.09) compared to neutral faces (Median = 13.63% (IQR = 9.09- 22.78); Figure 2B).”

In the Results:

“We decided to focus on recollection because of this response bias, our focus on the medial temporal lobe regions and the notion that recollection is more likely to elicit hippocampal activity than recognition.”

In the Materials and methods:

“This response bias was derived from the signal detection theory, a method that allows us to differentiate the signal from the noise. The bias measure indicates the extent to which a certain type of response is more probable than another, such as for example responding in a more conservative manner. The bias was regressed out from the emotional and neutral recollected scores and using their residuals this resulted in an adjusted emotional memory performance score.”

In the Materials and methods:

“Contrasts related to recognition were not our primary interest, however, we fit encoding regressors for emotional trials that were later successfully recognized, the neutral trials that were later successfully recognized, the emotional trials that were not recognized and the neutral trials that were not recognized. Contrasts of interests were emotional trials vs. neutral trials (Emotional > Neutral), successfully recognized trials vs. not successfully recognized (Successful recognition > Not successful recognition) and the interaction (Emotional successful recognition > neutral successful recognition). Results are presented in Figure 6—figure supplement 3.”

Figure 2 was updated. We also edited all the figures, tables, supplementary data and text replacing retrieval by recollection to avoid any confusion among the readers. We added the fMRI results for the recognition to the supplemental data. We also update the data file for this figure.

Preprocessing of fMRI data:

3) The authors mention comparing different preprocessing pipelines but (to my knowledge) don't show statistics on how these result in differences in the GLM results or SNR measures in their regions of interest. These should be added and are in particular critical as the authors chose to proceed with a denoising approach that does not include respiration and pulse regressors (maybe outline to what extent these might be captured even without explicit regressors in the chosen ICA denoising approach).

We agree with the reviewer that this is a crucial point for fMRI experiments in general and brainstem fMRI in particular. Overall, as described in our methods the approach we followed to mitigate the chance that our results are due to unmodeled physiological noise was to a) repeat the analysis across pipelines and b) repeat the analysis with a reference component that is adjacent to the LC. We further collected respiration and heart-timeseries and included them in a denoising scheme (using FSL’s PNM) along with a model-free physiological denoising based on FSL-FIX. Crucially, this was combined within a single regression to ensure that noise is not reintroduced in the data. Overall, when we examined the impact of our denoising pipelines on tSNR in the baseline resting state (Figure 2) we found that including FIX provided a sizable gain in tSNR, while including explicit physiological time series provided a more 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 (while the number of volumes where only 180 e.g. in the resting stage scans). 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 and FIXed +Explicit Resp and 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. Based on the above, we were not convinced that the explicit physiological regression provided a robust improvement over the implicit one (FIXed) (especially when considering that regressing-out even not very-meaningful regressors tends to reduce variance and thus increase the tSNR) and we chose to focus our report in the FIXed dataset.

The reason that the gain from the inclusion of physiological regressors is modest is that a) FSL-FIX already regresses out physiological-noise components b) respiration and heart-time series tend to get corrupted periods over lengthy scan sessions. Even so, we repeated the analysis across the pipelines to ensure the robustness of our results. Overall results between FIXed and the FIXed +Explicit Resp/Phys were largely consistent. Note that we changed the names of our pipelines as suggested by the reviewers in comment 18. See Figure 1—figure supplement 2, Figure 4—figure supplement 1, Figure 4—figure supplement 2 and Figure 5—figure supplement 1 for details.

LC variability - rMSSD relationship across preprocessing steps:

We performed the repeated measures correlations between LC BOLD variability and rMSSD across all preprocessing steps. Comparisons across pipelines show similar effects in the latter two pipelines (Figure 4—figure supplement 2B and C) as in the FIXed pipeline reported in our manuscript (Rrm=-0.387; p = 0.018).

Locus coeruleus - HRV coherence during task across preprocessing steps:

We also performed the coherence analysis between LC and HRV across all preprocessing steps.

Spatially normalized:

We ran linear mixed effect models to examine whether the relationship between frequency and coherence between the LC and heart rate variability varied across task stages. See supplementary file 4 and Figure 5—figure supplement 2.

FIXed:

We also ran linear mixed effect models to examine whether the relationship between frequency and coherence between the LC and heart rate variability varied across task stages. See Supplementary file 5 and Figure 5—figure supplement 3.

FIXed +Explicit Resp:

For the FIXed +Explicit Resp, we also ran linear mixed effect models to examine whether the relationship between frequency and coherence between the LC and heart rate variability varied across task stages. See Supplementary file 6 and Figure 5.

FIXed +Explicit Phys:

For the FIXed +Explicit Phys, we ran examined whether the relationship between frequency and coherence between the LC and heart rate variability varied across task stages using linear mixed effects models. See Supplementary file 7.

Overall, checking the reproducibility of our results showed that across all preprocessing pipelines – except the spatially normalized pipeline – the coherence between LC and HRV is greater during baseline compared to consolidation, during recollection compared to baseline, during recollection compared to encoding and during encoding compared to consolidation. These results are similar to the results reported in the original manuscript for the FIXed-pipeline.

Functional connectivity (FC) from the LC component to MTL structures across task stages.

We performed seed-to-voxel analyses to explore the cross-correlation at lag = 0 s (denoted as FC) between the LC (consisting of the LC component) and medial temporal lobe structures (set in advance to include hippocampus (HIP), amygdala (AMY) and entorhinal cortex (EC)). This revealed a significant decrease of FC between LC and left BLA, subiculum and entorhinal cortex across preprocessing pipelines, except for the spatially normalized data where no significant cluster-wise corrected changes in FC were detected. See Supplementary file 8 and Figure 7—figure supplement 1.

In the Results:

“To assess the specificity of our findings to the LC, we also investigated a control component in the pons (reference ROI). In addition, sensitivity analyses were done by evaluating the outcomes across four different preprocessing pipelines, each time adding in more physiological noise correction (Figure 4—figure supplement 1).”

In the supplemental data:

We added the results of each pipeline to the supplemental data, as well as the tSNR comparison across the pipelines. Following comment 6 of the reviewers, we used the FIX+PNM pipeline for the coherence analyses (instead of the FIX only pipeline).

The data file associated with this figure was updated.

4) Precision in the spatial coregistration/normalisation is paramount for group analyses on such a small structure. The authors should provide supplementary information that allows to evaluate the precision of the spatial transformations for every individual fMRI dataset.

We are happy to provide more information on the spatial coregistration. For the fMRI scans, it was crucial that quality registrations to the MNI template were obtained to facilitate integration with previous results and existing atlases. This is challenging with gradient-echo scans when aiming to image such a small region at high resolution and given the required coverage and SNR for memory experiments. Gradient-echo scans are inherently sensitive to dephasing and thus distortion in the phase-encoding direction. To tackle this, we combined careful placement of the field of view (so that a) the quickly-dephasing sinuses in the frontal and temporal lobes are largely excluded and b) the phase-encoding gradients are applied orthogonally to the main axis of the hippocampus and 45 degrees to the main axis of locus coeruleus so that their displacement is minimized) with parallel MRI (so that the slice thickness and thus the inclusion of quickly-dephasing regions within the relevant slices would be reduced). Having acquired scans that showed reduced distortions in the relevant ROIs, we calculated a warp field to minimize distortion in the rest of the brain and facilitate registration using FSL TOPUP.

The T1-weighted scans were preprocessed prior to the Freesurfer pipeline and subsequent registrations: typical MP2RAGE scans, while high-resolution and theoretically bias-corrected, show background noise outside the brain. To minimize this, the scans were recombined following the process described in [17] so that the background noise was suppressed. Then the scans were bias field corrected for remaining inhomogeneities using an N4 algorithm and brain extraction was performed. Freesurfer was performed using our own brain mask and both the T1-weighted image and T1-map (that was used as a substitute for a T2-weighted image) to improve white matter and grey matter delineation. The Freesurfer segmentations were then used as input to estimate a boundary-based registration transform from the fMRI images to the T1-weighted images using FSL. Subsequently, the N4-corrected T1-weighted scans were used to calculate diffeomorphic transforms to the MNI template with ANTs (symmetric normalization transformation model; SyN; antsRegistrationSyN.sh). The SyN algorithm has been shown to provide superior performance compared to most other registration algorithms used in neuroimaging [18, 19]. The linear transformations were transformed to the ITK-SNAP reference using c3d tools and combined with the warps, so that when transforming the data from native fMRI to MNI there was a single interpolation. The accuracy of the transforms was individually checked by NP.

To allow reviewers to examine the registrations, we are showing the mid-sagittal, coronal and axial slice of the MNI-warped T1-weighted and fMRI scans for every individual (Author response images 1-4). The quality of the registrations can be visually assessed. Furthermore, we created an average of all the individual T1-weighted and fMRI scans warped to the MNI space (Figure 1—figure supplement 1). The fact that small but relevant features of the hippocampus and brainstem can be clearly observed in both averages support our co-localization across group locus coeruleus and medial temporal lobe activation. Finally, we attempted to provide a quantitative metric of the quality of registrations between individuals. Typically, this is done with metrics of volume overlap or distance in previously-labelled datasets [19]. This is labour intensive to do manually, so we opted to use the automatic labelling of the Freesurfer pipeline. We created individual ROIs consisting of the Freesurfer’s segmentation of hippocampus and brainstem, as the most relevant anatomical features [20, 21]. The ROIs were projected to the MNI space. Then for each individual ROI A, we calculated a measure of volumetric overlap versus every other volumetric ROI B, such as overlap ratio = ((A)U(B))/(A). The distribution of the overlap ratio per individual is shown in Author response image 5. The overlap ratio across the group was high (median=0.93) with a very tight spread (IQR=[0.92-0.94]) indicating even registration performance across the group. We should note that this process likely underestimates the registration quality, since it relies on the automated segmentation of Freesurfer.

Finally, we would like to draw the attention of the reviewers to Figure 8 in the manuscript, where the locus coeruleus MT-weighted template is shown. Note that despite the very small size of the locus coeruleus (2mm wide), it can easily and specifically be discerned at group level, further attesting to our capacity to image the nucleus across group.

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.

In the Materials and methods:

“The T1 to MNI template registrations were calculated using a diffeomorphic transform with the antsRegistrationSyN.sh function (ANTS 2.1 (http://stnava.github.io/ANTs/) [18]). This algorithm has been shown to provide superior performance compared to most other registration algorithms used in neuroimaging [18, 19].”

In the Materials and methods:

“A boundary-based registration was calculated for the linear registration between the fMRI and T1data using FSL's epi_reg based on the FreeSurfer segmentations. The T1 to MNI template registrations were done using ANTS as described in the preprocessing of the structural data. The linear transformations were transformed to the ITK-SNAP reference using c3d tools and combined with the warps, so that when transforming the data from native fMRI to MNI there was a single interpolation. The accuracy of the transforms was individually checked by NP.”

Given that ELife also publishes the response letter, we decided to not overwhelm the reader with supplemental data. If this paper would be accepted for publication, the reader will be able review the individual images in this response letter. We added the T1 and fMRI average image to the supplemental data and made reference to the individual images in the response letter.

Physiological/functional measures:

5) it isn't clear why a relationship between a change in sAA across subtasks and task-specific measures of rMSSD is more relevant than a relationship between task-specific sAA levels and rMSSD measures.

We performed both: we examined changes in sAA and rMSSD across the different stages of the memory task using linear mixed effects models (Figure 3A and 3B in the manuscript) and we examined person-specific relationships between two measures along the entire setup using repeated measures correlations (Figure 3D in the manuscript). We don’t think that one set of analyses is more relevant than the other, but that they provide complementary information. The repeated measures correlation is similar to a linear mixed effects model, but allows for a time-varying predictor and time-varying outcome, with each having the same amount of measurements. Thus, in the first set of analyses, we examined whether one metric changes during specific parts of memory processing, whereas in the second set we explore whether the changes in one measure go together with changes in the other. Based on the physiology of the autonomic system and the homeostatic principles of the brain, we assume that sAA and rMSSD covary consistently across stages. We explained this in the Results section.

We like to point out that to enable the comparison of both measures along similar time points, we subtracted the sAA measures pre and post the condition (see Figure 1). For the rMSSD we calculated the variability during that time frame. Our data indicates that sAA during consolidation (note: this represents the differences between post-encoding versus post-consolidation) was elevated as compared to baseline (difference between pre-baseline and post-baseline). We did not see differences in rMSSD across the task stages. The fact that both sAA and rMSSD are correlated across the stages suggests that they are both measuring a similar biological construct and provides extra validity to our measures.

In the Results section:

“To facilitate analyzing rMSSD and sAA at similar time points along the paradigm, we calculated the difference in sAA (ΔsAA) levels before and after each of the stages of the fMRI task …”

In the Results section:

“Given that both measures relate to the autonomic system, we then correlated ΔsAA to RMSSD across the stages using bootstrapped (n=5000) repeated measures correlations”.

6) the significance of the coherence analyses linking LC BOLD variability and HRV/sAA are unclear (what does a higher coherence between HRV and BOLD variability mean? Why does this have to be related to a change in sAA?). Please only leave this in /a subset of analyses in) if the analyses can be reasonably physiologically motivated. Also, for these analyses, exclusively pulse-corrected fMRI data should be used.

We thank the reviewers for the opportunity to further demonstrate the importance of these analyses. The locus coeruleus neurons mediate arousal and autonomic activity. Autonomic activity is an intrinsic part of arousal or emotional experiences during a task. We already observed an overall correlation between locus coeruleus BOLD variability and HRV across the task stages (Figure 4 in the manuscript). But by combining our physiological data in a frequency-resolved manner with the LC BOLD data, we were able to examine the nature and contributions of autonomic regulations on LC neural activity across the different task stages.

The low frequency component (LF, 0.05–0.15Hz) is thought to be influenced by, though not entirely, sympathetic activity, while the high frequency component (HF, 0.15–0.40Hz) is influenced by parasympathetic, vagally mediated activity. Recent work demonstrated a frequency-specific relationship between the HF-HRV and brainstem activity [22, 23], which was interpreted to reflect the feedback loop between brainstem nuclei and parasympathetic system, which is expected to have a specific frequency profile (i.e. it is reflected in the high-frequency of HRV).

The coherence measure between HRV and LC BOLD is a measure of the magnitude of coupling between LC BOLD and HRV measures at specific frequencies.

The greater coherence between LC BOLD and HRV during recollection compared to baseline or encoding, indicates that as LC activity increases, the inhibition on the heart increases. This may reflect the feedback loop between cortical regions and the LC allowing for bidirectional communication between central memory areas and peripheral sites. This action allows conscious experience of the stimuli through top-down inhibitory influences on the LC, which shapes our arousal experiences and ultimately facilitates the appropriate response. Higher HF-HRV has been associated with better discrimination between emotional versus neutral faces, attentional resources and better recollection of context information from stored representation to perform memory task successfully [24, 25]. On the other hand, consolidation demonstrated lower coherence compared to encoding or baseline, which is consistent with the idea that lower HF-HRV is associated with higher arousal. The maximum coherence measure – mainly reflecting parasympathetic contributions – was negatively related to sAA, demonstrating that higher arousal-related metrics (lower coherence) is associated with greater sympathetic activity. The fact that our previous analyses demonstrated that sAA increased during consolidation and that this sAA increase related to memory performance confirms that consolidation is more likely to be influenced by sympathetic activity than parasympathetic activity. Such sympathetic influences may potentially reflect norepinephrine-related long-term potentiation.

Our sensitivity analyses, i.e. reproducing the same results using different processing pipelines highlight the robustness of these findings and exclude confounding by respiration.

We now clarified to the reader the rationale as well as the interpretation of these analyses in the Results section. We also added the results of the other pipelines to the supplementary data.

In the Results section::

“In the previous analyses we demonstrated a negative correlation between LC variability and rMSSD or ΔsAA across the entire paradigm. These results echo previous work demonstrating that blocking β-adrenoreceptor increases HRV [26]. However, autonomic fluctuations can alter the peak of frequencies in the HRV leading to complex patterns of variability. To understand the nature of autonomic regulations on LC activity at each task stage, we applied a time-frequency analysis by calculating the coherence magnitude squared and phase lag between the timeseries of the HRV and LC for each task stage (see example in Figure 5A). By using a within-subject design with a baseline measure we can eliminate interindividual differences in blood pressure, respiration and, by combining HRV fluctuations with other measures related to the autonomic system, we are well-positioned to differentiate autonomic contributions to LC activity due to emotion or arousal across the different memory stages. Here, we will use the FIXed+explicit Phys pipeline to remove respiratory confounding.”

In the Results section:

“These patterns indicate greater LC activity coupled to greater parasympathetic inhibition during baseline and recollection, which has been associated with better discrimination between emotional versus neutral faces, attentional resources and better recollection of context information from stored representation to perform memory task successfully. On the other hand, consolidation demonstrated lower coherence compared to encoding or baseline, which is consistent with the premise that lower HF-HRV is associated with higher arousal.”

In the Results section:

“The covariation between LC activity and HRV demonstrate that even though both emotion and cognition operate simultaneously, encoding and recollection are coupled to autonomic regulation supporting cognitive functions required for the task at hand. However, this LC-parasympathetic coupling is not sufficient to predict emotional memory performance. Conversely, LC BOLD signal during consolidation may be specifically tied to arousal-related sympathetic tone, an inference supported by our ΔsAA analyses and the at-trend correlation between ΔsAA during consolidation and memory performance.”

Figure 5: was updated to show the data with the FIXed+Explicit Phys pipeline

Supplemental data We added the results for these analyses with the other pipelines to the supplemental data.

7) Please scrutinize the GLM results for circularity in the ROI selection.

The ICA component of the locus coeruleus was identified in a separate resting-state period, that preceded the task (see Figure 4 —figure supplement 1 or Figure 4 in this response letter to comment 3). For the GLM results, we created a mask covering all regions of interest (the entire pons, medial temporal lobe as well as the 4th ventricle – the latter to allow for detection of motion related fMRI signal). The ICA component was used in the analyses leading up to Figure 4 and 5 but were not used in the GLM analyses. All GLM analyses were done voxel-based and hence, there was no circularity.

We added a figure to the supplement demonstrating the mask with the regions of interest for the GLM and functional connectivity analyses.

8) It is not clear what variability in the coherence measures between e.g. LC and MTL indicates functionally. It might be cautioned against interpreting these for areas/task conditions where the task-related GLMs did not show significant activations?

Coherence is a mathematical technique that quantifies the frequency and amplitude of the synchronicity of brain activity in different regions. It estimates the consistency of the amplitude and phase between both BOLD signals over time within a specific frequency band. The amplitude provides information on the strength of the coupling or connectivity between two regions. The phase provides information on their temporal relationship. These analyses are important as arousal and salience processing in the brain, typically involving contributions of the LC, exhibit variable behavior, as we demonstrate in the relationship between LC and the HRV (comment 6). In static fMRI analyses, such as for typical functional connectivity (Figure 7A in the manuscript), this is ignored, rendering them hard to interpret, given that they can dilute specific sources of information. However, the brain can show dynamic changes not only within the frequency domain but over time as well. Previous research suggested that attention-related variability of connectivity in the time-frequency domain explains anti-correlated functional connectivity [27]. The wavelet coherence is a method to disentangle differences in the strength of the coupling between regions across task stages in a time-resolved manner. In our study, we found anti-correlated functional connectivity between the LC and the MTL. Based on the coherence results, we suggest that this finding potentially relates to dynamic functional connectivity between the LC and the MTL, as demonstrated by the variability in the wavelet coherence measure (i.e. this reflects the extent to which the coherence between regions can transition across time at a given frequency).

This analysis provides complementary information to the static functional connectivity analyses and therefore, we decided to examine all regions of interest and applied an adjustment for multiple comparisons. We clarified the premise for these analyses and the interpretation in the Results section.

In the Results section:

“Areas related to arousal and salience processing, such as the LC, have particularly variable behavior that can result in a negative FC value, complicating their interpretation [28].”

In the Results section:

“Integrating these findings with our LC-HRV coherence analyses (Figure 5) reveals correspondence for the increase in coherence variability during consolidation for all LC-MTL pairs and suggests that these time-varying behaviors in the dynamics between regions reflect changes in arousal and possibly sympathetic tone.”

Discussion:

9) The Discussion should be scrutinized for claims that exceed the physiological significance of the results.

We ensured that our Discussion is reflective of our findings and whenever we speculated on the interpretations, we made that clear as well.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

We thank the authors for their work in revising this manuscript which has clarified several issues. There are still some concerns with the paper which we ask the authors to address:

1) As a general rule, we would like to ask the authors to go through the methods again and make sure that all analyses can be retraced by the reader. This concerns in particular the behavioral analyses. It is still not clear how the bias memory score was calculated. Please indicate which data were used for this (Hits-FAs of names related to which faces, old or new?) and give the precise formulas of all the calculations applied to these data.

We are happy to provide more information on our methods. We went through our Materials and methods section and added more information. As specifically requested by the reviewers, we have also provided the detailed methods relating to the bias scores and which data was used. To make this clearer, we have now also added a figure that explains the procedure to calculate the adjusted emotional memory performance score.

In the Materials and methods section:

“We calculated hits and false alarms for both the emotional and neutral condition during encoding. For the recollection (identification of the name), we calculated hit and false alarm rates for both the neutral and emotional condition. Reaction times were also analyzed. Performance in the recollection condition may be affected by a response bias in the recognition phase between emotional and neutral stimuli. Therefore, we calculated the likelihood ratio beta, as a measure of the response bias (beta = exp (d' x c)) (d' index calculated as by Stanislaw and Todorov, but corrected for ceiling effects [29]). This response bias was derived from the signal detection theory, a method that allows us to differentiate the signal from the noise [12]. The bias measure indicates the extent to which a certain type of response is more probable than another, such as for example responding in a more conservative manner.

We performed a paired t-test to compare the bias between emotional and neutral during the recognition phase. This bias was then regressed our from the emotional and neutral recollected hits in two separate regressions (Figure 2). The residuals of each regression represented the bias-corrected emotional and neutral recollection rate. Finally, to obtain an individual emotional recollection rate, we calculated an adjusted emotional memory performance score = (Emotional recollection rate)/(Emotional recollection rate+Neutral recollection rate) (Figure 2—figure supplement 1). This measure was used throughout the manuscript.”

2) Related to this, to add a more accessible measure of emotional memory effects (of names associated with faces), please add a table giving Hits, FAs, and Hits-FAs for names separately for old emotional and old neutral faces and separately for new emotional and new neutral new faces (incorrectly indicated as old (FA)), as well as statistics on the difference in Hit-FAs measures for names only on correct as old identified emotional versus neutral faces.

This is an excellent suggestion and we have now also added in a Table in the Supplementary files with the median and IQR for Hits, False Alarms, Hit-False Alarms, Bias scores and adjusted memory scores for each condition and provided statistical comparisons between emotional and neutral scores. The table contains the medians and interquartile range for the recognition hit rate for old faces only, the correct rejection rate of new faces during recognition, the false alarm rate during recognition (i.e. incorrectly identifying new faces as old) and the difference between the recognition hit rate and the false alarm. Furthermore, we also included the raw recollection hit rate of the raw data (i.e. correct selection of name over the total number of old faces), the recollection hit rate following correct recognition and the recollection hit rate corrected for the recognition response bias. The normalized emotional the recollection hit rate corrected for the recognition response bias was used for all analyses in the rest of the manuscript). While making the table, we also noticed we had reported the Wilcoxon’s W statistic instead of the z-value for the bias-corrected recollection differences. For consistency, we now report the z-value for all comparisons.

In the Results:

“…participants recollected on average more names with an emotional (median=38.68, IQR=36.68, 39.46) relative to those with a neutral valence (median=29.97, IQR=28.01, 33.39; paired Wilcoxon test: Z=4.41, p-value<0.001, Figure 2G),…”

3) We ask the authors to consider limiting the analyses exploring links between LC resting state variability and (indirect) physiological indicators of NA (sAA and HRV) to correlations across task stages and analyses with HRF-convolved low and high HRV events, which are very interesting. By contrast, the coherence analyses in particular frequencies of resting state fMRI data with HRV seem not well motivated by the cited literature (which physiological processes do these high-frequency-specific fMRI data likely represent?). Alternatively, to taking these results out, the authors can revise their motivation for these analyses and make the interpretation of the results clearer. The reviewers found these hard to understand in an already complicated paper.

We thank the reviewers and editors for an opportunity to clarify the importance of these analyses. Within this manuscript we aimed to examine the role of the LC during arousal-related memory across the various stages. We found a negative correlation across task stages between sAA and rMSSD, a metric of heart rate variability (HRV). The rMSSD is a frequency non-specific measure, though the parasympathetic modulation of HRV is known to be frequency-specific (0.15 to 0.4Hz). We therefore explored which HRV frequency band has the largest effect on the LC fMRI signal. Remarkably, we found that the confidence interval of the individual maximum coherence between LC and HRV at baseline overlaps with the known frequency band of parasympathetic modulation of HRV. Furthermore, coherence within that frequency band negatively correlated with sAA levels. Given that the LC plays a crucial role in both autonomic processes and arousal, these coherence analyses help us to disentangle contributions of emotion or arousal to LC activity during the various stages.

Using the coherence method is attractive because it provided us with a single-point metric of LC-related autonomic modulation across memory task stages. The power of the low and high frequencies are not fixed and can vary in response to autonomic changes [26, 30]. This analyses was motivated by the fact that higher arousal due to emotionally salient stimuli can shift the low and high frequency states of the HRV states and occurs in concert with changes in BOLD responses in brain regions involved in emotional processing [31, 24]. Regarding the high-frequency of LC fMRI signal, we were primarily focused on the frequency-specificity of the HRV and while brainstem regions seem to show more connectivity at higher frequency as compared to the often lower-frequency networks observed in the cortex, we agree that the real physiological relevance of high-frequency fMRI is still unclear.

We have clarified the rationale of these analyses in the text, provided additional literature references, rephrased the interpretation, and have also briefly discussed the fact that the neurophysiological basis of high-frequency fMRI signals remains unclear.

In the Results:

“However, rMSSD reflects beat-to-beat variance in HR in the time-domain and is relatively non-specific to the frequency-domain. Autonomic fluctuations due to emotionally salient stimuli can alter the power of frequencies in the HRV and this can occur in concert with changes in LC BOLD responses [26, 30, 24]. Therefore, we applied spectral analyses to assess the nature of autonomic modulations on LC activity at each task stage by calculating the coherence magnitude squared between the timeseries of the HRV and LC for each task stage (see example in Figure 5A).”

“The covariation between LC activity and HRV demonstrate that even though both emotion and cognition operate simultaneously, encoding and recollection are coupled to moment-to-moment autonomic regulation supporting cognitive functions required for the task at hand. However, this LC-parasympathetic coupling, appertaining to the central autonomic network, is not sufficient to predict accurate emotional memory recollection.”

“We should note that the physiological relevance of high-frequency fMRI signal remains unclear, but it has indeed been linked to rapid modulation of the brainstem driving information exchange [32].”

4) The significance of the exploratory coherence analyses, which report changes in the variability in coherence as their outcome measure seems to be also less clear in its physiological meaning (which processes that we can measure in fMRI are changing in coherence here and why should this be related to sAA?). Again, we invite the authors to remove these analyses. However, If the authors feel that this is adding valuable information, they are invited to leave these analyses in and make the interpretation of the results more accessible to the reader.

We understand the reviewers concern. Two different methods measuring coherence can be confusing to the reader and therefore we decided to remove the wavelet coherence variability analysis. Thus, throughout the manuscript, the same spectral coherence method is now used: first, to examine interactions between the HRV frequency band and the LC fMRI signal, and second, as an exploratory analysis relating to negative stationary FC between the LC and MTL regions.

The goal of both the wavelet analyses and the spectral coherence analyses was to further understand these negative functional connectivity patterns. Changes in arousal, mediated by the ascending arousal system, are associated with increases and decreases in the fMRI signal [33]. For example, the amplitude and extent of correlations in fMRI data vary with indicators of drowsiness and light sleep [34-37], are altered by sleep deprivation [38] and reveal changes due to caffeine-induced fluctuations in arousal [39].Thus, these variations in arousal can induce phase-differences or frequency-specific magnitude correlations between regions. This may result in negative correlations in stationary functional connectivity analyses, as also shown in our own functional connectivity analyses across the different task stages (Figure 7). Phase differences between states may vary over time, as a function of arousal [40]. Neuromodulators, such as noradrenalin modulate the activity of a neuron and consequently, coherent neural communications are based on the dynamics of neurotransmitters. In fact, pharmacological manipulation of catecholamines has led to alterations in non-stationary functional connectivity [41].

Spectral coherence analysis allows the detection of magnitude and phase correlations at frequency specific levels between timeseries and is therefore well-suited to investigate frequency-specific phase and magnitude changes between task stages. In support of our hypothesis, we found magnitude correlation changes between the HIP-AMY between task stages and phase modulations between the AMY-LC. Coherence between the MTL regions was also shown to relate to emotional memory performance. We have removed the wavelet coherence analyses (removed the methods, associated statistical analyses, results and Figure 7—figure supplement 2) and rewritten this section.

In the Results:

“Areas related to arousal and salience processing, such as the LC, have particularly variable behavior that can result in a negative FC value, complicating their interpretation [27]. As our FC analyses assume stationarity within the fMRI timeseries, we aimed to understand these negative FC changes by exploring the frequency-specific magnitude and phase correlation changes in the follow-up spectral coherence analyses. Phase or magnitude changes in the frequency plane can result in reduced cross-correlation values.”

5) The results from the analyses on changes in coherence across task stages are not yet clear. I am wondering whether such analyses are prone to spurious results: if HIP-AMY coherence is higher in consolidation > baseline and in baseline > encoding, shouldn't it also be higher in consolidation > encoding? What does it mean if this effect is not picked up by the analyses? If the authors have easy access to the data, they may consider adding an analysis in a control area that shouldn't be implemented in consolidation or retrieval to provide cut-offs for avoiding spurious results or suggest alternative approaches to control for spurious results. Again, however, an alternative is to discuss this issue.

The reviewer is indeed right: we inadvertently changed the order of the task stages in the results text. We found increased coherence in encoding > baseline and decreased coherence in Recollection > encoding. We did not find a difference between baseline and Recollection. We apologize for the error and we corrected it in the main text.

With respect to the reviewer’s other remark: regions that went into this analysis were chosen based on the stationary functional connectivity results, as this was an analysis to better understand the interactions between these specific regions. We did ensure that our results were not spurious by applying a FDR-adjustment for multiple comparisons. Furthermore, we created matched AR1 randomized timeseries as a null-distribution for our comparisons, which is a more meaningful and valid control metric than choosing a region that may not interact with the MTL. These null-distributions are similar to the bootstrap and hence, the resampled timeseries have similar properties as the original data (distribution, autocorrelation) and have less bias in the rejection rates of the null-hypothesis. We have explained this further in the Materials and methods.

In the Results:

“We observed an increase in HIP-AMY coherence for encoding compared to baseline (β=-0.13, t(821)=-4.39, p<0.001, 95%CI[-0.20, -0.052]), a decrease in recollection compared to encoding (β=0.10, t(821)=3.37, p=0.004, 95%CI[0.023, 0.171]) and a trend for consolidation decrease compared to baseline (β=-0.07, t(821)=-2.35, p=0.088 95%CI[-0.141, 0.007]) and a change in phase from AMY leading the LC in encoding to the LC leading the AMY in recollection (β=0.91, t(821)=2.73, p=0.033, 95%CI[0.051,1.772] Figure 7B).”

In the Materials and methods:

“These null-distributions are similar to the bootstrap and hence, the resampled timeseries have similar properties as the original data (distribution, autocorrelation) and have less bias in the rejection rates of the null-hypotheses.”

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

We thank the authors for implementing most of the requested changes. However, in their responses to Comment 1 and 2, they misunderstood or overlooked important aspects of the question. We would like to ask them to adjust their responses and changes in the paper as follows:

Re Comment 1: Please also indicate whether the names used for the calculation of the bias score were paired with new or old faces or whether this was collapsed across both old and new faces.

Re Comment 2: There was a misunderstanding in the response to comment 2. The statistics required in the table were the hit and false alarm rate for names, not for faces. Please report the Hit rate, FA rate and Hit-FA rate separately for the names paired in the memory tests with old emotional and old neutral faces and for those paired with new emotional and new neutral new faces.

We believe that there is misunderstanding in what these scores represent. The Recollection Hit Rate Raw represents the number of correct names to old faces identified during recollection divided by the total number of available old faces (whether or not they were identified as old). Thus, this is the %Hit rate. As indicated previously, 1 – Recollection Hit Rate (correction recognition) represents the false alarm. It should be noted that this %hit may be an underestimation as the participant may not have been able to indicate the name to the face, in case he/she identified the face as new. As explained in the previous revision and in Figure 1, only faces that were endorsed by the participant in the recognition stage would go to the recollection stage.

The %Hits as requested by the editor, would in our opinion be calculated as the number of correctly identified names belonging to old faces divided by the total number of presented old faces related to these names. This would then be expressed in % (multiplied by 100).

Therefore, the total number of old faces that were presented (denominator) equals the number of correctly recognized old faces in recognition. The correctly identified names (numerator) is the number of correct names to old faces during recollection.

The Recollection Hit Rate (correct Recognition) represents the number of correct names to old faces identified during recollection divided by the total number of old faces that were recognized correctly during recognition (presented as a %). This therefore represents the % Hit rate for names presented in the test with old faces that were correctly recognized as old and, to our understanding, is what the editor is requesting. 1 – Recollection Hit Rate (correct Recognition) represents the False Alarm for old faces.

We have added the Hit Rate – False Alarm rate to the table and the table is now included in the manuscript.

References

1. Khakpour-Taleghani B, Lashgari R, Aavani T, Haghparast A, Naderi N, Motamedi F. The locus coeruleus involves in consolidation and memory retrieval, but not in acquisition of inhibitory avoidance learning task. Behav Brain Res. 2008;189(2):257-62. doi: 10.1016/j.bbr.2008.01.004. PubMed PMID: 18295357.2. Wagatsuma A, Okuyama T, Sun C, Smith LM, Abe K, Tonegawa S. Locus coeruleus input to hippocampal CA3 drives single-trial learning of a novel context. Proc Natl Acad Sci U S A. 2018;115(2):E310-E6. doi: 10.1073/pnas.1714082115. PubMed PMID: 29279390; PubMed Central PMCID: PMCPMC5777050.3. Murchison CF, Zhang XY, Zhang WP, Ouyang M, Lee A, Thomas SA. A distinct role for norepinephrine in memory retrieval. Cell. 2004;117(1):131-43. Epub 2004/04/07. doi: 10.1016/s0092-8674(04)00259-4. PubMed PMID: 15066288.4. Hansen N, Manahan-Vaughan D. Locus Coeruleus Stimulation Facilitates Long-Term Depression in the Dentate Gyrus That Requires Activation of β-Adrenergic Receptors. Cereb Cortex. 2015;25(7):1889-96. doi: 10.1093/cercor/bht429. PubMed PMID: 24464942; PubMed Central PMCID: PMCPMC4459289.5. McIntyre CK, Hatfield T, McGaugh JL. Amygdala norepinephrine levels after training predict inhibitory avoidance retention performance in rats. Eur J Neurosci. 2002;16(7):1223-6. Epub 2002/10/31. doi: 10.1046/j.1460-9568.2002.02188.x. PubMed PMID: 12405982.6. Vazey EM, Moorman DE, Aston-Jones G. Phasic locus coeruleus activity regulates cortical encoding of salience information. Proc Natl Acad Sci U S A. 2018;115(40):E9439-E48. Epub 2018/09/21. doi: 10.1073/pnas.1803716115. PubMed PMID: 30232259; PubMed Central PMCID: PMCPMC6176602.7. Roozendaal B, Hermans EJ. Norepinephrine effects on the encoding and consolidation of emotional memory: improving synergy between animal and human studies. Curr Opin Behav Sci. 2017;14:115-22.8. Lemon N, Aydin-Abidin S, Funke K, Manahan-Vaughan D. Locus coeruleus activation facilitates memory encoding and induces hippocampal LTD that depends on β-adrenergic receptor activation. Cereb Cortex. 2009;19(12):2827-37. Epub 2009/05/14. doi: 10.1093/cercor/bhp065. PubMed PMID: 19435710; PubMed Central PMCID: PMCPMC2774396.9. Roozendaal B, Castello NA, Vedana G, Barsegyan A, McGaugh JL. Noradrenergic activation of the basolateral amygdala modulates consolidation of object recognition memory. Neurobiol Learn Mem. 2008;90(3):576-9. Epub 2008/07/29. doi: 10.1016/j.nlm.2008.06.010. PubMed PMID: 18657626; PubMed Central PMCID: PMCPMC2572617.10. Sara SJ, Devauges V. Priming stimulation of locus coeruleus facilitates memory retrieval in the rat. Brain Res. 1988;438(1-2):299-303. Epub 1988/01/12. doi: 10.1016/0006-8993(88)91351-0. PubMed PMID: 3345434.11. Sara SJ, Segal M. Plasticity of sensory responses of locus coeruleus neurons in the behaving rat: implications for cognition. Prog Brain Res. 1991;88:571-85. Epub 1991/01/01. doi: 10.1016/s0079-6123(08)63835-2. PubMed PMID: 1813935.12. Stanislaw H, Todorov N. Calculation of signal detection theory measures. Behavior Research Methods, Instruments, & Computers. 1999;31(1):137-49. doi: 10.3758/bf03207704.13. Yonelinas AP, Aly M, Wang WC, Koen JD. Recollection and familiarity: examining controversial assumptions and new directions. Hippocampus. 2010;20(11):1178-94. Epub 2010/09/18. doi: 10.1002/hipo.20864. PubMed PMID: 20848606; PubMed Central PMCID: PMCPMC4251874.14. Yonelinas AP, Otten LJ, Shaw KN, Rugg MD. Separating the brain regions involved in recollection and familiarity in recognition memory. J Neurosci. 2005;25(11):3002-8. Epub 2005/03/18. doi: 10.1523/JNEUROSCI.5295-04.2005. PubMed PMID: 15772360; PubMed Central PMCID: PMCPMC6725129.15. Eichenbaum H, Yonelinas AP, Ranganath C. The medial temporal lobe and recognition memory. Annu Rev Neurosci. 2007;30:123-52. Epub 2007/04/10. doi: 10.1146/annurev.neuro.30.051606.094328. PubMed PMID: 17417939; PubMed Central PMCID: PMCPMC2064941.16. Rugg MD, Vilberg KL. Brain networks underlying episodic memory retrieval. Current opinion in neurobiology. 2013;23(2):255-60. Epub 2012/12/05. doi: 10.1016/j.conb.2012.11.005. PubMed PMID: 23206590; PubMed Central PMCID: PMCPMC3594562.17. O'Brien KR, Kober T, Hagmann P, Maeder P, Marques J, Lazeyras F, Krueger G, Roche A. Robust T1-weighted structural brain imaging and morphometry at 7T using MP2RAGE. PLoS One. 2014;9(6):e99676. Epub 2014/06/17. doi: 10.1371/journal.pone.0099676. PubMed PMID: 24932514; PubMed Central PMCID: PMCPMC4059664.18. Avants BB, Tustison NJ, Song G, Cook PA, Klein A, Gee JC. A reproducible evaluation of ANTs similarity metric performance in brain image registration. NeuroImage. 2011;54(3):2033-44. Epub 2010/09/21. doi: 10.1016/j.neuroimage.2010.09.025. PubMed PMID: 20851191; PubMed Central PMCID: PMCPMC3065962.19. Klein A, Andersson J, Ardekani BA, Ashburner J, Avants B, Chiang M-C, Christensen GE, Collins DL, Gee J, Hellier P, Song JH, Jenkinson M, Lepage C, Rueckert D, Thompson P, Vercauteren T, Woods RP, Mann JJ, Parsey RV. Evaluation of 14 nonlinear deformation algorithms applied to human brain MRI registration. NeuroImage. 2009;46(3):786-802. doi: 10.1016/j.neuroimage.2008.12.037. PubMed PMID: PMC2747506.20. Iglesias JE, Augustinack JC, Nguyen K, Player CM, Player A, Wright M, Roy N, Frosch MP, McKee AC, Wald LL, Fischl B, Van Leemput K, Alzheimer's Disease Neuroimaging I. A computational atlas of the hippocampal formation using ex vivo, ultra-high resolution MRI: Application to adaptive segmentation of in vivo MRI. NeuroImage. 2015;115:117-37. doi: 10.1016/j.neuroimage.2015.04.042. PubMed PMID: 25936807; PubMed Central PMCID: PMC4461537.21. Iglesias JE, Van Leemput K, Bhatt P, Casillas C, Dutt S, Schuff N, Truran-Sacrey D, Boxer A, Fischl B, for the Alzheimer’s Disease Neuroimaging I. Bayesian segmentation of brainstem structures in MRI. NeuroImage. 2015;113:184-95. doi: 10.1016/j.neuroimage.2015.02.065. PubMed PMID: PMC4434226.22. Sclocco R, Beissner F, Desbordes G, Polimeni JR, Wald LL, Kettner NW, Kim J, Garcia RG, Renvall V, Bianchi AM, Cerutti S, Napadow V, Barbieri R. Neuroimaging brainstem circuitry supporting cardiovagal response to pain: a combined heart rate variability/ultrahigh-field (7 T) functional magnetic resonance imaging study. Philos Trans A Math Phys Eng Sci. 2016;374(2067). doi: 10.1098/rsta.2015.0189. PubMed PMID: 27044996; PubMed Central PMCID: PMCPMC4822448.23. Napadow V, Dhond R, Conti G, Makris N, Brown EN, Barbieri R. Brain correlates of autonomic modulation: combining heart rate variability with fMRI. NeuroImage. 2008;42(1):169-77. Epub 2008/06/06. doi: 10.1016/j.neuroimage.2008.04.238. PubMed PMID: 18524629; PubMed Central PMCID: PMCPMC2603289.24. Lane RD, McRae K, Reiman EM, Chen K, Ahern GL, Thayer JF. Neural correlates of heart rate variability during emotion. NeuroImage. 2009;44(1):213-22. Epub 2008/09/10. doi: 10.1016/j.neuroimage.2008.07.056. PubMed PMID: 18778779.25. Thayer JF, Ahs F, Fredrikson M, Sollers JJ, 3rd, Wager TD. A meta-analysis of heart rate variability and neuroimaging studies: implications for heart rate variability as a marker of stress and health. Neurosci Biobehav Rev. 2012;36(2):747-56. Epub 2011/12/20. doi: 10.1016/j.neubiorev.2011.11.009. PubMed PMID: 22178086.26. Bittiner SB, Smith SE. Beta-adrenoceptor antagonists increase sinus arrhythmia, a vagotonic effect. Br J Clin Pharmacol. 1986;22(6):691-5. Epub 1986/12/01. doi: 10.1111/j.1365-2125.1986.tb02959.x. PubMed PMID: 2882771; PubMed Central PMCID: PMCPMC1401218.27. Chang C, Glover GH. Time–frequency dynamics of resting-state brain connectivity measured with fMRI. NeuroImage. 2010;50(1):81-98. doi: https://doi.org/10.1016/j.neuroimage.2009.12.011.

28. Rohleder N, Nater UM. Determinants of salivary α-amylase in humans and methodological considerations. Psychoneuroendocrinology. 2009;34(4):469-85. doi: 10.1016/j.psyneuen.2008.12.004. PubMed PMID: 19155141.

29

. Jacobs HIL, Riphagen JM, Razat CM, Wiese S, Sack AT. Transcutaneous vagus nerve stimulation boosts associative memory in older individuals. Neurobiology of Aging. 2015;36(5):1860-7. doi: 10.1016/j.neurobiolaging.2015.02.023. PubMed PMID: WOS:000355100900008.30. Akselrod S, Gordon D, Ubel FA, Shannon DC, Berger AC, Cohen RJ. Power spectrum analysis of heart rate fluctuation: a quantitative probe of beat-to-beat cardiovascular control. Science. 1981;213(4504):220-2. Epub 1981/07/10. doi: 10.1126/science.6166045. PubMed PMID: 6166045.31. Wallentin M, Nielsen AH, Vuust P, Dohn A, Roepstorff A, Lund TE. Amygdala and heart rate variability responses from listening to emotionally intense parts of a story. NeuroImage. 2011;58(3):963-73. Epub 2011/07/14. doi: 10.1016/j.neuroimage.2011.06.077. PubMed PMID: 21749924.32. Billings JCW, Thompson GJ, Pan WJ, Magnuson ME, Medda A, Keilholz S. Disentangling Multispectral Functional Connectivity With Wavelets. Front Neurosci. 2018;12:812. Epub 2018/11/22. doi: 10.3389/fnins.2018.00812. PubMed PMID: 30459548; PubMed Central PMCID: PMCPMC6232345.33. Chang C, Leopold DA, Scholvinck ML, Mandelkow H, Picchioni D, Liu X, Ye FQ, Turchi JN, Duyn JH. Tracking brain arousal fluctuations with fMRI. Proc Natl Acad Sci U S A. 2016;113(16):4518-23. Epub 2016/04/07. doi: 10.1073/pnas.1520613113. PubMed PMID: 27051064; PubMed Central PMCID: PMCPMC4843437.34. Chang C, Liu Z, Chen MC, Liu X, Duyn JH. EEG correlates of time-varying BOLD functional connectivity. NeuroImage. 2013;72:227-36. Epub 2013/02/05. doi: 10.1016/j.neuroimage.2013.01.049. PubMed PMID: 23376790; PubMed Central PMCID: PMCPMC3602157.35. Fukunaga M, Horovitz SG, van Gelderen P, de Zwart JA, Jansma JM, Ikonomidou VN, Chu R, Deckers RH, Leopold DA, Duyn JH. Large-amplitude, spatially correlated fluctuations in BOLD fMRI signals during extended rest and early sleep stages. Magn Reson Imaging. 2006;24(8):979-92. Epub 2006/09/26. doi: 10.1016/j.mri.2006.04.018. PubMed PMID: 16997067.36. Horovitz SG, Fukunaga M, de Zwart JA, van Gelderen P, Fulton SC, Balkin TJ, Duyn JH. Low frequency BOLD fluctuations during resting wakefulness and light sleep: a simultaneous EEG-fMRI study. Hum Brain Mapp. 2008;29(6):671-82. Epub 2007/06/29. doi: 10.1002/hbm.20428. PubMed PMID: 17598166.37. Olbrich S, Mulert C, Karch S, Trenner M, Leicht G, Pogarell O, Hegerl U. EEG-vigilance and BOLD effect during simultaneous EEG/fMRI measurement. NeuroImage. 2009;45(2):319-32. Epub 2008/12/27. doi: 10.1016/j.neuroimage.2008.11.014. PubMed PMID: 19110062.38. De Havas JA, Parimal S, Soon CS, Chee MW. Sleep deprivation reduces default mode network connectivity and anti-correlation during rest and task performance. NeuroImage. 2012;59(2):1745-51. Epub 2011/08/30. doi: 10.1016/j.neuroimage.2011.08.026. PubMed PMID: 21872664.39. Wong CW, Olafsson V, Tal O, Liu TT. The amplitude of the resting-state fMRI global signal is related to EEG vigilance measures. NeuroImage. 2013;83:983-90. Epub 2013/08/01. doi: 10.1016/j.neuroimage.2013.07.057. PubMed PMID: 23899724; PubMed Central PMCID: PMCPMC3815994.40. Lurie DJ, Kessler D, Bassett DS, Betzel RF, Breakspear M, Kheilholz S, Kucyi A, Liegeois R, Lindquist MA, McIntosh AR, Poldrack RA, Shine JM, Thompson WH, Bielczyk NZ, Douw L, Kraft D, Miller RL, Muthuraman M, Pasquini L, Razi A, Vidaurre D, Xie H, Calhoun VD. Questions and controversies in the study of time-varying functional connectivity in resting fMRI. Netw Neurosci. 2020;4(1):30-69. Epub 2020/02/12. doi: 10.1162/netn_a_00116. PubMed PMID: 32043043; PubMed Central PMCID: PMCPMC7006871.41. Shine JM, van den Brink RL, Hernaus D, Nieuwenhuis S, Poldrack RA. Catecholaminergic manipulation alters dynamic network topology across cognitive states. Netw Neurosci. 2018;2(3):381-96. Epub 2018/10/09. doi: 10.1162/netn_a_00042. PubMed PMID: 30294705; PubMed Central PMCID: PMCPMC6145851.

https://doi.org/10.7554/eLife.52059.sa2

Article and author information

Author details

  1. Heidi IL Jacobs

    1. Gordon Center for Medical Imaging, Department of Radiology, Massachusetts General Hospital/Harvard Medical School, Boston, United States
    2. Faculty of Health, Medicine and Life Sciences, School for Mental Health and Neuroscience, Alzheimer Centre Limburg, Maastricht University, Maastricht, Netherlands
    3. Faculty of Psychology and Neuroscience, Department of Cognitive Neuroscience, Maastricht University, Maastricht, Netherlands
    Contribution
    Conceptualization, Data curation, Formal analysis, Supervision, Funding acquisition, Investigation, Methodology, Project administration, Data interpretation
    Contributed equally with
    Nikos Priovoulos
    For correspondence
    h.jacobs@maastrichtuniversity.nl
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-7620-3822
  2. Nikos Priovoulos

    Faculty of Health, Medicine and Life Sciences, School for Mental Health and Neuroscience, Alzheimer Centre Limburg, Maastricht University, Maastricht, Netherlands
    Contribution
    Data curation, Formal analysis, Investigation, Visualization, Methodology, Data interpretation
    Contributed equally with
    Heidi IL Jacobs
    Competing interests
    No competing interests declared
  3. Benedikt A Poser

    Faculty of Psychology and Neuroscience, Department of Cognitive Neuroscience, Maastricht University, Maastricht, Netherlands
    Contribution
    Software, Supervision, Data interpretation
    Competing interests
    No competing interests declared
  4. Linda HG Pagen

    Faculty of Health, Medicine and Life Sciences, School for Mental Health and Neuroscience, Alzheimer Centre Limburg, Maastricht University, Maastricht, Netherlands
    Contribution
    Investigation
    Competing interests
    No competing interests declared
  5. Dimo Ivanov

    Faculty of Psychology and Neuroscience, Department of Cognitive Neuroscience, Maastricht University, Maastricht, Netherlands
    Contribution
    Software, Data interpretation
    Competing interests
    No competing interests declared
  6. Frans RJ Verhey

    Faculty of Health, Medicine and Life Sciences, School for Mental Health and Neuroscience, Alzheimer Centre Limburg, Maastricht University, Maastricht, Netherlands
    Contribution
    Data interpretation
    Competing interests
    No competing interests declared
  7. Kâmil Uludağ

    1. Center for Neuroscience Imaging Research, Institute for Basic Science and Department of Biomedical Engineering, Sungkyunkwan University, Suwon, Republic of Korea
    2. Techna Institute & Koerner Scientist in MR Imaging, University Health Network, Toronto, Canada
    Contribution
    Data interpretation
    Competing interests
    No competing interests declared

Funding

Nederlandse Organisatie voor Wetenschappelijk Onderzoek (VENI 451-14-035)

  • Heidi IL Jacobs

Universiteit Maastricht (Intramural support FHML)

  • Heidi IL Jacobs

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We would like to thank Ms. Niky Vogt for her assistance in creating the experimental paradigm and recruiting participants.

This work is supported by a NWO VENI [451-14-035] to HILJ and supported by intramural support by the Faculty of Health Medicine and Life Sciences of Maastricht University to HILJ.

Ethics

Human subjects: All participants provided written informed. Approval of the experimental protocol was obtained from the local ethical committee of the Faculty of Psychology and Neuroscience at Maastricht University (#07_11_2014_A1).

Senior Editor

  1. Timothy E Behrens, University of Oxford, United Kingdom

Reviewing Editor

  1. Dorothea Hämmerer, University College London, United Kingdom

Reviewer

  1. Mara Mather, University of Southern California, United States

Publication history

  1. Received: September 20, 2019
  2. Accepted: June 24, 2020
  3. Accepted Manuscript published: June 24, 2020 (version 1)
  4. Version of Record published: July 8, 2020 (version 2)

Copyright

© 2020, Jacobs et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 817
    Page views
  • 116
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

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)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Neuroscience
    Neven Borak, Johannes Kohl
    Insight

    Unexplained repeated pregnancy loss is associated with an altered perception of male odors and differences in brain regions that process smells.

    1. Genetics and Genomics
    2. Neuroscience
    Marina Kovalenko et al.
    Research Article

    Somatic expansion of the Huntington's disease (HD) CAG repeat drives the rate of a pathogenic process ultimately resulting in neuronal cell death. Although mechanisms of toxicity are poorly delineated, transcriptional dysregulation is a likely contributor. To identify modifiers that act at the level of CAG expansion and/or downstream pathogenic processes, we tested the impact of genetic knockout, in HttQ111 mice, of Hdac2 or Hdac3 in medium-spiny striatal neurons that exhibit extensive CAG expansion and exquisite disease vulnerability. Both knockouts moderately attenuated CAG expansion, with Hdac2 knockout decreasing nuclear huntingtin pathology. Hdac2 knockout resulted in a substantial transcriptional response that included modification of transcriptional dysregulation elicited by the HttQ111 allele, likely via mechanisms unrelated to instability suppression. Our results identify novel modifiers of different aspects of HD pathogenesis in MSNs and highlight a complex relationship between the expanded Htt allele and Hdac2 with implications for targeting transcriptional dysregulation in HD.