Locus coeruleus-related insula activation supports implicit learning

1 The noradrenergic locus coeruleus and its neuromodulatory cortical projections are 2 critical for adaptive behavior, yet their contributions to implicit learning in novel 3 environments remain incompletely understood, due to challenges in non-invasive assessment . 4 Here, we combined multimodal neuroimaging—including locus-coeruleus-sensitive structural 5 MRI, concurrent pupillometry–fMRI, and PET -derived noradrenergic transporter maps—with 6 repeated behavioral assessments to investigate noradrenergic contributions to implicit 7 learning across younger and older adults (n = 77). 8 Salient expectation-violating stimuli elicited pupil dilation, indicating enhanced 9 neuromodulation, activated the action-mode network and deactivated the default-mode 10 network. Pupil-linked BOLD responses suggested a functional coupling between the locus 11 coeruleus and action-mode network, further supported by spatial overlap of activation patterns 12 with PET-derived noradrenergic transporter maps. Locus coeruleus MRI-guided functional 13 connectivity analyses demonstrated that locus coeruleus activity is coupled to anterior insula 14 activation, suggesting a noradrenergic role in shifting cortical dynamics toward action-15 oriented processing. 16 Behaviorally, participants implicitly learned the statistical task structure over time, as 17 evidenced by reaction time adjustments based on stimulus probabilities. Critically, stronger 18 locus coeruleus integrity, greater task-related anterior insula activation, and more pronounced 19 pupil dilation were associated with enhanced implicit learning, highlighting the behavioral 20 relevance of noradrenergic neuromodulation. Notably, noradrenergic responses and their link 21 to learning were preserved across age groups, suggesting a robust noradrenergic role in 22 supporting adaptive behavior throughout adulthood. 23 These findings provide novel insights into the neuromodulatory mechanisms 24 underlying learning and cognitive flexibility, emphasizing the pivotal role of locus coeruleus–25 action-mode network interactions in behavioral adaptation. 26

locus coeruleus, neuromodulation, noradrenaline, norepinephrine, MRI, 27 pupillometry, attention, memory, implicit learning, aging 28 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 3 3 1. Introduction 29 Optimal behavior requires dynamic adaptation to changing environments 1. On a 30 neurochemical level, unexpected environmental changes transiently increase noradrenaline 31 release from the locus coeruleus, a brainstem nucleus that facilitates attention and learning 2–4. 32 Specifically, noradrenaline release has long been conceptualized as a neural interrupt signal 33 that promotes rapid behavioral adaptation to new environmental imperatives5–8. The 34 momentary processing of salient stimuli is thereby facilitated by a noradrenergic enhancement 35 of neural gain, effectively increasing signal-to-noise ratios in neural circuits2,9–12. By contrast, 36 long-term adjustments to salient events are mediated by noradrenergic influences on 37 hippocampal synaptic plasticity3,13–16. 38 In a similar vein, more recent accounts propose that the locus coeruleus is activated 39 when predictions about the world are violated. The resulting noradrenaline release helps 40 refine internal model of the world by adjusting the rate of learning in cortical target regions17–41 21. 42 To enable these functions, t he locus coeruleus receives critical contextual information 43 about stimulus salience and utility from cortical structures such as the anterior cingulate and 44 insula2,9. In primates, these forebrain structures send major direct inputs to the locus coeruleus 45 to recruit noradrenergic neuromodulation and modulate brain-wide processing according to 46 environmental demands9. 47 The anterior cingulate and insular cortices have recently been grouped into the action-48 mode network22, overlapping with the cingulo-opercular network or ventral attention 49 network6,23,24. The action-mode network is closely linked to initiating and maintaining states 50 of heightened arousal and focused attention, as well as the generation and updating of actions 51 based on salient internal or external signals22,25. In this context, arousal is defined as a 52 continuum of sensitivity to environmental stimuli26, tightly regulated by noradrenergic 53 neuromodulation27–29. In line with this, it has been proposed that the locus coeruleus 54 modulates or is part of the action-mode network6,22,30, but direct in vivo evidence remains 55 limited. 56 Aging dysregulates the noradrenergic system 31–33, possibly due to the age-related 57 accumulation of pathology34–37. In accordance with this, converging evidence based on 58 cognitive assessments suggests that age differences in proxies of noradrenergic 59 neuromodulation contribute to late-life impairments in attention38–40, learning, and memory41–60 43; i.e., in situations when participants were explicitly instructed about the cognitive testing. 61 However, the role of the locus coeruleus in how younger and older adults implicitly learn to 62 adapt their behavior in new situations remains underexplored. 63 Here, we investigated the implications of locus coeruleus–cortical network 64 interactions for implicit behavioral adaptation using a multimodal age-comparative approach. 65 To this end, we used a modified conditioned oddball task, previously shown to reliably 66 increase locus coeruleus spiking in non-human primates44. In this variant of the oddball task, 67 participants detect infrequent, distinctive stimuli amid a series of repetitive standard stimuli. 68 To further increase recruitment of the locus coeruleus, infrequent oddball stimuli are pre-69 conditioned by repeatedly pairing them with appetitive or aversive outcomes40,44–46. Our 70 participants completed the conditioned oddball task twice on consecutive days. Trial-level 71 behavioral responses from the task allowed us to examine implicit learning within and across 72 sessions. In this context, implicit learning refers to how initially naïve participants pick up the 73 statistical regularities of the originally new but increasingly familiar task environment and 74 optimize their behavior over time (i.e., without explicit instructions). 75 To overcome challenges in non-invasive assessment posed by the locus coeruleus’ 76 small size and location in the brainstem47–49, we relied on several recent additions to the 77 neuroscience toolkit. In particular, dedicated structural magnetic resonance imaging (MRI) 78 sequences can reveal the locus coeruleus as a cluster of distinct bright (hyperintense) voxels 79 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 4 4 bordering the fourth ventricle41,49. Transgenic animal50,51 and human post-mortem52 validation 80 studies suggest this signal may serve as an in-vivo proxy for locus coeruleus integrity. Here, 81 we relied on neuromodulation-sensitive MRI to assess individual differences in locus 82 coeruleus integrity and to reliably localize the small brainstem nucleus in functional analyses. 83 To additionally track locus coeruleus activation patterns, we combined fMRI with concurrent 84 pupillometry, as increasing arousal-related neuromodulatory activity, including in the 85 noradrenergic system, dilates pupils53–55. Finally, we compared the observed pupillometry–86 fMRI activation maps to the brain-wide distribution of noradrenergic and other 87 neuromodulatory transporters56 to gauge potential dependencies. In sum, younger and older 88 adults completed a conditioned oddball task while we assessed multiple proxies for locus 89 coeruleus structure and function to explore their role in implicit learning. 90 In a brief synopsis of our findings, salient stimuli reliably dilated younger and older 91 participants’ pupils and activated the action-mode network. Pupil-indexed neuromodulation 92 correlated with brainstem and action-mode network activation, and was associated with 93 noradrenergic transporter distribution, suggesting potential links to the locus coeruleus, which 94 we confirmed using functional connectivity analyses. Initially naïve participants implicitly 95 learned the structure of the new task environment and adapted their behavior over time. 96 Behavioral adaptation (i.e., implicit learning) was positively associated with structural and 97 functional locus coeruleus and action-mode network measures. Together, these results suggest 98 that noradrenergic signals support refining internal models of environmental contingencies for 99 behavioral adaptation by modulating attention and memory following salient events. By 100 integrating advanced MRI, pupillometry, and functional connectivity analyses, we reveal 101 critical insights into how neuromodulatory signals dynamically drive behavioral flexibility. 102 103 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 5 5 104 Figure 1. Overview of the conditioned oddball task (a) and experimental phases (b). As part of a 105 three-stimulus oddball task, previously conditioned stimuli (CS+) are presented as infrequent oddball 106 stimuli while non-conditioned perceptually matched control stimuli (CS–) serve as frequent standards. 107 A second oddball category consists of perceptually matched but previously not shown stimuli, 108 hypothesized to elicit a novelty response. (b) Fear conditioning sessions are repeated over three 109 consecutive days while collecting eye tracking and EEG data. The conditioned oddball task takes 110 place on Day 2 (pupillometry-EEG) and Day 3 (pupillometry-fMRI). During the oddball tasks, no 111 electrical stimulation is applied. 112 113 2. Results 114 Conditioning reliably increases pupil-indexed neuromodulation across age groups 115 To increase recruitment of the locus coeruleus, experimental stimuli that were 116 subsequently shown in the main task were pre-conditioned by repeatedly pairing them with an 117 aversive electrical stimulation (Figure 1,44). Robust pupil dilation following presentation of 118 conditioned stimuli (CS) relative to perceptually matched control stimuli (CS–) suggests this 119 procedure enhanced neuromodulatory activation, including in the noradrenergic system55, 120 across age groups and assessment days (Figure 2a; all pcluster-corr ≤ 0.002). 121 122 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 6 6 123 Figure 2. Stimulus -related pupil responses during fear conditioning (a) and conditioned oddball 124 tasks (b) . During daily fear conditioning sessions, fear conditioned stimuli (CS+; red) elicit stronger 125 pupil dilation relative to perceptually matched control stimuli (CS–, black), suggesting 126 neuromodulatory activation53,54 . Note that both stimuli initially elicit comparable pupil constriction 127 due to higher luminance compared to the background (pupil light reflex after stimulus presentation at 128 time 0). (b) During subsequent pupillometry-EEG (day 1) and pupillometry-fMRI (day 2) conditioned 129 oddball tasks, oddball presentation (red) is linked to larger pupil dilation relative to perceptually 130 matched control stimuli (standard; black). Differences in pupil light reflexes between oddball 131 acquisitions result from different background luminance in the two laboratories (EEG, MRI). The 132 black horizontal lines indicate the temporal extent of the significant group level cluster, controlling for 133 multiple comparisons, which is comparable across days. Shaded areas represent standard errors of the 134 mean. 135 136 Conditioned oddball stimuli dilate the pupil and activate the action -mode network 137 During the subsequent main task, salient stimuli—oddballs that were either pre-138 conditioned or novel—reliably dilated participants’ pupils (pcluster-corr ≤ 0.002) and activated 139 the action-mode network relative to perceptually matched control stimuli (standards [CS–]; 140 Figures 2b and 3). 141 Action-mode network recruitment encompassed the bilateral anterior insula (T = 142 11.81, pFWE-corr < 0.001 MNI: 30, 22, 10), middle cingulate gyrus (T = 8.95, pFWE < 0.001 143 MNI: 3.5, 7.5, 31.5), thalamus (T = 7, pFWE < 0.001 MNI: 12, –20, 12) and cerebellum (T = 144 7.67, pFWE < 0.001 MNI: 30, –50, –24;22,25). In addition, salient stimuli elicited prominent 145 dorsal attention-network activation (bilateral superior parietal lobule; T = 13.8, pFWE-corr < 146 0.001 MNI: 42, –38, 50;6,25) and default-mode network deactivation (e.g., bilateral frontal 147 pole, T = –11.64, pFWE < 0.001 MNI: –4, 68, 4;22). The un/thresholded whole-brain contrast 148 maps are available at osf.io/4t5hj. 149 Together, this suggests task-related neuromodulatory activation, as indicated by pupil 150 dilation, as well as activation of the action-mode network, reflecting major cortical in- and 151 output regions of the noradrenergic system6,9,57. This may indicate that in response to salient 152 expectation-violating stimuli, there is a transition from default-mode to action-mode network 153 activation, likely facilitated by noradrenergic neuromodulation, which supports coping with 154 changing task demands5,17,22. 155 156 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 7 7 157 Figure 3. Salient oddball stimuli elicit action-mode network (a) and dorsal attention network 158 activation (b) and deactivate the default mode network (c). For visualization, group contrast maps 159 (oddballs vs standards) are thresholded at T = 3, main text results report family-wise error corrected 160 statistics. Contrast maps are available at: osf.io/4t5hj. Numbers below the labeled regions indicate 161 MNI XYZ-coordinates. (d-e) Overlap of group-level activation (d) and deactivation (e) maps with 162 three published functional network atlases23,58,59, quantified as Dice coefficients. Only networks with a 163 significant overlap based on permutation tests 24 and Dice coefficients > 0.2 are labeled. Note that the 164 Action-mode network has previously been termed Cingulo-opercular network22 and the Salience 165 network is also termed Ventral attention network24. 166 167 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 8 8 Pupil-indexed neuromodulation is associated with action-mode network and brainstem 168 activation 169 By integrating pupil time series into the fMRI analyses, we directly probe the 170 relationship between the detected cortical activation patterns and markers of 171 neuromodulation. This approach allowed us to test which brain regions’ activity correlated 172 with moment-to-moment pupil fluctuations, which are likely driven b y release of 173 neuromodulatory neurotransmitters53–55. 174 Supporting an association between noradrenergic neuromodulation and action-mode 175 activation, network hubs such as bilateral anterior insula (T = 7.31, pFWE-corr < 0.001 MNI: –176 32, 20, –8), middle cingulate cortex (T = 5.55, pFWE-corr < 0.001 MNI: –4 –0 44), and thalamus 177 (T = 7.88, pFWE-corr < 0.001 MNI: 14 –12 10) scaled their activity with pupil size (Figure 4). 178 The un/thresholded whole-brain contrast maps are available at osf.io/4t5hj.These findings 179 corroborate recent observations—linking human intracranial insula recordings with pupil 180 dynamics30—and extend them to a functional network level. 181 Importantly, these pupil–fMRI analyses also revealed a bilateral brainstem–midbrain 182 cluster (T = 7.27, pFWE-corr < 0.001 MNI: –4 –30 –14), overlapping with key noradrenergic, 183 serotonergic, and dopaminergic nuclei, in accordance with recent work in rodents55,60,61 and 184 primates62–64. 185 In sum, we found that neuromodulatory and action-mode network regions activated 186 when pupils dilated, suggesting neuromodulatory drive of network transitions. 187 188 Figure 4. Moment-to-moment changes in pupil size correlate with action-mode network (a) and 189 neuromodulatory brainstem–midbrain activation (b–c). For visualization, group contrast maps are 190 thresholded at T = 3, main text results report family-wise error corrected statistics. Contrast maps are 191 available at: osf.io/4t5hj. Numbers below the labeled regions indicate MNI XYZ-coordinates. 192 Noradrenergic transporter distribution overlaps with pupil-linked BOLD activation patterns 193 If indeed the observed activation pattern is supported by noradrenergic release in the 194 cortex, one would expect overlapping spatial distributions with the noradrenergic system. To 195 test this, we leveraged public PET-derived maps of the cortical and subcortical distribution of 196 the noradrenergic transporter56,65–69. We then assessed the voxel-wise association of 197 noradrenergic transporter expression and pupil-linked BOLD activation. We detected a robust 198 correlation (Hesse et al. map: r = 0.189; ppermutation-corr < 0.001), that we replicated using an 199 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 9 9 alternative noradrenergic transporter map (Ding et al. map: r = 0.161; ppermutation-corr < 0.001; 200 Figure 5). This suggests that those parts of the brain that activated when pupils dilated 201 overlapped with those housing noradrenergic receptors. 202 To further contextualize these results, we computed two negative control analyses. 203 Specifically, we linked two brain-wide GABAA receptor distribution maps70,71 to our pupil-204 linked BOLD maps, which we assumed to show no or even a negative association72. In line 205 with this, we found that GABAA receptor expression showed a non-significant negative 206 relation with the pupil–fMRI pattern (Nørgaard et al map: r = –0.028; ppermutation-corr = 0.063; 207 Dukart et al. map: r = –0.045; ppermutation-corr = 0.056; Figure 5). 208 Finally, given that large parts of the brainstem–midbrain, overlapping with 209 noradrenergic, but also serotonergic and dopaminergic nuclei, activated when pupils dilated, 210 we ran two additional analyses to determine whether similar effects extended to other 211 neuromodulatory systems. Both serotonergic (Beliveau et al. map: r = 0.229; ppermutation-corr < 212 0.001;73) and dopaminergic (Sasaki et al. map: r = 0.241; ppermutation-corr < 0.001;74) transporter 213 expression showed a comparable spatial correspondence with the pupil–fMRI activation 214 pattern. While the cortical innervation of dopaminergic, serotonergic and noradrenergic nuclei 215 is partially overlapping, there are also considerable differences between neuromodulatory 216 systems3. To test if the expression pattern of noradrenergic transporters explains unique 217 variance in the observed pupil-linked BOLD maps over and above the other neuromodulatory 218 systems, we estimated multiple regression analyses. Specifically, we compared a base model 219 (including serotonergic and dopaminergic transporter expression as predictors) to a full model 220 additionally including a predictor for the noradrenergic transporter (Table 1). Importantly, we 221 found the full model substantially outperformed the base model, suggesting a unique 222 association of noradrenergic transporter expression and pupil-linked BOLD activation (Δχ²(df 223 = 1) = 42909.5; ppermutation-corr < 0.001). 224 225 Table 1: Multiple regression predicting pupil-linked BOLD activation based on 226 neuromodulatory transporter maps 227 Predictor Estimate Standard error t-value p NAT 0.18997 0.0009112 208.48 <0.001 5HTT 0.084247 0.00087668 96.098 <0.001 DAT 0.20462 0.00081556 250.89 <0.001 Note: NAT, 5HTT, DAT refer to PET-derived noradrenergic, serotonergic and dopaminergic transporter 228 expression maps. To obtain a single noradrenergic predictor, the Hesse et al. and Ding et al. maps were 229 averaged. 230 231 Taken together, these analyses suggest that key ascending arousal-promoting systems, 232 including noradrenergic neuromodulation, contribute to the observed pupil-linked subcortical 233 and action-mode network activation. 234 235 236 237 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 10 10 238 Figure 5. Voxel-wise association of pupil-linked BOLD activation patterns with PET-derived 239 maps of noradrenergic transporter (a–b) and GABAergic receptor (c–d) distribution. (a) 240 Relative distribution of noradrenergic transporters based on69. (b) Brain regions activating when pupils 241 dilate overlap with noradrenergic transporter distribution based on two alternative PET maps. (c) 242 Relative distribution of GABAA receptors based on70. (d) Brain regions activating when pupils dilate 243 show no systematic overlap with GABAergic receptor distribution based on two alternative PET 244 maps. Observed correlation coefficients are tested against a reference distribution obtained using 245 phase-randomization permutation. 246 247 Locus coeruleus is functionally connected to action -mode network hubs 248 So far, our analyses suggest that pupil dilation, likely reflecting neuromodulatory 249 drive, is associated with brainstem and action-mode-network activations. Moreover, the 250 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 11 11 activated regions closely overlap with noradrenergic receptor distributions as derived from 251 PET atlases. These observations prompted us to test if locus coeruleus and action-mode 252 network activation were directly connected. 253 Using dedicated neuromodulation-sensitive MRI49, we estimated person-level proxies 254 for locus coeruleus integrity as well as a group image that visualizes the spatial location of the 255 locus coeruleus adjacent to the fourth ventricle (Figure 6 and osf.io/4t5hj). Notably, the locus 256 coeruleus-related hyperintensity was adequately captured by a previously generated high-257 confidence mask of this structure (75; available at: https://osf.io/sf2ky/). 258 After this validation step, we used the derived coordinates overlapping with the 259 masked region as seed region for trial-level whole-brain functional connectivity analyses, 260 probing which region’s activity co-fluctuated with the locus coeruleus during the conditioned 261 oddball task76,77. A bilateral anterior insula (T = 8.93, pFWE-corr < 0.001 MNI: –38 12 –10; 262 Figure 6) and thalamus (T = 6.79, pFWE-corr < 0.001 MNI: –6.5 –22.5 –2.5) cluster survived 263 correction for multiple comparisons, in addition to a broad cerebellar and pontine activation. 264 The un/thresholded whole-brain contrast maps are available at osf.io/4t5hj. These results 265 point to a co-fluctuation of activity in the locus coeruleus and key action-mode network hubs, 266 in line with known direct anatomical projections9, as well as our pupil–fMRI and 267 noradrenergic transporter findings. 268 However, the functional connectivity analyses do not speak to the directionality of this 269 association (brainstem→action-mode network vs . action-mode network→brainstem). Thus, 270 we ran a separate set of analyses, using the TR-level BOLD timeseries of the locus coeruleus 271 and anterior insula (i.e., on a sample level, averaged over voxels). We submitted these to a 272 cross-correlation analysis, which measures the temporal relationship between two signals by 273 quantifying how one variable aligns with shifts in another over time78, allowing us to assess 274 the temporal precedence of brainstem vs. action-mode network activation. First, we observed 275 a strong coinciding activation (mean cross-correlation coefficient at lag zero: 0.363; T = 276 11.892), replicating our functional connectivity findings, that stood out above the general 277 association of the two signals (mean cross-correlation coefficient across all lags: 0.217). Note 278 that to better judge the temporal relation of activations, we removed this overall background 279 association by demeaning before group-level cluster-statistics (Figure 6). 280 Adding to our connectivity analyses, we found that anterior insula activations 281 preceded and predicted subsequent locus coeruleus activations by up to four seconds (mean 282 cross-correlation coefficient at lag –4 s: 0.264; T = 4.69; at lag –2 s: 0.255; T = 2.896; cluster 283 permutation test: Tsum = 22.103; pcluster-corr ≤ 0.002; cluster including lags: (–4)–(+2) s). We 284 also observed that preceding locus coeruleus activity was followed by action mode network 285 activation two seconds later (mean cross-correlation coefficient at lag +2 s: 0.25; T = 2.626), 286 suggesting a potentially slightly later recurrent connection (brainstem↔action-mode network; 287 contrast of cross-correlation coefficient at lags [–4 to –2; anterior insula leading] vs [+2 to +4; 288 locus coeruleus]: T = 2.421; p = 0.018). While these analyses do not allow for causal 289 interpretations, they indicate that anterior insula activity may recruit noradrenergic 290 neuromodulation to facilitate subsequent action-mode network activation5,6. Taken together, 291 we observed that brainstem activation was functionally connected with action-mode network 292 activity and that this crosstalk may be initiated by the anterior insula, potentially helping 293 participants to efficiently respond to unexpected stimuli2. 294 295 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 12 12 296 Figure 6. Locus coeruleus MRI (a) guides whole-brain functional connectivity analyses (b–c), 297 revealing brainstem activation is coupled to and precedes action-mode network activity (d–e). 298 (a) Neuromodulation-sensitive MRI reveals locus coeruleus-related hyperintensity bordering the 299 fourth ventricle that is accurately captured by a high-confidence volume of interest (red;75). (b) Using 300 this region as seed for whole-brain task-based functional connectivity shows locus coeruleus–anterior 301 insula co-fluctuations. (d–e) Cross-correlations replicate the locus coeruleus–anterior insula 302 association and suggest this is initiated by the anterior insula, showing a temporal precedence of up to 303 4 s (contrast of lags [(–4)–(–2)] vs [(+2)–(+4)]). 304 305 Locus coeruleus and action -mode network support implicit learning of task structure 306 To investigate the implications of locus coeruleus–action-mode network interactions 307 during the conditioned oddball task, we analyzed behavioral data. As participants showed 308 high overall accuracy (group mean >0.95), our analyses focused on reaction time dynamics. 309 Participants entered the task without prior knowledge of its statistical structure, yet mean 310 reaction times systematically varied with trial likelihood across the fMRI session (Figure 7a; 311 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 13 13 F = 173.83; p < 0.001; mean reaction times per condition: stay-standard: –0.437±0.021, 312 switch-to standard: 0.173±0.031, stay -oddball: 0.442±0.043, switch -to oddball: 0.602±0.040; 313 note that all reaction times are expressed in Z-values after log-transformation and within-314 person standardization). This pattern suggests that participants prioritized responding quickly 315 to frequent trial types at the cost of slower responses to rare events. Supporting this, a strong 316 negative association of the reaction times for stay-standard and switch-to-oddball conditions 317 indicates a performance trade-off between these categories (Figure 7b; r =–0.795; p < 0.001). 318 We thus assumed that participants implicitly learned the statistical structure of the novel task 319 environment and adapted their behavior, which implies that this reaction time pattern should 320 emerge over the course of the experiment. Combining behavioral data from both repetitions of 321 the conditioned oddball task, we observed that early in the experiment, responses were 322 generally slow and did not differentiate between frequent and rare trials (Figure 7c). However, 323 as the experiment progressed, reaction times to common stimuli became faster, while rare 324 oddballs elicited increasingly slower responses. Accordingly, we found that the difference in 325 reaction times between switch-to-oddball and stay-standard categories increased with time, 326 suggesting implicit learning (Figure 7d; correlation of reaction-time difference and time bin 327 on a group level: r = 0.509; p = 0.003). This learning pattern detected on a group level was 328 also evident within participants (dependent samples t-test of within-participant time-bin beta 329 parameters against zero: T = 3.194, p = 0.002; mean beta parameter: 0.0097±0.00035; Figure 330 S1). Collectively, these results indicate that participants, initially unfamiliar with the task, 331 gradually internalized its structure and adjusted their behavior accordingly. 332 333 334 Figure 7. Participant focus their behavior on more likely trial categories (a) at the expense of less 335 likely categories (b). This bias emerges across the course of the experiments (c) indicating 336 implicit learning of the task structure (d). Day 1 indicates the pupillometry-EEG assessment, while 337 day 2 indicates the pupillometry-fMRI oddball assessment. (a–b) show day 2 data. (c–d) include data 338 of both days. 339 Finally, we combined imaging and behavioral data to probe if individual differences in 340 markers of the noradrenergic system and its cortical target regions explained implicit learning 341 performance. Specifically, we combined dedicated structural locus coeruleus imaging (Figure 342 6a and Figure 8a left), task-related pupil dilation (cf. Figure 2b) and locus coeruleus-related 343 anterior insula activation (cf. Figure 3a) in a multivariate partial least-squares correlation with 344 condition-wise reaction times ( stay-standard; switch-to standard; stay-oddball; switch-to 345 oddball; Figure 7 and Figure 8a right). We identified one reliable latent component (r = 0.488; 346 ppermutation-corrected = 0.001). Participants with stronger task-related anterior insula activation 347 (loading onto latent variable: 0.752; bootstrap ratio: 3.402 [interpreted analogous to Z-348 values]), higher locus coeruleus integrity (loading: 0.491; bootstrap ratio: 2.306) and task-349 related pupil dilation (loading: 0.44; bootstrap ratio: 1.873; trend level association) showed 350 more pronounced differentiation in reaction times between frequent and rare stimuli (cf. 351 Figure 8b). In particular, these participants tended to show quicker responses for frequent trial 352 categories (loadings; stay-standard: –0.468; switch-to-standard: –0.396) at the expense of 353 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 14 14 slower responses for infrequent trial categories (loadings; stay-oddball: 0.364, switch-to-354 oddball: 0.701)—recapitulating the implicit-learning pattern we have identified in behavioral 355 analyses (Figure 7c). Post-hoc age group comparisons revealed that younger and older adults 356 showed comparable latent behavioral (T = –1.36; p = 0.179) and neural scores (T = –0.209; p 357 = 0.835) and that the latent association did not differ across groups (Z = 1.111; p = 0.133). 358 As a sensitivity analysis, we replicated partial least squares-correlation findings using 359 a simpler analytical strategy (r = 0.466; p < 0.001; by averaging over standardized neural 360 indicators (i.e., insula BOLD activity, locus coeruleus integrity, and pupil dilation, see PLSC 361 analyses) and correlating this composite score with the difference in reaction times to oddball 362 and standard stimuli). 363 In summary, participants adapted their responses over time to reflect the statistical 364 regularities of the experimental task, demonstrating implicit learning. This implicit learning 365 pattern was stronger expressed in those participants with higher insula activation, higher 366 structural locus coeruleus integrity, and a stronger pupil response to salient stimuli. Together 367 those findings collectively support a role of the locus coeruleus-noradrenergic system in 368 guiding attention and learning to adapt behavior in novel environments3,6,41. 369 370 371 Figure 8. Individual differences in noradrenergic markers (insula BOLD activity, locus 372 coeruleus integrity, pupil dilation) are associated with implicit learning. (a) Multivariate partial 373 least squares correlation links noradrenergic markers to reaction times for frequent and infrequent trial 374 categories (standards, oddballs, respectively), which reflect that participants implicitly learned the 375 statistical regularities of the task. (b) Noradrenergic markers positively contribute to the latent brain 376 variable (e.g., the more activation, the higher scores). On a behavioral level, quicker reaction times for 377 frequent trial categories at the expense of slower reaction times for rare categories are linked to higher 378 latent behavioral scores. (c) Participants with higher values in markers of the noradrenergic system 379 demonstrate more implicit learning, a pattern that is comparable across age groups. 380 381 3. Discussion 382 This study investigated the role of the locus coeruleus and its cortical targets in how 383 younger and older participants adapt their behavior in dynamic environments with novel 384 observations. We took advantage of repeated assessments of an experimental task shown to 385 modulate noradrenergic activity in primates44 and a dedicated multimodal imaging protocol to 386 identify neuromodulatory contributions to implicit learning. 387 We found that salient expectation-violating stimuli dilated participants’ pupils, 388 indicating heightened neuromodulatory activity, and triggered a shift from default-mode to 389 action-mode network activation. Combining pupil and BOLD data, we observed that 390 neuromodulatory brainstem–midbrain as well as action-mode network regions scaled their 391 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 15 15 activity with pupil size. This suggests an interplay between the noradrenergic system and the 392 action-mode network, which we confirmed using PET-derived transporter maps and locus 393 coeruleus-centered functional connectivity analyses. Behavioral data showed that participants 394 implicitly learned the statistical structure of the novel task environment and focused their 395 behavior on more prevalent trial categories, resulting in quicker responses at the expense of 396 slower responses to rare oddballs. Linking brain and behavior, we found that this implicit 397 learning pattern was more pronounced in participants with higher levels of markers associated 398 with the noradrenergic system and its cortical targets. Taken together, this study suggests 399 noradrenergic neuromodulation may upregulate the action-mode network to facilitate 400 (implicit) learning for behavioral adaptation. 401 402 Fear conditioned stimuli evoke robust neuromodulatory responses 403 To increase recruitment of the locus coeruleus, a subset of our experimental stimuli 404 was pre-conditioned before the main task by repeatedly pairing them with an aversive 405 outcome40,44–46. Conditioned stimuli (CS+) elicited greater pupil dilation than perceptually 406 matched control stimuli (CS–), suggesting increased neuromodulatory drive. Moreover, these 407 effects remained stable across age groups and assessment days, highlighting the robustness of 408 the conditioned response. Our findings are in line with earlier work in rodents demonstrating 409 increased locus coeruleus spiking79 and noradrenergic axonal activation45 during conditioning, 410 assumed to support fear learning. 411 412 Noradrenergic neuromodulatory activation is coupled with the action-mode network 413 Building on this, during the conditioned oddball task, salient stimuli dilated 414 participants’ pupils and activated the action-mode network, including the anterior insula, 415 middle cingulate cortex, thalamus, and cerebellum. This aligns with prior findings implicating 416 the brain’s action-mode of function with states of externally focused attention, heightened 417 arousal, and the processing of action-relevant bottom up signals22,25. Importantly, these 418 activations were accompanied by default-mode network deactivation, suggesting a functional 419 transition from introspective to externally oriented processing6. This is in line with the 420 proposed antagonism of the default-mode and action-mode networks22 and the more general 421 noradrenergic role in facilitating dynamic reorganizations of target neural networks (also 422 termed network reset5). 423 Combined analyses of pupil and BOLD time series revealed that moment-to-moment 424 fluctuations in pupil size systematically co-varied with activation in the action-mode network 425 (anterior insula, middle cingulate cortex, thalamus). This supports previous intracranial work 426 in humans linking anterior insula activations to variations in pupil size30 and extends them to 427 a network level. In addition, a brainstem–midbrain cluster encompassing the locus coeruleus 428 and other neuromodulatory nuclei activated when pupils dilated, in line with recent animal 429 research55,60–62, suggesting that pupil dilation reflects broad neuromodulatory engagement 430 across cortical and subcortical regions63,64. 431 An analysis of the spatial overlap between publicly available PET-derived 432 noradrenergic transporter maps and pupil-linked BOLD activation further underscores a role 433 for noradrenergic neuromodulation in regulating these activation patterns56,65–69. We found a 434 robust positive correlation, replicating across independent transporter maps, suggesting that 435 brain regions with increased activity during pupil dilation overlap with areas rich in 436 noradrenergic transporters. Negative control analyses with GABAergic receptor maps70,71 437 showed no such association, further strengthening the specificity of the association between 438 pupil size, locus coeruleus and action-mode network activation. In line with the broad 439 brainstem–midbrain activation, also serotonergic73 and dopaminergic74 transporter maps were 440 positively coupled with the pupil-linked BOLD pattern 63,64. Crucially, however, noradrenergic 441 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 16 16 transporter expression explained unique variance in pupil-linked BOLD activation over and 442 above the other neuromodulatory systems. 443 To more precisely map locus coeruleus activation to its cortical targets, we employed 444 functional connectivity analyses guided by dedicated neuromodulation-sensitive MRI 75. 445 These analyses revealed that locus coeruleus activity co-fluctuated with action-mode network 446 hubs, particularly the anterior insula and thalamus, consistent with established direct 447 projections in primates9. Cross-correlation analyses further elucidated the temporal dynamics 448 of this relationship, suggesting that the anterior insula may recruit locus coeruleus activation 449 up to four seconds later. This finding together with evidence of a recurrent connection support 450 the notion that the insula-initiated recruitment of noradrenergic signaling may facilitate a 451 cortical shift toward an action-oriented state, optimizing adaptive responses to environmental 452 changes5,6,22. In this circuit, the anterior insula likely provides crucial contextual signals 453 without which the locus coeruleus would be blind to salient events that require gain-454 modulation2. Our findings are in broad agreement with a recent rodent study in which direct 455 stimulation of the cingulate cortex, another action-mode network hub region, elicited pupil 456 dilation through locus coeruleus activation, suggesting top-down recruitment of brainstem 457 neuromodulation80. However, we cannot yet rule out the possibility that differences in the 458 shape of subcortical and cortical hemodynamic responses may contribute to the observed 459 effects, though previous findings do not point in this direction81. 460 Notably, beyond the action-mode network, also the salience network includes hub 461 regions in the anterior insula and cingulate cortex82, and in some atlases networks with these 462 names are overlapping24. Similar to the action-mode network, salience network activity has 463 been linked to ascending neuromodulatory systems, including noradrenaline83. While our 464 study cannot conclusively disentangle the spatially adjacent and functionally linked networks, 465 the more posterior cingulate and more dorsal anterior insula activation favors an action-mode 466 network involvement (22,82, see Figure 3). 467 468 Locus coeruleus and action -mode network together support implicit learning 469 Finally, we examined behavioral adaptations over the course of the experiments. 470 Participants, initially unfamiliar with the task structure, gradually optimized their reaction 471 times to reflect stimulus probabilities, indicating implicit learning. Optimized reaction times 472 may arise from preparatory movement initiation, resulting in quicker responses for expected 473 stimuli, but slower responses for unexpected stimuli. Crucially, individuals with greater MRI-474 indexed locus coeruleus integrity, stronger task-related anterior insula activation, and more 475 pronounced pupil dilation showed greater differentiation in reaction times between frequent 476 and infrequent stimuli. This suggests that the locus coeruleus and its cortical targets play a 477 key role in learning and adapting to novel environments, matching previous findings from 478 rodents demonstrating that noradrenergic activity is causal for task execution and 479 optimization84. In addition, our observations are in line with a recent study in humans relating 480 low MRI-indexed locus coeruleus integrity to diminished practice effects 85. These result 481 moreover support and extend earlier work using explicit cognitive assessments to link 482 noradrenergic markers to attention38,40, learning, and memory41,43. 483 Post-mortem research identified the locus coeruleus as one of the starting points of 484 Alzheimer’s-related tau accumulation36, which may dysregulate noradrenergic activity with 485 increasing age31,34 and impair late-life cognition13,33. Here we found that a latent measure 486 reflecting locus coeruleus integrity as well as pupil dilation and insula activation to salient 487 stimuli did not differ between younger and older adults. This may indicate that externally-488 driven noradrenergic responses may be comparatively less impaired in aging38,86. In addition, 489 we found that this noradrenergic latent variable similarly explained implicit learning across 490 age groups, highlighting its behavioral relevance over the adult lifespan. 491 492 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 17 17

493 By combining neuromodulation-sensitive MRI, concurrent pupillometry–fMRI, and 494 PET-derived neuromodulatory transporter maps, our study suggests the noradrenergic locus 495 coeruleus may contribute to a transition from default-mode to action-mode network 496 activation. This mechanism may support adapting behavior to novel environments, 497 reinforcing the importance of neuromodulatory systems in attention, learning and memory. 498 499 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 18 18 4. Material and methods 500 501 Study design 502 Data were collected as part of a larger project investigating neural and cognitive 503 correlates of age-related differences in the noradrenergic system (for details, see38). 504 Participants attended three consecutive days of lab-based testing, with the conditioned oddball 505 task administered on the second and third days (Figure 1). On the second day, concurrent 506 pupillometry and electroencephalography (EEG) data were recorded during task performance, 507 while third-day assessments incorporated MRI and pupillometry measurements. To enhance 508 the salience of oddball stimuli, each assessment day included a preceding fear conditioning 509 session, during which the oddball stimuli were repeatedly paired with aversive electrical 510 stimulation. The institutional review board of the German Psychological Association 511 approved the study protocol. 512 513 Participants 514 The final sample included 39 younger adults (Y As; mean age ± standard deviation: 515 25.23 ± 3.23 years; range: 20.17–31 years) and 38 older adults (OAs; mean age: 70.61 ± 2.71 516 years; range: 65.50–75.92 years), all of whom were male. Two younger adults were excluded 517 from the study after the first assessment day due to low-quality pupil data. Six participants 518 (three per age group) did not participate in the fMRI part of the study (n = 4) or had 519 incomplete eye tracking data (n = 2), leaving a sample of n = 71 for fMRI–pupillometry 520 analyses. All participants were healthy, right-handed, MRI -compatible, fluent German 521 speakers with normal or corrected-to-normal vision, for further characteristics, see 38. All 522 participants provided written informed consent and were reimbursed for their participation. 523 Exclusion criteria included the use of centrally active drugs, particularly medications affecting 524 the noradrenergic system (e.g., beta-blockers). 525 526 Experimental procedures and stimuli 527 Fear conditioning 528 Participants underwent fear conditioning on each testing day to increase the salience of 529 experimental stimuli and their likelihood of eliciting locus coeruleus activation 40,44–46. During 530 this phase, a sinusoidal luminance pattern (Gabor patch; CS+) with either horizontal or 531 vertical orientation (0° or 90°) was paired with an aversive electrical shock (US) on 80% of 532 trials. A perceptually matched control stimulus (CS–) with the alternate orientation (vertical 533 or horizontal) was never paired with the shock. Stimulus orientation–reinforcement pairings 534 remained consistent for each participant across days and were counterbalanced within age 535 groups (younger adults [YA]: 21:18; older adults [OA]: 20:18). 536 Each conditioning session comprised 40 trials, with 20 presentations each of CS+ and 537 CS– in pseudorandomized order across assessment days. Trials began with a one-second 538 fixation cross, followed by a two-second visual stimulus (CS+ or CS–). Shocks were 539 delivered for 0.5 seconds immediately after CS+ offset via a bipolar current stimulator (DS5; 540 Digitimer) connected to a ring electrode affixed to participant’s left or right index finger. 541 Hand assignment was counterbalanced within age groups (Y A: 19:20; OA: 19:19). Each trial 542 concluded with a six-second inter-trial interval, allowing pupil responses to return to baseline. 543 Participants selected an individually calibrated stimulation intensity deemed 544 unpleasant but not painful before each session38. Throughout all conditioning sessions, pupil 545 dilation and gaze position were recorded, and external distractions were minimized. 546 A procedural error occurred for one older participant, resulting in a reversal of 547 stimulus orientation–shock assignment between the first and second assessment days. This 548 error reduced the likelihood of detecting differences between conditions ( CS+/CS–; i.e., 549 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 19 19 worked against our hypothesis), but the participant’s data were retained for analysis to 550 maintain a larger sample size. 551 552 Conditioned oddball task 553 On the second and third assessment days (EEG–pupillometry and fMRI–pupillometry, 554 respectively), participants completed a modified version of a conditioned oddball task 555 previously shown to increase locus coeruleus activity44. 556 Specifically, on each trial participants viewed a Gabor patch in one of three possible 557 orientations (horizontal, diagonal, vertical; [0°, 45°, 90°]). Two of these orientations (0°, 90°) 558 were extensively familiarized during preceding fear conditioning (see above) and other tasks 559 within the larger project (see 38), whereas the third orientation (45°) benefited from relative 560 novelty, hypothesized to enhance neuromodulatory activity87. 561 During the conditioned oddball task, one of the stimulus orientations was presented 562 frequently (standard trials; 0° or 90°), while the two other orientations were rarely presented 563 (oddballs). In total, the task consisted of 160 trials presented in pseudorandomized order that 564 differed over assessment days, with the standard, non-conditioned stimulus (CS–) appearing 565 on 70% of trials (112 trials), while the two oddball stimuli (CS+ and the novel 45° 566 orientation) appeared less frequently, each comprising 15% of trials (24 each). 567 Gabor patches (0°, 45°, and 90° orientation) were displayed for two seconds followed 568 by an inter-trial interval ranging from 2–12 seconds (mean 3.625 seconds; skewed distribution 569 generated with OptSeq2, http://surfer.nmr.mgh.harvard.edu/optseq/). 570 During Gabor patch presentation, participants were instructed to respond as quickly 571 and accurately as possible by pressing a button corresponding to the orientation of each 572 stimulus using the hand not used for fear conditioning (see above). Accuracy and reaction 573 times were recorded for each trial. 574 575 Physiological dat a recording and preprocessing 576 Eye tracking 577 During all experiments, participants’ pupil dilation was measured as an indirect 578 marker of central neuromodulatory activity53–55 alongside gaze position, using a video-based 579 infrared eye tracker (SR Research EyeLink 1000). On the second day of assessments (during 580 EEG), the desktop mount configuration was used, while on the third day (during fMRI), the 581 long-range mount was employed. The systems operated in a monocular configuration with a 582 spatial resolution of up to 0.25° and a sampling rate of 1000 Hz. 583 To minimize head movements, participants were seated with their forehead and chin 584 stabilized at a fixed distance of 53.5 cm from the display during the second day. For the MRI 585 session (third day), the eye tracker was positioned within the scanner bore, and data were 586 collected via a mirror affixed to the head coil, which redirected participants’ view to the 587 stimulus display. 588 Participants were instructed to maintain central fixation throughout the experiments. 589 Before each experiment, the eye tracker was (re)calibrated using a standard 5-point grid. 590 Calibration was considered successful if fixation errors were under 0.5°. 591 Eye tracking data were preprocessed and missing samples were imputed using 592 published standardized routines88. The percentage of imputed samples was logged as an 593

quality measure for potential subsequent participant exclusion (proportion of valid 594 eye tracking samples during the fMRI–pupillometry assessment (mean ± standard error): 595 0.803 ± 0.024). 596 597 Magnetic resonance imaging 598 MRI data were collected using a 3T Magnetom TIM Trio (Siemens Healthcare) with a 599 12-channel head coil. 600 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 20 20 An axial, T1-weighted neuromodulation-sensitive Fast Spin Echo (FSE) sequence was 601 collected to assess markers for locus coeruleus integrity. The following parameters were 602 used41,89: acquisition matrix: 440 × 512, 10 slices, voxel size: 0.5 × 0.5 × 2.5 mm, 20 % gap 603 between slices, repetition time (TR): 600 ms, echo time (TE): 11 ms, flip angle (FA): 120 °, 604 acquisition time: 11:48 min. To increase signal to noise ratios in brainstem imaging, the 605 sequence included four online averages and yielded two images, extracted locus coeruleus 606 parameters were averaged to boost stability of estimates48. 607 In addition, a sagittal T1-weighted Magnetization-Prepared Gradient Echo 608 (MPRAGE) sequence with the following parameters was collected to facilitate co-registration 609 to standard space: acquisition matrix: 256 × 256 × 192, voxel size: 1 mm isotropic, repetition 610 time (TR): 2500 ms, echo time (TE): 4.77 ms, inversion time (TI): 1100 ms, flip angle (FA): 611 10 °, acquisition time: 9:20 min. 612 Finally, a T2*-weighted whole-brain Echo Planar Imaging (EPI) sequence was used to 613 track blood oxygenation level dependent (BOLD)-changes during the conditioned oddball 614 task. The following imaging parameters were used: acquisition matrix: 72 × 72, 36 slices, 450 615 images, voxel size: 3 mm isotropic, repetition time (TR): 2000 ms, echo time (TE): 30 ms, 616 flip angle (FA): 80 °, acquisition time: 15 min. 617 MPRAGE and EPI data were preprocessed using standardized routines as 618 implemented in HeuDiConv90, MRIQC 91,92, and fMRIprep93 and as detailed in the 619 supplementary information. Data were transformed to 2 mm isotropic MNI 152 non-linear 620 2009c asymmetric space and smoothed with an isotropic 8 mm full-width half-maximum 621 (FWHM) kernel for further analyses, following recommendations for smoothing kernels to be 622 at least twice the voxel resolution94. A time series of confounds was derived from head -623 motion estimates, and global signals (within the white matter, the cerebrospinal fluid (CSF) 624 and whole brain mask). These confounds were expanded with the inclusion of temporal 625 derivatives and quadratic terms for each, yielding a total of 36 confounds that were included 626 in within-participant analyses 95,96. 627 628 Physiological data analyses 629 Pupillometry analysis 630 Pupil time series were analyzed using the EEGlab97, eye -EEG98, and fieldtrip99 631 toolboxes. The continuous pupil time series were segmented into trials surrounding 632 presentation of the Gabor patches (for conditioning and oddball experiments). Within-633 participant trial-level analyses then contrasted pupil responses across experimental conditions 634 after baseline correction (–500–0 ms relative to stimulus onset). That is, for conditioning 635 experiments, CS+ and CS– trials were contrasted, while for the oddball experiments standard 636 and oddball trials were compared (i.e., collapsing the two oddball categories) using sample-637 wise independent samples t-tests. Within-participant analyses were repeated over n Bootstraps = 638 50 and the resulting t-value time series were averaged38. Subsequent across-participant 639 analyses tested for the consistency of within-participant effects on a group level (dependent 640 samples t-test against zero), while controlling for multiple comparisons using cluster-based 641 random permutation tests100. For more details, see38. 642 643 Functional magnetic resonance imaging analysis 644 fMRI data were analyzed using the Statistical Parametric Mapping (SPM) 12 645 toolbox101 to identify brain responses to oddball stimuli. Within-participant event -related 646 analyses included stick functions for each stimulus onset (standards, oddballs) convolved with 647 a canonical hemodynamic response function as well as its temporal and dispersion 648 derivatives102. General linear models contrasted BOLD responses to standard and oddball 649 stimuli (aggregating over the two oddball categories to increase power). Subsequent across-650 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 21 21 participant analyses evaluated the consistency of within-participants effects on a group level 651 with family-wise error correction for multiple comparisons. 652 Overlap of observed group-level effects with several published functional network 653 atlases (see Figure 3,23,58,59) was evaluated using Dice coefficients and permutation tests, as 654 implemented in the Network Correspondence Toolbox24. For overlap estimations, statistical 655 maps were thresholded at T = 3. 656 Follow-up within-participant functional connectivity analyses first calculated trial -by-657 trial activation patterns for each voxel76 and next assessed which brain regions’ time series 658 correlated with the locus coeruleus77. For this, a locus coeruleus consensus mask was applied 659 as seed region75 that reliably segmented the locus coeruleus-related hyperintensity in this 660 study, as shown in the results. That is, our functional connectivity analyses were centered on 661 the region in which we localized the locus coeruleus using dedicated high-resolution 662 structural scans. 663 We detected a reliable association of locus coeruleus and anterior insula time series. To 664 further characterize the temporal properties of this association, we estimated cross-665 correlations between TR-level locus coeruleus and insula time series on a participant level 666 (using crosscorr in Matlab;30). To this end, we averaged BOLD signals across voxels within 667 the locus coeruleus consensus mask75, as well as anterior insula functional-connectivity 668 cluster. Cluster-based random permutation tests probed the consistency of the participant-level 669 cross-correlation estimates on a group level after de-meaning100. 670 671 Structural magnetic resonance imaging analysis 672 Neuromodulation-sensitive FSE images were analyzed using a previously established 673 semi-automatic procedure to extract proxies for locus coeruleus integrity41,75,89 using 674 Advanced Normalization Tools (version 2.3.3.; ANTs 103,104). 675 In brief, first MPRAGE scans were linearly resampled to 0.5 mm isotropic resolution 676 before estimating a whole-brain group template (six iterations; 677 antsMultivariateTemplateConstruction, including N4BiasFieldCorrection ). Next, native-678 space FSE scans were non-linearly aligned to within-person temp late-space MPRAGE images 679 using antsRegistrationSyNQuick to facilitate subsequent brainstem template construction 680 based on the outputs of this transformation (six iterations; 681 antsMultivariateTemplateConstruction, including N4BiasFieldCorrection ). Finally, the 682 brainstem template was moved to 0.5 mm linear MNI space using the whole-brain template as 683 an intermediate target (antsRegistration, facilitated by a custom coregistration mask). All 684 transformation matrices were then concatenated and applied to individual neuromodulation-685 sensitive FSE scans, bringing them directly from native to MNI space in a single step 686 (antsApplyTransforms ). 687 To extract proxies for locus coeruleus integrity, standard space FSE images were 688 masked using a high-confidence locus coeruleus consensus mask75 that captured the observed 689 locus coeruleus hyperintensity well (see Figure 6). To normalize intensity values for across-690 participant analyses, scans were also masked with a pontine reference area mask75. The peak 691 voxel intensity for each slice of each region of interest was then automatically identified. 692 Finally, intensity ratios for each slice of the locus coeruleus were computed as a measure of 693 structural integrity (see Equation 1). For each slice, these ratios were derived by dividing the 694 difference between the peak intensity in the locus coeruleus and the corresponding reference 695 region by the peak intensity of the reference region, following previously established 696 methods41,75,89. 697 698 Equation 1: 𝐿𝑜𝑐𝑢𝑠𝐶𝑜𝑒𝑟𝑢𝑙𝑒𝑢𝑠𝑅𝑎𝑡𝑖𝑜= max(𝐿𝑜𝑐𝑢𝑠𝐶𝑜𝑒𝑟𝑢𝑙𝑒𝑢𝑠) – max(𝑅𝑒𝑓𝑒𝑟𝑒𝑛𝑐𝑒) max (𝑅𝑒𝑓𝑒𝑟𝑒𝑛𝑐𝑒) 699 700 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 22 22 For subsequent analyses, extracted slice-wise values were averaged over hemispheres 701 and then the peak value across slices was selected, while excluding the two most rostral slices 702 to avoid unreliable intensity values at the edge of the acquisition41. 703 704 Behavioral data analys is 705 In the conditioned oddball task, participants were instructed to press a specific button 706 corresponding to the orientation of the presented Gabor stimulus. The majority of trials (70%) 707 featured the standard stimulus, creating a high probability that a standard stimulus would 708 follow another standard stimulus. We assumed that due to this consistency participants would 709 build up a tendency (bias) to respond with the corresponding key. We termed these stay-710 standard trials. Occasionally, this expectation was violated by the presentation of one of the 711 oddball stimuli, requiring participants to switch response keys (switch-to-oddball trials). 712 Following an oddball stimulus, most trials returned to the standard stimulus (switch-to-713 standard), while less frequently, another oddball appeared (stay-oddball ). Following this 714 reasoning, we estimated mean reaction times for each of these four trial categories for each 715 participant (2 (stay/switch) × 2 (standard/oddball)) after log-transformation and Z-716 standardization of reaction times across trials. 717 Crucially, participants began the task without prior knowledge of the statistical 718 structure or trial distribution. The extent to which participants adjusted their behavior over 719 time in response to the predictability of the stimuli is thus interpreted as an implicit measure 720 of (task structure) learning. To quantify if participant indeed learned the statistical structure of 721 the task, we down-sampled the trial-level reaction-time time series of the repeated oddball 722 experiments (320 trials in total, two repetitions of 160 trials each; cf. Figure 7). That is, we 723 averaged reaction times over 10 adjacent trials within the four trial categories, yielding 32 724 reaction-time time bins, which we contrasted across trial categories. Finally, we evaluated if 725 participants over the course of the experiment developed a tendency to preferentially focus on 726 one trial category at the expense of another. To this end, we probed if the contrast of reaction 727 times (switch-to-oddball – stay-standard) increased over time. We repeated these analyses on 728 a group level (correlation of group-averaged reaction time contrast and time bin) and on a 729 single-participant level (linear regression predicting reaction time contrasts by time bin). 730 Finally, participant-level beta esti mates (quantifying implicit learning) were contrasted against 731 zero using dependent-samples t-test to evaluate the consistency of implicit learning on a 732 group level. 733 734 Multimodal data analyses 735 Combined pupillometry –functional magnetic resonance imaging analyses 736 Presentation of oddball stimuli elicited pupil dilation and distributed cortical 737 activation, including the action-mode network. To test which brain regions’ activation scaled 738 with pupil-indexed neuromodulation, we estimated a separate set of general l inear models. 739 First, we down-sampled each participant’s pupil time series to the fMRI temporal resolution 740 (TR = 2 s) and added it as regressor to the participant-level models 105. The same confounds as 741 in unimodal fMRI analyses were applied (see above), while task regressors were omitted. 742 Subsequent analyses across participants assessed the consistency of within-participant effects 743 at the group level, applying family-wise error correction to account for multiple comparisons. 744 These analyses were restricted to participants with high-quality pupil data (>50% valid non-745 imputed samples; 64/71 participants88). 746 747 Association of pupillometry –functional magnetic resonance imaging patterns and PET-748 derived transporter maps 749 Combined pupillometry–fMRI analyses showed that brainstem –midbrain and insula 750 activation correlated with pupil size variations, suggesting neuromodulatory influences. To 751 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 23 23 more directly test this, we leveraged publicly available maps of neuromodulatory transporter 752 distributions56. PET -derived maps were resampled to the spatial resolution of our pupil–fMRI 753 activation patterns (2 mm MNI-ICBM 152 non-linear 2009 version c space). Next, we probed 754 the voxel-wise association of noradrenergic transporter expression and pupil-linked BOLD 755 activation using two alternative transporter maps using Pearson’s correlations56,65–69. To judge 756 the reliability of these effects, we ran two negative control analyses, using two maps of 757 GABAergic receptor distribution acquired from two different sources70,71 as well as two 758 comparative analyses, using maps of the dopaminergic 74 and serotonergic transporter73 759 distribution. For each of these, observed effects were contrasted to reference distributions of 760 correlation coefficients estimated with permuted pupillometry-fMRI data while preserving the 761 spatial autocorrelation of the observed data. Specifically, we applied a phase-randomization 762 approach, which retains the power spectrum of the original data while randomizing its phase 763 in the Fourier domain106. This method generates surrogate datasets that maintain the 764 autocorrelation properties of the original data (i.e., the pupil–fMRI activation map) while 765 breaking voxel-wise correspondences with t he PET-derived transporter/receptor maps. We 766 computed correlation coefficients between the shuffled pupil-linked BOLD activation pattern 767 and the transporter/receptor maps across 10000 iterations, forming a null distribution, to 768 which we compared the observed correlation coefficients. 769 Finally, to determine if the voxel-wise noradrenergic transporter expression explained 770 variance in pupil-linked BOLD patterns ( PupilBOLD), we used multiple regression analyses 771 (using fitglm in Matlab). Specifically, we compared a base model (PupilBOLD ~ 1 + 5HTT + 772 DAT) incorporating the serotonergic (5HTT) and dopaminergic (DAT) transporter expression 773 to a full model including all three neuromodulatory systems (PupilBOLD ~ 1 + 5HTT + DAT 774 + NAT). To obtain a single noradrenergic transporter predictor (NAT), the two available maps 775 were averaged65,69. Differences in model fit between base and full models were evaluated 776 using a likelihood-ratio test and compared to a null distribution obtained from adding shuffled 777 noradrenergic transporter predictors (10000 iterations). 778 779 Combined brain –behavior analyses 780 Our physiological analyses yielded multiple locus coeruleus-related measures (MRI-781 indexed locus coeruleus integrity; pupil-indexed neuromodulation; locus coeruleus -linked 782 action-mode network activation). Given the prominent involvement of the noradrenergic 783 system in learning and memory 3, we next tested if these explained individual differences in 784 implicit learning during the conditioned oddball task, using partial l east squares correlations 785 (PLSC107,108). 786 In short, partial least squares correlation is a multivariate statistical technique used to 787 identify patterns of covariance between two sets of variables. In this study, behavioral partial 788 least squares correlation was applied to explore the relationship between locus coeruleus -789 related measures and task-derived measures of implicit learning. By decomposing the 790 covariance structure with Singular V alue Decomposition, partial least squares correlation 791 identifies latent variables (LVs) that maximize the shared variance between the datasets (i.e., 792 locus coeruleus-related measures and implicit learning). Each latent variable is characterized 793 by a pair of weighted combinations of variables—one from each dataset —allowing for the 794 assessment of how strongly these patterns are expressed across participants. Statistical 795 significance of the latent variables was determined using permutation tests (npermutations = 796 10,000), and the reliability of individual variable contributions was evaluated with 797 bootstrapping (nbootstraps = 10,000). A ratio of a variable’s weight and its bootstrapped standard 798 error, termed bootstrap ratio, (BSR), allows identifying indicators reliably contributing to the 799 latent variable (|BSR| ≥ 1.96, interpreted analogous to Z-scores). Taken together, partial least 800 squares correlation provides a robust method for examining complex, high-dimensional 801 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 24 24 relationships, integrating locus coeruleus-related and learning-related measures. For further 802 details, see 75. 803 The following neural and behavioral variables were used as input to the partial least-804 squares analyses: (1) individual task-related left anterior insula activations, extracted using 805 MarsBar109, averaged over voxels of the cluster that showed reliable associations with pupil-806 linked neuromodulation (Figure 4); (2) individual task-related pupil dilation, extracted using 807 the cluster shown in Figure 2 and averaged over samples; (3) peak locus coeruleus contrast 808 ratios, averaged over hemispheres; and (4–8) individual reaction times for each trial category, 809 2 (stay/switch) × 2 (standard/oddball), averaged over trials after log-transformation and Z-810 standardization, as shown in Figure 7. All input variables were standardized across 811 individuals, and participants with absolute values >3 were dropped before analyses (n = 3). 812 Post-hoc independent samples t-tests probed age differences in the latent brain and behavioral 813 variables. As a sensitivity analysis, we replicated partial least squares-correlation findings 814 using a simpler analytical strategy that averaged over standardized neural indicators (i.e., 815 calculated composite scores based on (1)-(3), see PLSC analyses), estimated the difference in 816 response times for oddball and standard trials, and then correlated these measures. 817 818 819 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 25 25

820 MW-B received support from the German Research Foundation (DFG, WE 4269/5-1) 821 and the Jacobs Foundation (Early Career Research Fellowship 2017–2019). 822 MM’s work was supported by an Alexander von Humboldt fellowship and by National 823 Institutes of Health grant R01AG025340 and R01AG080652. 824 During the work on this article, MJD was a member of the International Max Planck 825 Research School on Computational Methods in Psychiatry and Ageing Research (IMPRS 826 COMP2PSYCH, https://www.mps-uclcentre.mpg.de/comp2psych. Participating institutions: 827 Max Planck Institute for Human Development, University College London). At the time of 828 writing, TL was a PhD student in the Max Planck School of Cognition. 829 Acknowledgment is made to the donors of the ADR A2024006F, a program of the 830 BrightFocus Foundation, for support of this research (MJD). 831 This work was in part conducted at the Max Planck Dahlem Campus of Cognition 832 (MPDCC) of the Max Planck Institute for Human Development, Berlin, Germany. 833 834 Author contributions 835 MJD, MM and MWB designed the study. MJD performed the experiments together 836 with a team of research assistants. MJD and TL analyzed the data. MJD wrote the paper. All 837 authors gave conceptual advice and revised the paper. 838 839 Competing interests 840 The authors declare no competing interests. 841 Data availability 842 Participants in this study did not consent to the public sharing of their data. To obtain 843 access to participant-level data, please contact the corresponding author (MJD) with a signed 844 data sharing agreement (template available at: https://osf.io/4t5hj/). Un/thresholded group -845 level data are available at https://osf.io/4t5hj/. 846 847

and correspondence 848 Correspondence and requests for materials should be addressed to Martin J. Dahl. 849 850 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 26 26

851 1. Uddin, L. Q. Cognitive and behavioural flexibility: neural mechanisms and clinical 852 considerations. Nat. Rev. Neurosci. 1–13 (2021) doi:10.1038/s41583-021-00428-w. 853 2. Mather, M., Clewett, D., Sakaki, M. & Harley, C. W. Norepinephrine ignites local hotspots of 854 neuronal excitation: How arousal amplifies selectivity in perception and memory. Behav. Brain 855 Sci. 39, e200 (2016). 856 3. Sara, S. J. The locus coeruleus and noradrenergic modulation of cognition. Nat. Rev. Neurosci. 857 10, 211–223 (2009). 858 4. Dahl, M. J., Mather, M. & Werkle-Bergner, M. Noradrenergic modulation of rhythmic neural 859 activity shapes selective attention. Trends Cogn. Sci. 26, 38–52 (2022). 860 5. Bouret, S. & Sara, S. J. Network reset: A simplified overarching theory of locus coeruleus 861 noradrenaline function. Trends Neurosci. 28, 574–582 (2005). 862 6. Corbetta, M., Patel, G. & Shulman, G. L. The reorienting system of the human brain: From 863 environment to theory of mind. Neuron 58, 306–324 (2008). 864 7. Dayan, P. & Yu, A. J. Phasic norepinephrine: A neural interrupt signal for unexpected events. 865 Netw. Comput. Neural Syst. 17, 335–350 (2006). 866 8. Nassar, M. R. Toward a computational role for locus coeruleus/norepinephrine arousal 867 systems. Curr. Opin. Behav. Sci. 59, 101407 (2024). 868 9. Aston-Jones, G. & Cohen, J. D. An integrative theory of locus coeruleus-norepinephrine 869 function: Adaptive gain and optimal performance. Annu. Rev. Neurosci. 28, 403–450 (2005). 870 10. Thiele, A. & Bellgrove, M. A. Neuromodulation of attention. Neuron 97, 769–785 (2018). 871 11. Servan-Schreiber, D., Printz, H. & Cohen, J. D. A network model of catecholamine effects: 872 Gain, signal-to-noise ratio, and behavior. Science (80-. ). 249, 892–5 (1990). 873 12. Ghosh, S. & Maunsell, J. H. R. Locus coeruleus norepinephrine contributes to visual-spatial 874 attention by selectively enhancing perceptual sensitivity. Neuron 112, 2231-2240.e5 (2024). 875 13. Dahl, M. J., Kulesza, A., Werkle-Bergner, M. & Mather, M. Declining locus coeruleus-876 dopaminergic and noradrenergic modulation of long-term memory in aging and Alzheimer’s 877 disease. Neurosci. Biobehav. Rev. 105358 (2023) doi:10.1016/J.NEUBIOREV.2023.105358. 878 14. Hagena, H., Hansen, N. & Manahan-Vaughan, D. β-adrenergic control of hippocampal 879 function: Subserving the choreography of synaptic information storage and memory. Cereb. 880 Cortex 26, 1349–1364 (2016). 881 15. Eschenko, O. The role of the locus coeruleus in cellular and systems memory consolidation. in 882 Handbook of behavioral neuroscience: Handbook of in vivo neural plasticity techniques (ed. 883 Manahan-Vaughan, D.) vol. 28 327–347 (Elsevier, 2019). 884 16. Hagena, H. & Manahan-Vaughan, D. Oppositional and competitive instigation of hippocampal 885 synaptic plasticity by the VTA and locus coeruleus. Proc. Natl. Acad. Sci. 122, e2402356122 886 (2024). 887 17. Jordan, R. The locus coeruleus as a global model failure system. Trends Neurosci. 47, 92–105 888 (2024). 889 18. Jepma, M. et al. Catecholaminergic Regulation of Learning Rate in a Dynamic Environment. 890 PLOS Comput. Biol. 12, e1005171 (2016). 891 19. Silvetti, M., Vassena, E., Abrahamse, E. & Verguts, T. Dorsal anterior cingulate-brainstem 892 ensemble as a reinforcement meta-learner. PLoS Comput. Biol. 14, e1006370 (2018). 893 20. Li, T., Marble, H., Chen, T., Razmi, N. & Nassar, M. R. Fluctuations in arousal reflect latent 894 state transitions that facilitate behavioral optimization. bioRxiv 2025.02.06.636887 (2025) 895 doi:10.1101/2025.02.06.636887. 896 21. Nassar, M. R., Wilson, R. C., Heasly, B. & Gold, J. I. An Approximately Bayesian Delta-Rule 897 Model Explains the Dynamics of Belief Updating in a Changing Environment. J. Neurosci. 30, 898 12366–12378 (2010). 899 22. Dosenbach, N. U. F., Raichle, M. E. & Gordon, E. M. The brain’s action-mode network. Nat. 900 Rev. Neurosci. 2025 1–11 (2025) doi:10.1038/s41583-024-00895-x. 901 23. Yeo, B. T. T. et al. The organization of the human cerebral cortex estimated by intrinsic 902 functional connectivity. J. Neurophysiol. 106, 1125–1165 (2011). 903 24. Kong, R. et al. A network correspondence toolbox for quantitative evaluation of novel 904 neuroimaging results. Nat. Commun. 2025 161 16, 1–16 (2025). 905 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 27 27 25. Dosenbach, N. U. F., Fair, D. A., Cohen, A. L., Schlaggar, B. L. & Petersen, S. E. A dual-906 networks architecture of top-down control. Trends Cogn. Sci. 12, 99–105 (2008). 907 26. Berridge, C. W. Noradrenergic modulation of arousal. Brain Res. Rev. 58, 1–17 (2008). 908 27. Carter, M. E. et al. Tuning arousal with optogenetic modulation of locus coeruleus neurons. 909 Nat. Neurosci. 13, 1526–1535 (2010). 910 28. Hayat, H. et al. Locus coeruleus norepinephrine activity mediates sensory-evoked awakenings 911 from sleep. Sci. Adv. 6, eaaz4232 (2020). 912 29. Osorio-Forero, A. et al. Infraslow noradrenergic locus coeruleus activity fluctuations are 913 gatekeepers of the NREM–REM sleep cycle. Nat. Neurosci. (2024) doi:10.1038/s41593-024-914 01822-0. 915 30. Kucyi, A. & Parvizi, J. Pupillary dynamics link spontaneous and task-evoked activations 916 recorded directly from human insula. J. Neurosci. 40, 6207–6218 (2020). 917 31. Kelberman, M. A. et al. Diversity of ancestral brainstem noradrenergic neurons across species 918 and multiple biological factors. bioRxiv Prepr. Serv. Biol. (2024) 919 doi:10.1101/2024.10.14.618224. 920 32. Theofilas, P. et al. Locus coeruleus volume and cell population changes during Alzheimer’s 921 disease progression: A stereological study in human postmortem brains with potential 922 implication for early-stage biomarker discovery. Alzheimer’s Dement. 13, 236–246 (2017). 923 33. Mooraj, Z. et al. Toward a functional future for the cognitive neuroscience of human aging. 924 Neuron 113, 154–183 (2025). 925 34. Weinshenker, D. Long road to ruin: Noradrenergic dysfunction in neurodegenerative disease. 926 Trends Neurosci. 41, 211–223 (2018). 927 35. Mather, M. & Harley, C. W. The locus coeruleus: Essential for maintaining cognitive function 928 and the aging brain. Trends Cogn. Sci. 20, 214–226 (2016). 929 36. Braak, H., Thal, D. R., Ghebremedhin, E. & Del Tredici, K. Stages of the pathologic process in 930 Alzheimer disease: Age categories from 1 to 100 years. J. Neuropathol. Exp. Neurol. 70, 960–931 969 (2011). 932 37. Ehrenberg, A. J. et al. Priorities for research on neuromodulatory subcortical systems in 933 Alzheimer’s disease: Position paper from the NSS PIA of ISTAART. Alzheimer’s Dement. 8, 934 21 (2023). 935 38. Dahl, M. J., Mather, M., Sander, M. C. & Werkle-Bergner, M. Noradrenergic responsiveness 936 supports selective attention across the adult lifespan. J. Neurosci. 40, 4372–4390 (2020). 937 39. Kosciessa, J. Q., Mayr, U., Lindenberger, U. & Garrett, D. D. Broadscale dampening of 938 uncertainty adjustment in the aging brain. Nat. Commun. 2024 151 15, 1–18 (2024). 939 40. Lee, T.-H. et al. Arousal increases neural gain via the locus coeruleus–noradrenaline system in 940 younger adults but not in older adults. Nat. Hum. Behav. 2, 356–366 (2018). 941 41. Dahl, M. J. et al. The integrity of dopaminergic and noradrenergic brain regions is associated 942 with different aspects of late-life memory performance. Nat. Aging 2023 1–16 (2023) 943 doi:10.1038/s43587-023-00469-z. 944 42. Bueichekú, E. et al. Spatiotemporal patterns of locus coeruleus integrity predict cortical tau and 945 cognition. Nat. Aging 2024 1–13 (2024) doi:10.1038/s43587-024-00626-y. 946 43. Jacobs, H. I. L. et al. In vivo and neuropathology data support locus coeruleus integrity as 947 indicator of Alzheimer’s disease pathology and cognitive decline. Sci. Transl. Med. 13, 948 eabj2511 (2021). 949 44. Aston-Jones, G., Rajkowski, J., Kubiak, P. & Alexinsky, T. Locus coeruleus neurons in 950 monkey are selectively activated by attended cues in a vigilance task. J. Neurosci. 14, 4467–951 4480 (1994). 952 45. Deitcher, Y., Leibner, Y., Kutzkel, S., Zylbermann, N. & London, M. Nonlinear relationship 953 between multimodal adrenergic responses and local dendritic activity in primary sensory 954 cortices. bioRxiv 814657 (2019) doi:10.1101/814657. 955 46. Wilmot, J. H. et al. Phasic locus coeruleus activity enhances trace fear conditioning by 956 increasing dopamine release in the hippocampus. Elife 12, (2023). 957 47. Astafiev, S. V., Snyder, A. Z., Shulman, G. L. & Corbetta, M. Comment on ‘Modafinil shifts 958 human locus coeruleus to low-tonic, high-phasic activity during functional MRI’ and 959 ‘Homeostatic sleep pressure and responses to sustained attention in the suprachiasmatic area’. 960 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 28 28 Science (80-. ). 328, 309 (2010). 961 48. Forstmann, B. U., De Hollander, G., Van Maanen, L., Alkemade, A. & Keuken, M. C. 962 Towards a mechanistic understanding of the human subcortex. Nat. Rev. Neurosci. 18, 57–65 963 (2016). 964 49. Betts, M. J. et al. Locus coeruleus imaging as a biomarker for noradrenergic dysfunction in 965 neurodegenerative diseases. Brain 142, 2558–2571 (2019). 966 50. Pérot, J.-B. et al. Longitudinal neuromelanin changes in prodromal and early Parkinson’s 967 disease in humans and rat model. bioRxiv 2024.10.22.619619 (2024) 968 doi:10.1101/2024.10.22.619619. 969 51. Watanabe, T., Tan, Z., Wang, X., Martinez-Hernandez, A. & Frahm, J. Magnetic resonance 970 imaging of noradrenergic neurons. Brain Struct. Funct. 224, 1609–1625 (2019). 971 52. Keren, N. I. et al. Histologic validation of locus coeruleus MRI contrast in post-mortem tissue. 972 Neuroimage 113, 235–245 (2015). 973 53. Joshi, S. & Gold, J. I. Pupil size as a window on neural substrates of cognition. Trends Cogn. 974 Sci. 24, 466–480 (2020). 975 54. Grujic, N., Polania, R. & Burdakov, D. Neurobehavioral meaning of pupil size. Neuron 0, 976 (2024). 977 55. Privitera, M. et al. A complete pupillometry toolbox for real-time monitoring of locus 978 coeruleus activity in rodents. Nat. Protoc. 15, 2301–2320 (2020). 979 56. Hansen, J. Y. et al. Mapping neurotransmitter systems to the structural and functional 980 organization of the human neocortex. Nat. Neurosci. 2022 1–13 (2022) doi:10.1038/s41593-981 022-01186-3. 982 57. George, M. S. & Aston-Jones, G. Noninvasive techniques for probing neurocircuitry and 983 treating illness: vagus nerve stimulation (VNS), transcranial magnetic stimulation (TMS) and 984 transcranial direct current stimulation (tDCS). Neuropsychopharmacol. 2010 351 35, 301–316 985 (2009). 986 58. Gordon, E. M. et al. Precision Functional Mapping of Individual Human Brains. Neuron 95, 987 791-807.e7 (2017). 988 59. Ji, J. L. et al. Mapping the human brain’s cortical-subcortical functional network organization. 989 Neuroimage 185, 35–57 (2019). 990 60. Cazettes, F., Reato, D., Morais, J. P., Renart, A. & Mainen, Z. F. Phasic activation of dorsal 991 raphe serotonergic neurons increases pupil size. Curr. Biol. 31, 192–197.e4 (2020). 992 61. Reimer, J. et al. Pupil fluctuations track rapid changes in adrenergic and cholinergic activity in 993 cortex. Nat. Commun. 7, 13289 (2016). 994 62. Joshi, S., Li, Y., Kalwani, R. M. & Gold, J. I. Relationships between pupil diameter and 995 neuronal activity in the locus coeruleus, colliculi, and cingulate cortex. Neuron 89, 221–234 996 (2016). 997 63. Meissner, S. N. et al. Self-regulating arousal via pupil-based biofeedback. Nat. Hum. Behav. 998 2023 1–20 (2023) doi:10.1038/s41562-023-01729-z. 999 64. de Gee, J. W. et al. Dynamic modulation of decision biases by brainstem arousal systems. Elife 1000 6, 1–36 (2017). 1001 65. Hesse, S. et al. Central noradrenaline transporter availability in highly obese, non-depressed 1002 individuals. Eur. J. Nucl. Med. Mol. Imaging 44, 1056–1064 (2017). 1003 66. Belfort-Deaguiar, R. et al. Noradrenergic Activity in the Human Brain: A Mechanism 1004 Supporting the Defense Against Hypoglycemia. J. Clin. Endocrinol. Metab. 103, 2244–2252 1005 (2018). 1006 67. Sanchez-Rangel, E. et al. Norepinephrine transporter availability in brown fat is reduced in 1007 obesity: a human PET study with [11C] MRB. Int. J. Obes. 2019 444 44, 964–967 (2019). 1008 68. Li, C. R. et al. Decreased norepinephrine transporter availability in obesity: Positron Emission 1009 Tomography imaging with (S,S)-[11C]O-methylreboxetine. Neuroimage 86, 306–310 (2014). 1010 69. Ding, Y. S. et al. PET imaging of the effects of age and cocaine on the norepinephrine 1011 transporter in the human brain using (S,S)-[(11)C]O-methylreboxetine and HRRT. Synapse 64, 1012 30–38 (2010). 1013 70. Dukart, J. et al. Cerebral blood flow predicts differential neurotransmitter activity. Sci. Reports 1014 2018 81 8, 1–11 (2018). 1015 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 29 29 71. Nørgaard, M. et al. A high-resolution in vivo atlas of the human brain’s benzodiazepine 1016 binding site of GABAA receptors. Neuroimage 232, (2021). 1017 72. Breton-Provencher, V. & Sur, M. Active control of arousal by a locus coeruleus GABAergic 1018 circuit. Nat. Neurosci. 22, 218–228 (2019). 1019 73. Beliveau, V. et al. A High-Resolution In Vivo Atlas of the Human Brain’s Serotonin System. 1020 J. Neurosci. 37, 120 (2017). 1021 74. Sasaki, T. et al. Quantification of dopamine transporter in human brain using PET with 18F-1022 FE-PE2I. J. Nucl. Med. 53, 1065–1073 (2012). 1023 75. Dahl, M. J. et al. Locus coeruleus integrity is related to tau burden and memory loss in 1024 autosomal-dominant Alzheimer’s disease. Neurobiol. Aging 112, 39–54 (2022). 1025 76. Mumford, J. A., Turner, B. O., Ashby, F. G. & Poldrack, R. A. Deconvolving BOLD activation 1026 in event-related designs for multivoxel pattern classification analyses. Neuroimage 59, 2636–1027 2643 (2012). 1028 77. Rissman, J., Gazzaley, A. & D’Esposito, M. Measuring functional connectivity during distinct 1029 stages of a cognitive task. Neuroimage 23, 752–763 (2004). 1030 78. Box, G. E. P., Jenkins, G. M. & Reinsel, G. C. Time series analysis: forecasting and control . 1031 (Prentice Hall, 1994). 1032 79. Uematsu, A. et al. Modular organization of the brainstem noradrenaline system coordinates 1033 opposing learning states. Nat. Neurosci. 20, 1602–1611 (2017). 1034 80. Salvi, V. et al. Cingulate cortex stimulation drives distinct pupillary responses in rat via 1035 recruitment of noradrenergic neurons in the locus coeruleus. Cereb. Cortex 35, (2025). 1036 81. Prokopiou, P. C. et al. Lower novelty-related locus coeruleus function is associated with Aβ-1037 related cognitive decline in clinically healthy individuals. Nat. Commun. 2022 131 13, 1–14 1038 (2022). 1039 82. Seeley, W. W. The Salience Network: A Neural System for Perceiving and Responding to 1040 Homeostatic Demands. J. Neurosci. 39, 9878–9882 (2019). 1041 83. Hermans, E. J. et al. Stress-related noradrenergic activity prompty large-scale neural network 1042 reconfiguration. Science (80-. ). 334, 1151–1153 (2011). 1043 84. Breton-Provencher, V., Drummond, G. T., Feng, J., Li, Y. & Sur, M. Spatiotemporal dynamics 1044 of noradrenaline during learned behaviour. Nat. 2022 1–7 (2022) doi:10.1038/s41586-022-1045 04782-2. 1046 85. Smegal, L. F. et al. Lower locus coeruleus integrity is associated with diminished practice 1047 effects in clinically unimpaired older individuals. Neurobiol. Aging 152, 13–24 (2025). 1048 86. Lindenberger, U. & Mayr, U. Cognitive aging: Is there a dark side to environmental support? 1049 Trends Cogn. Sci. 18, 7–15 (2014). 1050 87. Duszkiewicz, A. J., McNamara, C. G., Takeuchi, T. & Genzel, L. Novelty and dopaminergic 1051 modulation of memory persistence: A tale of two systems. Trends Neurosci. 42, 102–114 1052 (2019). 1053 88. Kret, M. E. & Sjak-Shie, E. E. Preprocessing pupil size data: Guidelines and code. Behav. Res. 1054

51, 1336–1342 (2019). 1055 89. Dahl, M. J. et al. Rostral locus coeruleus integrity is associated with better memory 1056 performance in older adults. Nat. Hum. Behav. 3, 1203–1214 (2019). 1057 90. Halchenko, Y. O. et al. HeuDiConv — flexible DICOM conversion into structured directory 1058 layouts. J. Open Source Softw. 9, 5839 (2024). 1059 91. Esteban, O. et al. Crowdsourced MRI quality metrics and expert quality annotations for 1060 training of humans and machines. Sci. data 6, 30 (2019). 1061 92. Esteban, O. et al. MRIQC: Advancing the automatic prediction of image quality in MRI from 1062 unseen sites. PLoS One 12, e0184661 (2017). 1063 93. Esteban, O. et al. fMRIPrep: a robust preprocessing pipeline for functional MRI. Nat. Methods 1064 16, 111–116 (2019). 1065 94. Worsley, K. J. & Friston, K. J. Analysis of fMRI Time-Series Revisited—Again. Neuroimage 1066 2, 173–181 (1995). 1067 95. Satterthwaite, T. D. et al. An improved framework for confound regression and filtering for 1068 control of motion artifact in the preprocessing of resting-state functional connectivity data. 1069 Neuroimage 64, 240–256 (2013). 1070 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint LOCUS COERULEUS SUPPORTS IMPLICIT LEARNING 30 30 96. Friston, K. J., Williams, S., Howard, R., Frackowiak, R. S. J. & Turner, R. Movement-Related 1071 effects in fMRI time-series. Magn. Reson. Med. 35, 346–355 (1996). 1072 97. Delorme, A. & Makeig, S. EEGLAB: An open source toolbox for analysis of single-trial EEG 1073 dynamics including independent component analysis. J. Neurosci. Methods 134, 9–21 (2004). 1074 98. Dimigen, O., Sommer, W., Hohlfeld, A., Jacobs, A. M. & Kliegl, R. Coregistration of eye 1075 movements and EEG in natural reading: Analyses and review. J. Exp. Psychol. Gen. 140, 552–1076 572 (2011). 1077 99. Oostenveld, R., Fries, P., Maris, E. & Schoffelen, J. M. FieldTrip: Open source software for 1078 advanced analysis of MEG, EEG, and invasive electrophysiological data. Comput. Intell. 1079 Neurosci. 2011, 1–9 (2011). 1080 100. Maris, E. & Oostenveld, R. Nonparametric statistical testing of EEG- and MEG-data. J. 1081 Neurosci. Methods 164, 177–190 (2007). 1082 101. Friston, K. J., Ashburner, J., Kiebel, S., Nichols, T. & Penny, W. Statistical Parametric 1083 Mapping: The Analysis of Functional Brain Images. Statistical Parametric Mapping: The 1084 Analysis of Functional Brain Images (Academic Press, 2007). doi:10.1016/B978-0-12-372560-1085 8.X5000-1. 1086 102. Friston, K. J. et al. Event-Related fMRI: Characterizing Differential Responses. Neuroimage 7, 1087 30–40 (1998). 1088 103. Avants, B. B., Tustison, N. & Song, G. Advanced Normalization Tools: V1.0. Insight J. 2, 1089 (2009). 1090 104. Tustison, N. J. et al. The ANTsX ecosystem for quantitative biological and medical imaging. 1091 Sci. Reports 2021 111 11, 1–13 (2021). 1092 105. Murphy, P. R., O’Connell, R. G., O’Sullivan, M., Robertson, I. H. & Balsters, J. H. Pupil 1093 diameter covaries with BOLD activity in human locus coeruleus. Hum. Brain Mapp. 35, 4140–1094 4154 (2014). 1095 106. Theiler, J., Eubank, S., Longtin, A., Galdrikian, B. & Doyne Farmer, J. Testing for nonlinearity 1096 in time series: the method of surrogate data. Phys. D Nonlinear Phenom. 58, 77–94 (1992). 1097 107. Krishnan, A., Williams, L. J., McIntosh, A. R. & Abdi, H. Partial Least Squares (PLS) methods 1098 for neuroimaging: A tutorial and review. Neuroimage 56, 455–475 (2011). 1099 108. McIntosh, A. R. & Lobaugh, N. J. Partial least squares analysis of neuroimaging data: 1100 Applications and advances. in NeuroImage vol. 23 (Neuroimage, 2004). 1101 109. Brett, M., Anton, J.-L., Valabregue, R. & Poline, J.-B. Region of interest analysis using an 1102 SPM toolbox. in Human Brain Mapping conference (2002). 1103 1104 .CC-BY-NC 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted May 11, 2025. ; https://doi.org/10.1101/2025.05.10.653174doi: bioRxiv preprint