Behavioral state-dependent noradrenergic dynamics in the primary somatosensory and prefrontal cortices during tactile detection tasks

Norepinephrine, locus coeruleus, pupil-linked arousal, GLM-HMM modeling, the primary somatosensory cortex, the prefrontal cortex, tactile detection task Contact Info: Correspondence should be addressed to: Professor Qi Wang Department of Biomedical Engineering Columbia University ET 351, 500 W. 120 th Street, New York, NY 10027, Phone: 212.854.3657 Email: [email protected] .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint

Animals must integrate sensory information, ignore behaviorally irrelevant stimuli, and respond to behaviorally relevant stimuli to find food, find mates, avoid predators, and ultimately survive. In mammals, these goal-directed behaviors require the coordinated activity of many brain regions, including sensory and prefrontal cortices as well as neuromodulatory brainstem nuclei like the locus coeruleus (LC), which is the brain’s primary source of norepinephrine. Norepinephrine resulting from LC activity and arousal indexed by pupil size exert strong influences on goal- directed behavior. We explored the relationships between pupil size, cortical noradrenergic dynamics, and behavior in a tactile signal detection task. We monitored pupil dynamics and fluorescent GRAB NE signals in somatosensory and prefrontal cortices simultaneously during task execution and found that pupil size and synchronization of GRAB NE signals at baseline were strong predictors of whether animals chose to respond but not baseline cortical GRAB NE levels. We also employed a generalized linear model - hidden Markov model (GLM-HMM) framework to identify distinct, stable behavioral states throughout the task that optimally account for task performance. We found distinct psychometric curves, task-evoked pupil dynamics, and cortical NE dynamics across these behavioral states. .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint Significance Statement Behavioral state strongly shapes goal-directed behavior, which in turn depends on the coordinated activity of distributed brain regions, including the sensory and prefrontal cortices. By simultaneously measuring pupil size and cortical norepinephrine dynamics, and by identifying psychophysically distinct behavioral states during a tactile detection task, this study establishes a mechanistic link between pupil-linked arousal, norepinephrine signaling in somatosensory and prefrontal cortices, and behavioral outcomes. These findings provide new insight into how the locus coeruleus – norepinephrine system regulates perception and decision-making. .CC-BY-NC 4.0 International licenseavailable 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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint

Integrating sensory information, responding to behaviorally relevant stimuli, and ignoring behaviorally irrelevant stimuli is essential to animal survival (Wimmer et al., 2015; Speed and Haider, 2021). Goal-directed behaviors require the coordinated activity of multiple brain regions: thalamus and sensory cortices process and encode stimuli; prefrontal cortex is involved in decision making, planning, and working memory; and neuromodulatory brainstem nuclei dynamically modulate processing in other brain regions through attention and arousal (Pezzulo et al., 2014; Dahl et al., 2022; Cerpa et al., 2023; Jordan and Keller, 2023; Ghosh and Maunsell, 2024; Hansen et al., 2024; Osorio-Forero et al., 2024; Jensen et al., 2025; Kelley et al., 2025). In particular, activity in the locus coeruleus, the brain's primary source of norepinephrine, has been shown to modulate feedforward processing of behaviorally-relevant stimuli and correlate with performance in goal-directed behaviors (Devilbiss et al., 2006; de Gee et al., 2017; Clewett et al., 2018; Totah et al., 2018; Vazey et al., 2018; Rodenkirch et al., 2019; Orlando et al., 2023; Grimm et al., 2024; Rodenkirch and Wang, 2024). Recent developments in genetically encoded fluorescent sensors in vivo have opened a window into the dynamics of norepinephrine released in brain regions targeted by LC projections during sensory and cognitive processing, learned behaviors, and sleep (Feng et al., 2019; Breton-Provencher et al., 2022; Kjaerby et al., 2022; Osorio-Forero et al., 2024; Liu et al., 2025). Pupil diameter is a reliable physiological biomarker of arousal and predictor of task performance in perceptual and cognitive tasks (Hess and Polt, 1960; Kahneman and Beatty, 1966; Nassar et al., 2012; de Gee et al., 2014; Reimer et al., 2014; Ebitz and Platt, 2015; McGinley et al., 2015; Vinck et al., 2015; Urai et al., 2017; Schriver et al., 2020; Narasimhan et al., 2023). There is also strong evidence of a link between neural activi ty in LC and dynamic changes in pupil size; however, the relationships between LC activity, pupil size, and arousal are strongly nonlinear and nonexclusive (Joshi et al., 2016; Liu et al., 2017). Although large pupil dilations often coincided with bursts of LC firing, reflecting arousal or salient stimuli, whether pupil size alone reliably indexes spiking activity in LC remains hotly debated (Megemont et al., 2022). Additionally, while arousal is a strong predictor of behavior, mice also exhibit distinct strategies during goal directed behaviors (Ashwood et al., 2022). However, the relationship between those behavioral strategies and arousal, indexed by pupil size or noradrenergic dynamics, remains unclear. Here we explored the relationships between pupil size fluctuations, noradrenergic dynamics, and behavior in a tactile stimulus detection task. Mice were trained to respond to behaviorally relevant .CC-BY-NC 4.0 International licenseavailable 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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint tactile signals in pursuit of water rewards. We simultaneously monitored pupil dynamics and fluorescent GRABNE signals in somatosensory and prefrontal cortices during the task. We found that baseline pupil size and baseline synchronization of GRABNE signals, but not baseline cortical GRABNE levels, were strong predictors of whether animals decided to respond. We also employed a generalized linear model - hidden Markov model (GLM-HMM) framework to identify behavioral states throughout the behavioral task. The behavioral states identified by the GLM-HMM were psychometrically distinct and temporally stable. Moreover, we found distinct task-evoked pupil dynamics and cortical NE dynamics across these behavioral states.

Surgery All experiments were approved by the Institutional Animal Care and Use Committee (IACUC) at Columbia University and were c onducted in compliance with NIH guidelines. Mice (B6, Jax #: 000664) underwent survival surgery for AAV injection and implantation of optical fibers and head plate at age of 3-6 months. Mice were initially anesthetized with isoflurane in oxygen (5% induction, 2% maintenance) and then secured in a stereotaxic frame. Body temperature was maintained at ~37 ℃ using a feedback-controlled heating pad (FHC, Bowdoinham, ME). After the mouse’s condition stabilized but before an incision was made on the scalp, lidocaine hydrochloride and buprenorphine (0.05 mg/kg) were administered subcutaneously to ensure analgesics were on board during the whole surgery. To measure cortical NE dynamics during tactile signal detection tasks, AAVs encoding GRAB NE (AAV9-hSyn-NE2h, WZ Biosciences) were injected into the PFC (AP: +2.3 mm, ML: 1.2 mm, DV: -2.0 mm) and S1 (AP: -1.30 mm, ML: +3.3 mm, DV: -0.72). Before the injection, small burr holes were drilled above the PFC and S1 and saline was applied to each craniotomy to prevent exposed brain surface from drying out. Pulled capillary glass micropipettes were back-filled with AAV solution, which was subsequently injected into the target brain regions (~200 nL each site) at 0.8 nL/s using a precision injection system (Nanoliter 2020, World Precision Instruments, Sarasota, FL). The micropipette was left in place for at least 10 minutes following each injection and then slowly withdrawn. Following GRAB NE AAV injection, an optical fiber (200 μm diameter, NA = 0.39) was implanted with the tip of the fiber placed approximately 0.15 mm above the injection site. Metabond (Parkell Inc., Edgewood, NJ) was used to build a headcap to bond the fibers. The ferrules and headplate were then cemented in place with dental acrylic. At the conclusion of the surgery, Baytril (5 mg/kg) and Ketoprofen (5 mg/kg) were administered. Four additional doses of Baytril and two additional doses of Ketoprofen .CC-BY-NC 4.0 International licenseavailable 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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint were provided every 24 hours after the surgery day. Animals’ weight was measured at least once per day for 5 days. Fiber photometry recording Fiber photometry recording was performed approximately 3 weeks following surgery to allow sufficient time for viral expression. Fluorescence signals mediated by the GRAB NE sensors were recorded using a 2-channel fiber photometry system (Doric Lenses). For each channel, the excitation light with 465 nm wavelength was generated by an LED (CLED_465, Doric Lenses) and passed through a MiniCube (iFMC4_AE (405)E(460–490)_F(500–550)_S, Doric Lenses). Emission fluorescence from the GRAB NE sensor was measured by an integrated PMT detector in the MiniCube. The fiber photometry recordings were run in a ‘Lock-in’ mode controlled by Doric Neuroscience Studio (V5.4.1.12), where the intensity of the excitation lights was modulated at frequencies of 208.62 Hz and 572.21 Hz to avoid contamination from other light sources in the room and crosstalk between the excitation lights (Liu et al., 2025). The demodulated signal processed by the Doric fiber photometry console was low-pass filtered at 25 Hz and sampled at 12 kHz with a 16-bit ADC. The fiber photometry system was synchronized with the behavioral apparatus through TTLs generated by the xPC target real-time system (MathWorks, Massachusetts). All photometry data were dec imated to 120 Hz by Doric Neuroscience Studio software and saved for offline analysis. We used signal background estimation with a 3 s moving window to remove drifts in the baseline signal while preserving peaks and oscillations (MATLAB function msbackadj). Pupillometry Pupil recordings were obtained using a custom pupillometry system (Schriver et al., 2018). The camera captured images at 10 frames per second via TTL triggers from the xPC real-time system. Pupil images were streamed to a high-speed solid-state drive for offline analysis. As previously described, the DeepLabCut toolbox was used for automated segmentation of the pupil contour (Weiss et al., 2025). Elliptical regression was then applied to fit the labeled points, enabling the computation of pupil size based on the fitted contour. To ensure segmentation accuracy, approximately 5% of segmented images were randomly selected and inspected. Pupil size during periods of blinks was estimated by interpolating over frames immediately preceding and following the blinks. If DeepLabCut did not recognize pupil contour due to either poor video quality or animal’s eyelid covering a significant portion of pupil in >33% of the recorded video frames, the session was excluded from pupillometry analysis. Prior to further analysis, a fourth- .CC-BY-NC 4.0 International licenseavailable 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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint order non-causal low-pass filter with a cutoff frequency of 3.5 Hz was applied to the pupil size data (Joshi et al., 2016) . Behavioral task Mice (n=4) were trained to respond to tactile stimuli applied to their whiskers via air puffs by licking a water spout. Each trial began with an onset tone followed by a delay period lasting 4-10 s. During the delay period, 1-3 distractor air puffs (not aimed at the mouse or its whiskers, but close enough to hear) were delivered. Responses to a distractor stimulus (licking the water spout within 500 ms of the distractor) resulted in restarting the trial. The mouse needed to withhold licks for 2 s at the end of the delay period in order for a stimulus to be delivered. On 40% of trials, no stimulus (0 PSI) was delivered (i.e. catch trials). On the remaining 60% of trials, air puffs were delivered to left whiskers at different pressures (0.5 – 20 PSI) with equal probability. Stimulus delivery was followed by a 500 ms window of opportunity, during which the mouse could respond to tactile stimulus by licking the water spout. Responses to tactile stimuli within the window of opportunity were rewarded with 2.3 µL of sweetened water (10% sucrose) followed by a 2 s cool down period before the start of the next trial. Responses to 0 PSI stimuli on catch trials resulted in a 15 s timeout before the start of the next trial. Sessions lasted 129 ± 5 trials (56.4 ± 0.1 mins). Histology At the end of the study, mice were transcardially perfused with PBS followed immediately by ice-cold 4% paraformaldehyde. The brain was removed carefully and post-fixed overnight at 4 °C in 4% paraformaldehyde, and then cryopreserv ed in a 30% sucrose (wt/vol) in PBS solution for 3 days at 4 ℃. Brains were embedded in Optimum Cutting Temperature Compound, and 30- μm coronal slices were sectioned using a cryostat. Brain slices were washed 4x in PBS and then incubated in 10% normal goat serum contained with 0.5% Triton X-100 in PBS for 2 hours. This was followed by primary antibody incubation overnight at room temperature using either a chicken anti-GFP primary antibody (1:500), or a mixture of chicken anti-TH (1:500) and rabbit anti-ChAT (1:300) primary antibodies. On the next day, slices were washed 3x in PBS + Tween (0.0005%) solution followed by secondary antibody incubation for 2 hours at room temperature using an Alexa Fluor 488-conjugated goat anti-chicken (1:800), or a mixture of Alexa Fluor 647-conjugated goat anti-rabbit (1:500) and Alexa Fluor 488-conjugated goat anti-chicken (1:800). The slices were then washed 3x in PBS + Tween solution and 1x with PBS only followed by coverslipping using Fluoromount-G medium with DAPI. Slices were imaged using 8X objective in a slide scanner (Nikon AZ100) for verification of AAV transfection. .CC-BY-NC 4.0 International licenseavailable 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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint Analyses All analyses were performed using custom code written in MATLAB (Mathworks, Natick, Ma) and Python available at https://github.com/Neural-Control-Engineering/Neurodynamic- control-toolbox. All pupillometry and fiber photometry recordings were z-scored prior to further analyses. Baseline measurements were averaged over 0.5 s prior to stimulus delivery. The magnitude of stimulus-induced pupil dilations and increases in NE signal were computed by subtracting the baseline value from the average pupil size during dilations or from GRAB NE signal over the 6 s following stimulus delivery. When computing cross correlations between GRAB NE in S1 and PFC, we used 4 s of baseline activity. We employed an approach combining generalized linear models and hidden Markov models (GLM-HMM) developed by Ashwood et al. (2022) to predict responses to tactile stimuli and identify behavioral states. For each mouse, we trained GLM-HMM models with 2-5 states using response on the previous trial (lick or no lick) and stimulus strength on the current trial, outcome on the previous trial (correct or incorrect) and stimulus strength on the current trial, or baseline pupil area and stimulus strength on the current trial as regressors. Trials were shuffled for training, and we used five-fold cross validation for training and testing throughout. We also compared performance on the GLM-HMM models to a classical lapse model (Prins, 2012; Ashwood et al., 2022) All analyses were performed on the 4-state model using baseline pupil area and stimulus strength on the current trial as regressors. Statistics All statistical tests were two-sided. A one-sample Kolmogorov-Smirnov test was first used to assess the normality of data before performing statistical tests. If the samples were normally distributed, a paired or unpaired t-test was used. Otherwise, the two-sided Mann-Whitney U-test was used for unpaired samples or the two-sided Wilcoxon signed-rank test for paired samples. Bonferroni correction was used for multiple comparisons.

Behavioral performance in tactile detection tasks To understand behaviorally relevant NE dynamics in the prefrontal cortex (PFC) and the Barrel cortex (S1), we trained 4 mice to perform tactile detection tasks (Stuttgen and Schwarz, 2008; Ollerenshaw et al., 2012). In the task, the animals were required to lick within the window of opportunity (500 ms) to indicate the perception of whisker stimulation induced by a brief air puff with different pressures (Figure 1A, B, see Methods). To gauge false alarm rate for the animals, approximately 40% of all trials were catch trials, in which no air puff was delivered. Following the .CC-BY-NC 4.0 International licenseavailable 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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint presentation of the tactile stimulus, the animals usually licked within approximately 400 ms to indicate the detection of the tactile stimulus. We found that the intensity of tactile stimuli had no effects on reaction time ( Figure 1C , p=0.58, one-way ANOVA test). As expected, response probability increased as the intensity of tactile stimulus increased ( Figure 1D,E), resulting in a monotonically increasing perceptual sensitivity with stimulus intensity (Figure 1F). Task-evoked pupil dynamics in whisker detection tasks We have previously showed that task-evoked phasic pupil dilation reflects different cognitive components in a perceptual decision making task in rats (Schriver et al., 2020). In this study, we also imaged pupil size of the mice throughout the behavioral task. Similar to rats (Schriver et al., 2018), the pupil of the mice fluctuated throughout the task (Figure 2A). Moreover, the animals’ pupils dilated following the presentation of a tactile stimulus, and the amplitude of task-evoked pupil dilation is positively correlated with the intensity of tactile stimulus ( Figure 2B, p<8.7e-24, one-way ANOVA test). Pupil dynamics around stimulus presentation varied across the four behavioral outcomes (i.e. hit, correct rejection, miss, and false alarm) (Figure 2C). Similar to rats’ pupil dynamics in a tactile discrimination task, pupil dilations are dependent upon the baseline pupil size: the larger the baseline pupil size, the smaller the pupil dilation, resulting in negative correlation coefficients between the baseline pupil size and pupil dilation (Figure 2D-F). This negative correlation was more profound in response trials than in withheld trials (Figure 2F, p<3.8e-4, Wilcoxon signed-rank test). We found that the baseline pupil size was smallest for false alarm trials (-0.43±0.12), followed by hit (-0.24±0.02), correct rejection (-0.14±0.04), and miss trials (0.16±0.06) ( Figure .CC-BY-NC 4.0 International licenseavailable 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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint 2G, p<8e-5, one-way ANOVA test), suggesting that pupil-linked arousal modulates tactile detection performance. Interestingly, in contrast to rats, the baseline pupil size of the mice was smaller on response trials than on withheld trials ( Figure 2G, -0.24±0.02 vs. 0.06±0.04, p<2.2e- 7, Wilcoxon signed-rank test, see Discussion). However, the task evoked pupil dilation across the four behavioral outcomes is similar to that in rats performing tactile discrimination tasks, with largest pupil dilation in hit trials (0.48±0.04), followed by false alarm (-0.35±0.07), miss (0.01±0.03), and correct rejection trials (-0.03±0.02) (Figure 2H, p<7.7e-8, one-way ANOVA test). Moreover, task evoked pupil dilation is larger in response trials than in withheld trials (Figure 2H, 0.48±0.04 vs. 0.01±0.02, p<8.7e-14, Wilcoxon signed-rank test). Consistent with previous work, the perceptual behavior of our animals was dependent upon baseline pupil size (Schriver et al., 2018). Although the reaction time did not depend on the baseline pupil size (Figure 2I, p=0.27, one-way ANOVA test), the animals had the highest .CC-BY-NC 4.0 International licenseavailable 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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint detection performance during the lowest tercile of baseline pupil size, and had the lowest detection performance during the highest tercile of baseline pupil size ( Figure 2J). On average, the higher the baseline pupil size, the lower response rate, but this trend was not significant (p=0.10, two-way ANOVA). There was a similar trend toward a dependence of perceptual sensitivity on baseline pupil size (Figure 2K, p=0.14, two-way ANOVA). Correlated NE dynamics between the primary somatosensory cortex and prefrontal cortex We next examined NE dynamics in the prefrontal cortex and the barrel cortex during the tactile detection task. Extracellular NE dynamics were measured through fiber photometry recording of fluorescent signals of GRAB NE sensors, which were expressed in the whisker primary somatosensory cortex (S1) and the prefrontal cortex (PFC) (Feng et al., 2019) ( Figure 3A). NE signals in the S1 and PFC prior to stimulus presentation were strongly synchronized as indicated by a peak in the shuffle corrected cross correlogram at a lag of 0 s between the two signals (Figure 3B ). We next examined if the correlation between NE dynamics in the S1 and PFC differed across the four behavioral outcomes ( Figure 3C). There was a significant difference in peak correlation coefficient across the behavioral outcomes ( Figure 3D, p=0.03, ANOVA test). There was a significant difference in peak correlation coefficient between response and withheld trials (p=0.004, Wilcoxon signed-rank test) but not between correct and incorrect trials ( Figure 3D, p=0.17, Wilcoxon signed-rank test). .CC-BY-NC 4.0 International licenseavailable 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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint NE dynamics in S1 and PFC during the detection task Task-evoked NE dynamics in S1 and PFC were dependent on the intensity of tactile stimulus (Figure 4A). The NE dynamics in the S1 were bimodal, with the first peak emerging at around 0.6 s following stimulus presentation. The second peak, which is comparable with the first peak in amplitude, occurred at around 3.5 s following stimulus presentation (Figure 4A). Although there were no significant differences in baseline NE levels in S1 across hit, correct rejection, false alarm, and miss trials (p=0.37, one-way ANOVA test), the task evoked increase in S1 NE levels were different across the four behavioral outcomes (p<4.3e-18, one-way ANOVA test) ( Figure 4B-D). In general, NE increases on action trials were higher than on withheld trials (p<2.5e-15, Wilcoxon signed-rank test) (Figure 4D). We next examined if baseline NE levels in S1 had effects on the detection behavior by assessing reaction time and response probability on trials in the different baseline NE terciles. We found that reaction time was not dependent on baseline NE levels (Figure 4E, p=0.39, one-way ANOVA test). However, baseline NE levels in S1 appeared to have a strong influence on the response probability, with highest response probability occurred on trials in the medium baseline NE tercile (p=0.022 for NE effect, p<10e-32 for stimulus effect, p=0.73 for interaction; two-way ANOVA test, Figure 4F ). This indicated an inverted-U shaped relationship between tonic NE level in S1 and detection performance (Wekselblatt and Niell, 2015). NE dynamics in PFC were slightly different than in S1: the decay of the first peak was faster and the amplitude of the second peak was about the half of the first peak ( Figure 4G), suggesting that the noradrenergic projection to the S1 and PFC may originate from different subpopulations of LC neurons (see Discussion). Similar to NE dynamics in S1, there were no significant differences in baseline NE levels in PFC across hit, correct rejection, false alarm, and miss trials (p=0.15, one-way ANOVA test) (Figure 4H&I). The task evoked increase in NE levels in PFC were different across the four behavioral outcomes (p<4.8e-16, one-way ANOVA test), with NE increase in action trials being higher than in withheld trials (p<3.7e-13, Wilcoxon signed- rank test) ( Figure 4J). Reaction time was not dependent on baseline NE levels in PFC either (Figure 4K, p=0.92, one-way ANOVA test). The inverted-U shaped relationship between tonic NE levels and detection performance was also evident in PFC: mice had the highest detection performance when the baseline NE level is medium (p=0.014 for NE effect, p<10e-32 for stimulus effect, p=0.86 for interaction; two-way ANOVA test, Figure 4L). .CC-BY-NC 4.0 International licenseavailable 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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint GLM-HMM modeling of behavioral state Recent work has revealed that mice switch between several strategies, identified as the state of HMM models, in perceptual decision-making tasks (Ashwood et al., 2022). Because behavioral state exerts strong influence on behavioral performance, we then examined if GLM- HMM models would be able uncover the behavioral states that the mice were in during the detection task (Figure 5A). We compared the prediction accuracy of the lapse model and GLM- HMM models with different inputs and numbers of states (see Methods). The elbow plot .CC-BY-NC 4.0 International licenseavailable 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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint illustrating the relationship between model perform ance and number of states suggested that a 4-state GLM-HMM model using baseline pupil size and stimulus strength as inputs had an optimal tradeoff between model accuracy and risk of over-fitting (Figure 5B ). During the task, the mice were mostly dwelling in state 3, which was associated with the highest detection performance, while spending the least numbers of trials in state 1 ( Figure 5C&D , p<0.031, ANOVA test). Although the mice appeared to randomly switch between different behavioral states, the probability of remaining in the same state (0.86±0.04) was substantially higher than probability of switching to other states (Figure 5E&F). We did not find any difference in reaction time across the 4 states. Mice had the highest decision criterion (i.e. most conservative) during the state 0 and the lowest decision criterion (i.e. most liberal) during the state 3, but the differences were marginally significant (p=0.07, one-way ANOVA test). Behavioral state dependent task-evoked pupil dilation and NE dynamics Since we incorporated baseline pupil-linked arousal levels into the GLM-HMM model to reveal latent behavioral states, we next examined task-evoked pupil dilation during the different behavioral states ( Figure 6A). The baseline pupil size was different across the four behavioral states for each behavioral outcome (Figure 6B, p < 2.8e-5 ANOVA tests). However, we failed to find a consistent trend for baseline pupil size across the four states except that the baseline pupil size was always largest in withheld trials during state 3. Task-evoked pupil dilation was also different across the four behavioral states for each behavioral outcome ( Figure 6C; p < 0.024, ANOVA tests). Pupil dilation appeared to be largest on response trials during state 0. .CC-BY-NC 4.0 International licenseavailable 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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint NE dynamics in S1 barely exhibited any differences in baseline NE levels across the four states for all behavioral outcomes ( Figure 7A,B). However, we found different task-evoked NE dynamics across the four states for hit, miss and correct rejection trials. For hit trials, task-evoked NE dynamics in S1 were similar across states ( Figure 7A ). For miss trials, task-evoked NE dynamics in state 1 were more similar to in state 3 than in the other 2 states ( Figure 7A,C). For CR trials, task-evoked NE dynamics in state 0 were distinct from those in the other 3 states (Figure 7A,C). Similarly, NE dynamics in PFC did not exhibit any differences in baseline NE levels across the four states for all behavioral outcomes (Figure 7D,E ). However, task-evoked PFC NE dynamics in states 1, 2 and 3 were similar to each other but differed from task-evoked NE dynamics in state 0 in hit trials, demonstrating the different feature of NE dynamics between S1 and PFC ( Figure 7D,F). For miss trials, task-evoked NE dynamics in state 1 were similar to in state 3 while task-evoked NE dynamics in state 0 were similar to in state 2 (Figure 7D,F). .CC-BY-NC 4.0 International licenseavailable 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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint Synchronization of NE dynamics between S1 and PFC differed significantly between states ( Figure 8 ), Specifically, during baseline activity, there was significantly higher shuffle- .CC-BY-NC 4.0 International licenseavailable 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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint corrected cross correlation between NE in S1 and PFC at a lag of 0 s during state 0 (putatively disengaged, low response probability) compared to the other states (p < 6.1e-4, ANOVA test).

In this study we trained mice to perform a whisker-based tactile signal detection task while simultaneously monitoring pupil dynamics and NE activity in S1 and PFC using GRAB NE fiber photometry. We found that both pupil size and synchronization of cortical NE signals were tightly linked to whether the animal reported the stimulus (hit and false alarm trials) versus withheld a response (miss and correct rejection trials), but that baseline pupil size and NE dynamics carried distinct information about behavioral state. Together, these results support a nuanced view in which pupil size and cortical NE are related but not interchangeable readouts of arousal and engagement (Joshi and Gold, 2020; Cazettes et al., 2021; Liu et al., 2021; Yang et al., 2021; Megemont et al., 2022; Grujic et al., 2024; Weiss et al., 2025). Consistent with previous work in rodents, we observed large, reliable pupil dilations following stimulus onset and during licking on hit trials, as well as similarly robust dilations on false alarm trials, with minimal changes on miss and correct rejection trials. The latency from stimulus onset to pupil dilation in our detection task was relatively short (on the order of ~100 ms), shorter than reported in whisker-based discrimination paradigms that require texture or direction judgments, where dilation emerges later (Lee and Margolis, 2016; Schriver et al., 2018). In addition, baseline pupil size also differed systematically across behavioral outcomes between this study and a previous study using rat models. Mice showed larger baseline pupils on trials in which they subsequently withheld a response (miss and correct rejection) and smaller baseline pupils on trials in which they responded (hit and false alarm). Although this finding supports the use of baseline pupil size as an index of arousal or engagement, it runs opposite to what we previously .CC-BY-NC 4.0 International licenseavailable 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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint observed in rats performing a more complex tactile discrimination task, where baseline pupil size tended to be higher on responded trials (Schriver et al., 2018). While this may reflect some genuine difference between species, these differences are likely related to task demands: detection requires only deciding whether a stimulus is present, whereas discrimination requires additional sensory evaluation and decision processes that may delay the arousal-related pupil response. Using GRAB NE, we observed rapid, stimulus-locked increases in NE in both cortical regions on hit and false alarm trials, and negligible changes on miss and correct rejection trials, mirroring the behavioral outcome dependence of the pupil response. However, the temporal profiles of NE signals are distinct between S1 and PFC. On hit trials, NE in PFC showed a sharp transient peak followed by a broader, lower-amplitude elevation, whereas NE in S1 exhibited two broader peaks of comparable magnitude. These distinct temporal patterns suggest that noradrenergic signaling from the LC is not broadcast uniformly to all cortical targets; rather, distinct subpopulations of noradrenergic neurons in the LC project with distinct task-related spiking dynamics project to different cortical areas (Totah et al., 2018; Breton-Provencher et al., 2021; Kelberman et al., 2024). In support of this notion, recent work has shown that distinct subpopulations in LC project to basal forebrain and PFC with only 30% overlap in neurons that project to both (Liu et al., 2025). The feedback from PFC to LC may shape this divergence in NE dynamics (Poe et al., 2020; Totah et al., 2021; Kelberman et al., 2024). PFC provides the densest cortical input to LC and has been proposed to use LC as a hub to broadcast control signals to other brain regions (Jodoj et al., 1998; Mashour et al., 2022). Stimulation of PFC can elicit LC spiking, whereas stimulation of more lateral frontal regions is less effective, suggesting a privileged PFC–LC pathway (Jodoj et al., 1998). In this framework, excitatory projections from PFC onto inhibitory LC interneurons could transiently suppress LC output back to PFC, producing a sharp peak followed by a rapid decline in NE in PFC (Breton-Provencher and Sur, 2019). At the same time, excitatory projections from PFC onto LC noradrenergic neurons that project to S1 could maintain a more sustained elevation of NE in S1, consistent with the longer-lasting S1 responses that we observed (Figure 4). Alternatively, noradrenergic projections from LC to PFC may exhibit increased expression of presynaptic inhibitory α2 receptors compared to those that project to S1. Although our data are correlational, this model provides a testable circuit-level explanation for the observed differences between PFC and S1. .CC-BY-NC 4.0 International licenseavailable 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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint Interestingly, in contrast to baseline pupil size, baseline NE levels in S1 and PFC did not predict whether the animal would respond on the trial. Instead, we found that the correlation of GRABNE signals between S1 and PFC during the baseline period was higher on withheld trials than on responded trials (Figure 3). If LC contains partially segregated subpopulations projecting to S1 and PFC, greater baseline synchrony between these projections may reflect a more homogeneous, low-information state of noradrener gic drive during behavioral disengagement (Wang et al., 2010). In contrast, lower cross-correlation before responded trials could indicate more differentiated, target-specific NE signaling when the animal is engaged and prepared to report stimuli. Thus, whereas baseline pupil area appears to index a global arousal or engagement, the baseline coordination of NE across cortical targets may capture a more circuit- specific facet of the animal’s internal state. Here we showed that HMM-GLM framework developed by Ashwood et al. (2022) could be effectively extended to a simple stimulus detection task with model outputs corresponding to Go vs. No-Go rather than a forced choice (left vs. right). Previous work has extended the HMM- GLM framework to reversal learning tasks and unforced two choice tasks (e.g. left, right, no response) (Le et al., 2023; Hulsey et al., 2024). Interestingly, the model achieved better predictive accuracy by using baseline pupil area as the model input rather than the outcome or action taken on the previous trial ( Figure 5). We also found distinct task-related pupil dynamics across the four states (Figure 6). These differences in pupil dynamics (not just baseline pupil area) between states suggest that the latent states identified in the HMM-GLM framework corresponded to genuine brain states. Differences in noradrenergic dynamics between states were modest, however (Figure 7 ). In state 0, the disengaged low-response-probability state, we observed significantly higher cross correlation during baseline between the GRAB NE signals recorded in S1 and PFC ( Figure 8), consistent with our findings showing higher GRAB NE signal correlations between S1 and PFC on withheld trials than on responded trials. There were, however, no differences in the correlation of GRAB NE signals between S1 and PFC across the other behavioral states. Future work is warranted to elucidate how norepinephrine differentially modulates task- relevant neural information processing across these latent states.

This work was supported by NIH R01NS119813, R01AG075114, R21MH125107, and NSF CBET 1847315. .CC-BY-NC 4.0 International licenseavailable 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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint Data availability All data and code are available upon request. Disclaimer Q.W. is a co-founder of Sharper Sense.

Ashwood ZC, Roy NA, Stone IR, Urai AE, Churchland AK, Pouget A, Pillow JW, The International Brain L (2022) Mice alternate between discrete strategies during perceptual decision- making. Nature Neuroscience 25:201–212. Breton-Provencher V, Sur M (2019) Active contro l of arousal by a locu s coeruleus GABAergic circuit. Nature Neuroscience 22:218–228. Breton-Provencher V, Drummond GT, Sur M (2021) Locus Coeruleus Norepinephrine in Learned Behavior: Anatomical Modularity and Spatiotemporal Integration in Targets. Front Neural Circuits 15:638007. Breton-Provencher V, Drummond GT, Feng J, Li Y, Sur M (2022) Spatiotemporal dynamics of noradrenaline during learned behaviour. Nature 606:732–738. Cazettes F, Reato D, Morais JP, Renart A, Mainen ZF (2021) Phasic Activation of Dorsal Raphe Serotonergic Neurons Increases Pupil Size. Curr Biol 31:192–197.e194. Cerpa JC, Piccin A, Dehove M, Lavigne M, Kremer EJ, Wolff M, Parkes SL, Coutureau E (2023) Inhibition of noradrenergic signalling in rodent orbitofrontal cortex impairs the updating of goal-directed actions. eLife 12:e81623. Clewett D, Huang R, Velasco R, Lee TH, Mather M (2018) Locus coeruleus activity strengthens prioritized memories under arousal. J Neurosci. Dahl MJ, Mather M, Werkle-Bergner M (2022) Noradrenergic modulation of rhythmic neural activity shapes selective attention. Trends Cogn Sci 26:38–52. de Gee JW, Knapen T, Donner TH (2014) Decision-related pupil dilation reflects upcoming choice and individual bias. Proceedings of the National Academy of Sciences of the United States of America 111:E618–E625. de Gee JW, Colizoli O, Kloosterman NA, Knapen T, Nieuwenhuis S, Donner TH (2017) Dynamic modulation of decision biases by brainstem arousal systems. eLife 6:e23232. Devilbiss DM, Page ME, Waterhouse BD (2006) Locus Ceruleus Regulates Sensory Encoding by Neurons and Networks in Waking Anim als. The Journal of Neuroscience 26:9860– 9872. Ebitz RB, Platt Michael L (2015) Neuronal activity in primate dorsal anterior cingulate cortex signals task conflict and predicts adjustments in pupil-linked arousal. Neuron 85:628–640. Feng J, Zhang C, Lischinsky JE, Jing M, Zhou J, Wang H, Zhang Y, Dong A, Wu Z, Wu H, Chen W, Zhang P, Zou J, Hires SA, Zhu JJ, Cui G, Lin D, Du J, Li Y (2019) A Genetically Encoded Fluorescent Sensor for Rapid and Specific In Vivo Detection of Norepinephrine. Neuron 102:745–761.e748. Ghosh S, Maunsell JHR (2024) Locus coeruleus norepinephrine contributes to visual-spatial attention by selectively enhancing perceptual sensitivity. Neuron 112:2231–2240.e2235. Grimm C, Duss SN, Privitera M, Munn BR, Karalis N, Frässle S, Wilhelm M, Patriarchi T, Razansky D, Wenderoth N, Shine JM, Bohacek J, Zerbi V (2024) Tonic and burst-like locus coeruleus stimulation distinctly shift net work activity across the cortical hierarchy. Nature Neuroscience 27:2167–2177. Grujic N, Polania R, Burdakov D (2024) Neurobehavioral meaning of pupil size. Neuron 112:3381–3395. .CC-BY-NC 4.0 International licenseavailable 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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint Hansen JY, Cauzzo S, Singh K, García-Gomar MG, Shine JM, Bianciardi M, Misic B (2024) Integrating brainstem and cortical functional architectures. Nature Neuroscience 27:2500– 2511. Hess EH, Polt JM (1960) Pupil Size as Related to Interest Value of Visual Stimuli. Science 132:349–350. Hulsey D, Zumwalt K, Mazzucato L, McCormick DA, Jaramillo S (2024) Decision-making dynamics are predicted by arousal and uninstructed movements. Cell reports 43:113709. Jensen KT, Doohan P, Sablé-Meyer M, Reinert S, Baram A, Akam T, Behrens TEJ (2025) A mechanistic theory of planning in prefrontal cortex. bioRxiv:2025.2009.2023.677709. Jodoj E, Chiang C, Aston-Jones G (1998) Potent excitatory influence of prefrontal cortex activity on noradrenergic locus coeruleus neurons. Neuroscience 83:63–79. Jordan R, Keller GB (2023) The locus coeruleus broadcasts prediction errors across the cortex to promote sensorimotor plasticity. eLife 12:RP85111. Joshi S, Gold JI (2020) Pupil Size as a Window on Neural Substrates of Cognition. Trends Cogn Sci 24:466–480. Joshi S, Li Y, Kalwani Rishi M, Gold Joshua I (2016) Relationships between pupil diameter and neuronal activity in the locus coeruleus, colliculi, and cingulate cortex. Neuron 89:221– 234. Kahneman D, Beatty J (1966) Pupil diameter and load on memory. Science 154:1583–1585. Kelberman MA et al. (2024) Diversity of ancestral brainstem noradrenergic neurons across species and multiple biological factors. bioRxiv. Kelley C, Slater C, Sorrentino M, Noone D, Hung J, Sajda P, Wang Q (2025) Alpha Modulation of Spiking Activity Across Multiple Brain Regions in Mice Performing a Tactile Selective Detection Task. European Journal of Neuroscience 62:e70356. Kjaerby C, Andersen M, Hauglund N, Untiet V, Dall C, Sigurdsson B, Ding F, Feng J, Li Y, Weikop P, Hirase H, Nedergaard M (2022) Memory-enhancing properties of sleep depend on the oscillatory amplitude of norepinephrine. Nature Neuroscience 25:1059–1070. Le NM, Yildirim M, Wang Y, Sugihara H, Jazayeri M, Sur M (2023) Mixtures of strategies underlie rodent behavior during reversal learning. PLoS Comput Biol 19:e1011430. Lee CR, Margolis DJ (2016) Pupil dynamics reflect behavioral choice and learning in a Go/NoGo tactile decision-making task in mice. Frontiers in Behavioral Neuroscience 10:200. Liu Y, Narasimhan S, Schriver BJ, Wang Q (2021) Perceptual Behavior Depends Differently on Pupil-Linked Arousal and Heartbeat Dynamics-Linked Arousal in Rats Performing Tactile Discrimination Tasks. Frontiers in systems neuroscience 14:614248–614248. Liu Y, Rodenkirch C, Moskowitz N, Schriver B, Wang Q (2017) Dynamic Lateralization of Pupil Dilation Evoked by Locus Coeruleus Activation Results from Sympathetic, Not Parasympathetic, Contributions. Cell reports 20:3099–3112. Liu YA, Nong Y, Feng J, Li G, Sajda P, Li Y, Wang Q (2025) Phase synchrony between prefrontal noradrenergic and cholinergic signals indexes inhibitory control. Nature Communications 16:7260. Mashour GA, Pal D, Brown EN (2022) Prefrontal cortex as a key node in arousal circuitry. Trends in Neurosciences 45:722–732. McGinley Matthew J, David Stephen V, McCormick David A (2015) Cortical membrane potential signature of optimal states for sensory signal detection. Neuron 87:179–192. Megemont M, McBurney-Lin J, Yang H (2022) Pupil diameter is not an accurate real-time readout of locus coeruleus activity. Elife 11. Narasimhan S, Schriver BJ, Wang Q (2023) Adaptive decision making depends on pupil-linked arousal in rats performing tactile discrim ination tasks. Journal of Neurophysiology 130:1541–1551. Nassar MR, Rumsey KM, Wilson RC, Parikh K, Heasly B, Gold JI (2012) Rational regulation of learning dynamics by pupil-linked arousal systems. Nature Neuroscience 15:1040–1046. .CC-BY-NC 4.0 International licenseavailable 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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint Ollerenshaw DR, Bari BA, Millard DC, Orr LE, Wang Q, Stanley GB (2012) Detection of tactile inputs in the rat vibrissa pathway. Journal of Neurophysiology 108:479–490. Orlando IF, Shine JM, Robbins TW, Rowe JB, O’Callaghan C (2023) Noradrenergic and cholinergic systems take centre stage in neuropsychiatric diseases of ageing. Neuroscience & Biobehavioral Reviews 149:105167. Osorio-Forero A, Foustoukos G, Cardis R, Cherrad N, Devenoges C, Fernandez LMJ, Lüthi A (2024) Infraslow noradrenergic locus coeruleus activity fluctuations are gatekeepers of the NREM–REM sleep cycle. Nature Neuroscience. Pezzulo G, van der Meer MAA, Lansink CS, Pennartz CMA (2014) Internally generated sequences in learning and executing goal-directed behavior. Trends in Cognitive Sciences 18:647–657. Poe GR, Foote S, Eschenko O, Johansen JP, Bouret S, Aston-Jones G, Harley CW, Manahan- Vaughan D, Weinshenker D, Valentino R, Berridge C, Chandler DJ, Waterhouse B, Sara SJ (2020) Locus coeruleus: a new look at the blue spot. Nat Rev Neurosci 21:644–659. Prins N (2012) The psychometric function: the lapse rate revisited. J Vis 12. Reimer J, Froudarakis E, Cadwell Cathryn R, Yatsenko D, Denfield George H, Tolias Andreas S (2014) Pupil fluctuations track fast switching of cortical states during quiet wakefulness. Neuron 84:355–362. Rodenkirch C, Wang Q (2024) Optimization of Temporal Coding of Tactile Information in Rat Thalamus by Locus Coeruleus Activation. Biology (Basel) 13. Rodenkirch C, Liu Y, Schriver BJ, Wang Q (2019) Locus coeruleus activation enhances thalamic feature selectivity via norepinephrine regulation of intrathalamic circuit dynamics. Nature Neuroscience 22:120–133. Schriver B, Bagdasarov S, Wang Q (2018) Pupil-linked arousal modulates behavior in rats performing a whisker deflection direction discrimination task. Journal of Neurophysiology 120:1655–1670. Schriver BJ, Perkins SM, Sajda P, Wang Q (2020) Interplay between components of pupil-linked phasic arousal and its role in driving behavioral choice in Go/No-Go perceptual decision- making. Psychophysiology:e13565. Speed A, Haider B (2021) Probing mechanisms of visual spatial attention in mice. Trends Neurosci 44:822–836. Stuttgen MC, Schwarz C (2008) Psychophysica l and neurometric detection performance under stimulus uncertainty. Nature Neuroscience 11:1091–1099. Totah NK, Logothetis NK, Eschenko O (2021) Synchronous spiking associated with prefrontal high γ oscillations evokes a 5-Hz rhythmic modulation of spiking in locus coeruleus. Journal of Neurophysiology 125:1191–1201. Totah NK, Neves RM, Panzeri S, Logothetis NK, Eschenko O (2018) The Locus Coeruleus Is a Complex and Differentiated Neuromodulatory System. Neuron 99:1055–1068.e1056. Urai AE, Braun A, Donner TH (2017) Pupil-linked arousal is driven by decision uncertainty and alters serial choice bias. Nature Communications 8:14637. Vazey EM, Moorman DE, Aston-Jones G (2018) Phasic locus coeruleus activity regulates cortical encoding of salience information. Proceedings of the National Academy of Sciences 115:E9439. Vinck M, Batista-Brito R, Knoblich U, Cardin Jessica A (2015) Arousal and Locomotion Make Distinct Contributions to Cortical Activity Patterns and Visual Encoding. Neuron 86:740– 754. Wang Q, Webber R, Stanley GB (2010) Thalamic Synchrony and the Adaptive Gating of Information Flow to Cortex. Nature Neuroscience 13:1534–1541. Weiss E, Liu Y, Wang Q (2025) The contribution of the locus coeruleus – norepinephrine system to the coupling between pupil-linked arousal and cortical state. The Journal of Neuroscience:e0898252025. .CC-BY-NC 4.0 International licenseavailable 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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint Wekselblatt Joseph B, Niell Cristopher M (2015) Behavioral state—getting “in the zone”. Neuron 87:7–9. Wimmer RD, Schmitt LI, Davidson TJ, Nakajima M, Deisseroth K, Halassa MM (2015) Thalamic control of sensory selection in divided attention. Nature 526:705–709. Yang H, Bari BA, Cohen JY, O'Connor DH (2021) Locus coeruleus spiking differently correlates with S1 cortex activity and pupil diameter in a tactile detection task. eLife 10:e64327. .CC-BY-NC 4.0 International licenseavailable 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 made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.699887doi: bioRxiv preprint