Locus coeruleus spiking differently correlates with somatosensory cortex activity and pupil diameter

We examined the relationships between activity in the locus coeruleus (LC), activity in the primary somatosensory cortex (S1), and pupil diameter in mice performing a tactile detection task. While LC spiking consistently preceded S1 membrane potential depolarization and pupil dilation, the correlation between S1 and pupil was more heterogeneous. Furthermore, the relationships between LC, S1 and pupil varied on timescales of sub-seconds to seconds within trials. Our data suggest that pupil diameter can be dissociated from LC spiking and cannot be used as a stationary index of LC activity.

LC activity (measured in a 0.5-s window prior to stimulus onset) was slightly but significantly lower 51 than Miss trials, where mice failed to lick to the whisker stimulus (Fig. 2e). We note that on Miss 52 trials LC responded weakly to whisker stimulus alone (< 0.5 sp/s above baseline, Fig. S1). LC 53 activity measured in a short window (0.2-s) after stimulus onset was larger on Hits compared with 54 Misses (Fig. 2e; the same trend holds for 0.1-s window, data not shown). Ideal-observer analysis 55 showed that both pre-and post-stimulus LC activity significantly predicted perceptual reports of 56 the mice on a trial-by-trial basis, with choice probabilities 20 of 0.47 ± 0.014 (P = 0.032, n = 43) for 57 pre-stimulus and 0.59 ± 0.017 (P = 4.6e-6, n = 43) for post-stimulus LC activity, respectively (Fig.  58 2e). LC activity aligned to the time of licking showed that spiking responses began ~200 ms prior 59 to licking (Fig. 2f). 60 In striking contrast, pupil diameter minimally increased in response to the tone. Instead, 61 pupil strongly dilated on Hit and False Alarm trials, in which mice made Go (licking) responses 62 (Fig. 2a, c, d; tone vs. Go: P = 6.4e-5, n = 36, Methods) 15 . Interestingly, pupil response to the tone 63 was larger on Misses compared to Hits, and significantly predicted perceptual choices of the mice 64 (Fig. S2). Pupil diameter changes (∆Pupil) aligned to the time of licking showed that pupil 65 responses occurred after licking (Fig. 2f). 66 Together, these data show that LC and pupil responses were positively correlated. Both 67 LC activity and pupil diameter increased during licking responses, but LC also strongly responded 68 to the tone, a salient sensory cue that alerted mice of trial onsets. Thus, LC activity and pupil 69 diameter appear to reflect different sets of task events during this behavior. 70 Next, we analyzed recordings where we simultaneously measured membrane potential 71 (Vm) of S1 neurons (mostly from layer 2/3, Fig. S3) along with LC spiking and/or pupil diameter 72 during the detection task. Our goal was to determine how LC spiking related to cortical activity 73 and to pupil diameter during task performance. We used spike-triggered averages (STAs) to 74 quantify how individual spikes from single LC units correlated with changes in Vm and pupil 75 diameter. LC spike-triggered Vm analyses revealed that LC spikes were associated with a 76 depolarization in cortical neurons (1.39 ± 0.35 mV, n = 12, Fig. 3a-c). On average, Vm 77 depolarization associated with an LC spike peaked after the spike, with short time lags from an 78 LC spike to peak depolarization in S1 (0.17 ± 0.06 s, n = 12, Fig. 3a-c, also see Fig. S4).

79
Consistent with the previous cross-correlogram analysis based on a larger set of LC-pupil 80 recordings (Fig. 2b), here STA analysis showed that pupil diameter increased in association with 81 individual spikes from LC single units (0.03 ± 0.01 mm, n = 7), with peak dilation occurring roughly 82 ten-fold slower than peak Vm depolarization (time lags from an LC spike to peak pupil dilation: 83 1.89 ± 0.25 s, n = 7, Fig. 3d-f). 84 Given that pupil diameter and LC activity are positively correlated, and that pupil diameter 85 has been often considered to index LC activity 16,21 , we next tested whether the pupil-S1 86 relationship resembled the LC-S1 relationship. Cross-correlogram analyses revealed 87 heterogeneous correlations between pupil diameter and S1 Vm, with both positive and negative 88 correlations as well as positive and negative time lags (peak correlation coefficient: 0.05 ± 0.04; 89 time lags: -0.22 ± 1.01 s, n = 19, Fig. 3g Pupil'-Vm, P = 7.1e-7). On the other hand, the LC-Pupil' relationship was very similar to that of 99 LC-Pupil (compare Fig. S5a,b with Fig. 3e,f). 100 Together, these data show that LC spikes preceded S1 depolarizations and pupil dilations.

101
LC spiking correlated with both Vm and pupil diameter changes, but on vastly different timescales 102 (~0.2 s vs. ~2 s). Our data also show that the time derivative of pupil diameter, but not the absolute 103 pupil size, is a good predictor of S1 Vm fluctuations. However, LC spiking can track fast Vm 104 fluctuations better than either pupil and pupil'. 105 Individual trials in our detection task contained distinct events, including the tone that 106 alerted mice of the trial start ("Tone"), the whisker stimulus on Go trials ("Stimulus"), and licks 107 ("Lick"), as well as other periods in which mice did not receive stimuli or make lick responses 108 ("Quiet"). For a more granular perspective on how LC spiking correlated with changes in Vm and 109 pupil diameter, we computed LC spike-triggered averages separately in these different event 110 windows (task epochs, Methods). 111 While single LC spikes were associated with prominent changes in both cortical Vm and 112 pupil diameter, we found that these associations strikingly depended on task epochs: Vm 113 depolarization associated with an LC spike had the biggest response to tone/licking and almost 114 no response during the quiet periods (Fig. 4a). In contrast, pupil dilation associated with an LC 115 spike had the biggest response to licking and almost no response to the tone (Fig. 4b). The pupil' 116 associated with an LC spike had an intermediate response to the tone (Fig. 4c). In addition, peak 117 pupil dilation, pupil' and Vm depolarization appeared to have different dependencies on LC spike 118 counts, with a roughly monotonic relationship between pupil and LC, and a much weaker 119 dependence between Vm and LC (Fig. S6). Thus, the correlations between LC spiking and Vm, 120 and between LC spiking and pupil diameter, are non-stationary, even on the timescale of a few 121 seconds. Importantly, these epoch-dependencies were different for Vm and pupil -with the biggest 122 response occurring to the tone for Vm, and the smallest response occurring to the tone for pupil -123 suggesting that the correlations between LC activity and Vm and pupil each reflect distinct 124 unmeasured underlying processes. 125 126 DISCUSSION 127 128 We found that pre-stimulus baseline LC spiking predicted behavioral responses. Thus, 129 fluctuations in LC/NE activity may in part underlie perceptual task performance. However, the 130 effect was weak, possibly due to the use of an auditory cue that puts the mice in a more 131 homogeneous arousal state. As a result, factors other than fluctuations of arousal also likely 132 contribute to cortical choice probabilities observed in prior work with this task 20 . In other tasks 133 without such alerting cues, task performance may have a stronger dependence on arousal and 134 pre-stimulus LC activity.

135
LC responded strongly to an auditory cue (tone) meant to alert the mice to the beginning 136 of a trial. While this tone carried no information about the presence of a tactile stimulus or reward 137 on any given trial, and therefore was not associated with a particular movement response, it did 138 inform the mice about the time when a tactile stimulus could occur (in our task the duration 139 between the tone and stimulus onset was fixed). The robust LC spiking responses to this cue are 140 therefore consistent with LC's role in promoting alertness or preparedness to detect a weak 141 stimulus. We also found that LC responded to operant licking responses, which is consistent with 142 earlier work showing that LC encoded overt decision execution 22 . 143 Our data show that while LC spiking and pupil diameter correlate well at long timescales, 144 and both can predict changes in cortical dynamics, LC does so an order of magnitude faster. 145 Moreover, the correlation between pupil and Vm is much more heterogeneous than between LC 146 and Vm. In support of previous studies, our results suggest that compared with change in the 147 absolute size of pupil diameter, its time derivative is a better predictor of cortical states 10,12 . 148 Importantly, the relationships between LC activity, S1 Vm and pupil depended on task epoch.

149
Because these epochs changed on the timescale of a few seconds, our data imply that pupil 150 diameter can be dissociated from LC spiking and cannot be used as a stationary index of LC 151 activity. However, comparing across repeats of similar epochs should yield a more accurate 152 prediction of LC spiking by pupil diameter. That is, in attempting to use pupil diameter as a proxy 153 for LC spiking, our data suggest it would be useful to separately normalize distinct task epochs. All procedures were performed in accordance with protocols approved by the Johns Hopkins 244 University Animal Care and Use Committee. Mice were DBH-Cre (B6.FVB(Cg)-Tg(Dbh-cre) 245 KH212Gsat/Mmucd, 036778-UCD, MMRRC); Ai32 (RCL-ChR2(H134R)/EYFP, 012569, JAX), 246 singly housed in a vivarium with reverse light-dark cycle (12 hr each phase). Male and female 247 mice of 6-12 weeks were implanted with titanium head posts as described previously 20 . After 248 recovery, mice were trained to perform a Go/NoGo single whisker detection task as described 249 previously 20 .

251
Custom microdrives with eight tetrodes and an optic fiber 23 (0.39 NA, 200 um core) were built to 252 make extracellular recordings from LC neurons. Each tetrode comprised four nichrome wires 253 (100-300 KΩ). A ~1 mm diameter craniotomy was made (centered at -5.2 mm caudal and 0.85 254 mm lateral relative to bregma) for implanting the tetrodes to a depth of 2.7 mm relative to the 255 brain surface. The microdrive was advanced in steps of ~100 um each day until reaching LC, 256 identified by optogenetic tagging of DBH+ neurons expressing ChR2, tail pinch response, wide 257 extracellular spike waveforms and post-hoc electrolytic lesions. Broadband voltage traces were 258 acquired at 30 kHz (Intan Technologies) and filtered between 0.1 and 10 kHz. Signals were then 259 bandpass filtered between 300 and 6000 Hz, and spikes were detected using a threshold of 4-6 260 standard deviations. The timestamp of the peak of each detected spike, as well as a 1-ms 261 waveform centered at the peak were extracted from each channel for offline spike sorting using 262 MClust 24 . At the conclusion of the experiments, brains were perfused with PBS followed by 4% 263 PFA, post-fixed overnight, then cut into 100 μm coronal sections and stained with anti-Tyrosine 264 Hydroxylase (TH) antibody (Millipore AB152).

266
Pupil video was acquired at 50 Hz using a PhotonFocus camera and StreamPix 5 software. 267 Light from a 940 LED was passed through a condenser lens and directed to the right eye, 268 reflected off a mirror, and directed into a 0.25X telecentric lens. WaveSurfer 269 (https://www.janelia.org/open-science/wavesurfer) triggered individual camera frames 270 synchronized with electrophysiological recordings.

272
In a subset of animals, we performed simultaneous intracellular current clamp (whole-cell) 273 recordings in conjunction with LC recording and/or pupil tracking during behavior. A craniotomy 274 over the C2 barrel was made based on intrinsic signal imaging 20 . In some cases, we also made 275 craniotomies over nearby barrels based on the known somatotopy of S1 25,26 to increase yield. 276 Whole-cell recording procedures, quality control and data processing were performed as 277 described previously 20 . 278 279 For Fig. 2d, LC responses to the tone were calculated using a 300-ms window starting at tone 280 onset, and LC responses to Go were calculated using a 300-ms window starting 200 ms after 281 stimulus onset to capture peak responses. These estimates were based on LC response 282 profiles in Fig. 2c. Pupil responses to the tone were calculated using a 1-s window starting 1 s 283 after tone onset. This estimate was primarily based on pupil response profile during CR trials 284 (e.g., Fig. 2a, c, indicated by the grey bar), where there was no whisker stimulus or licking 285 response. Pupil responses to Go (licking) were calculated using a 1-s window starting 1.5 s after 286 stimulus onset (e.g., FA trials in Fig. 2a, c, indicated by the black bar). Based on the temporal 287 profiles of pupil diameter in different trial types shown in Fig. 2a, c, and that the whisker stimulus 288 started 1 s after tone onset, pupil responses to tone and Go can be segregated. These 289 estimates were consistent with the results showing that pupil dilated 1-2 s after LC spikes (Fig.  290 2b, and Fig. 3d-f). 291 292 For Fig. 2e, pre-stimulus LC baseline activity was calculated using a 500-ms window ending 50 293 ms before stimulus onset. Post-stimulus activity was calculated using a 200-ms window starting 294 20 ms after stimulus onset, before licking responses 20 . Choice probabilities were computed as 295 described previously 20 .

297
To compute lick-aligned changes in LC spiking and pupil diameter, we only used licks that 298 occurred at least 0.5 s after the previous lick. To compute LC spike triggered S1 Vm and pupil, 299 we only used LC spikes that occurred at least 0.5 s after the previous spike. For STA analysis, 300 peak ∆Vm, ∆Pupil or the time derivative of Pupil (Pupil') was defined as the largest positive or 301 negative value within the observed window (± 1 s or ± 10 s, respectively). 302

303
For cross-correlogram analysis, each LC spike train was convolved with a 400-ms wide 304 Gaussian kernel (results hold for 200-ms kernel, data not shown). Peak correlation coefficients 305 were defined as the largest positive or negative value within the observed window (± 1 s or ± 10 306 s). To examine how well LC spiking and pupil diameter could predict cortical Vm fluctuations at 307 different timescales (Fig. 3l), Vm was high-pass filtered at 0, 0.2, 0.4, 0.6, 0.8, 1, 2, 3, 4 and 5 Hz 308 separately. Cross-correlogram analysis between the filtered Vm and LC (pupil) activity was then 309 performed as described above, and largest absolute values of peak correlation coefficients were 310 taken.

312
Task epochs were defined as: "Tone" epochs: -0.2 s to 0.3 s with respect to tone onset; 313 "Stimulus" epochs: -0.2 s to 0.3 s with respect to stimulus onset (i.e., only on trials with whisker 314 stimulation); "Licking" epochs: -0.2 s to 0.3 s with respect to licks that occurred at least 0.5 s 315 after the previous lick; "Quiet" epochs: non-overlapping 0.5 s segments excluding the three 316 types of epoch defined previously during the entire session.

318
Thirty-nine LC-pupil pairs were included in Fig. 2b, including single-and multi-units, with and 319 without S1 recordings. For the rest of Fig. 2, LC analysis included forty-three recordings, each 320 with at least 4 Hit and 4 Miss trials. Among those, thirty-six were with pupil recordings, and were 321 used for pupil analysis. Twelve pairs of S1 whole-cell and LC single-unit recordings were 322 included in Fig. 3a-c, 4a, seven of which were with pupil recordings and included in Fig. 3d-f. 323 Nineteen S1-pupil recordings were included in Fig. 3g-l. Twenty pairs of LC SU and pupil 324 recordings were included in Fig. 4b,c, with and without S1 recordings.

326
Data were reported as mean ± s.e.m. unless otherwise noted. Statistical tests were by two-tailed 327 Wilcoxon signed rank unless otherwise noted. We did not use statistical methods to predetermine 328 sample sizes. Sample sizes are similar to those reported in the field. We assigned mice to 329 experimental groups arbitrarily, without randomization or blinding. (e) Example simultaneously recorded LC activity, S1 Vm, and pupil.

FIGURE 2: LC and pupil responses during behavior.
(a) Example LC recording with pupil tracking. Left: LC spike raster separated by trial types. Right: Mean pupil diameter (± s.e.m.) separated by trial types. Grey and black arrows indicate tone and stimulus onsets, respectively. Grey and black bars indicate the time windows during which pupil responses to tone and to Go (behavioral responses) were quantified, respectively. We note that based on the temporal profiles of pupil diameter in different trial types (i.e., in the presence or absence of tactile stimulus or licking), and that tactile stimulus starts 1 s after tone onset, pupil responses to tone and Go can be segregated (Methods). (b) Top: Cross-correlogram between LC spike train and pupil diameter. Individual LC spikes were convolved with a 400-ms wide Gaussian kernel. Spike times were shuffled and LC-pupil correlations computed to establish controls (narrow grey band around zero). Bottom: Histogram of peak correlation coefficient (left), and time lags (right) between LC spike train and pupil diameter for each paired recording (magenta dot: mean). Both distributions are significantly larger than 0 (peak correlation coefficient: 0.15 ± 0.02, P = 8.3e-7, Signed rank = 743; time lags: 2.61 ± 0.39 s, P = 7.8e-7, Signed rank = 744, n = 39).   .1, P = 1.3e-9, n = 20. Post-hoc Tukey-Kramer tests revealed that peak ∆Pupil in lick and stimulus epochs were larger than in tone and quiet epochs. Lick vs. Stim, P = 0.10; Tone vs. Quiet, P = 0.76; Lick vs. Tone, P = 3.7e-7; Lick vs. Quiet, P = 6.2e-4; Stim vs. Tone, P = 1.1e-4; Stim vs. Quiet, P = 0.0027. (c) Top: LC spike-triggered pupil' separated by task epoch. Bottom: Bar graphs of peak pupil' for each epoch. Dots indicate individual paired recordings. Repeated-measures ANOVA, F(3, 57) = 35.3, P = 4.9e-13, n = 20. Post-hoc Tukey-Kramer tests revealed that peak pupil' in lick and stimulus epochs were larger than in tone, and peak pupil' in quiet epochs was the lowest. Lick vs. Stim, P = 0.46; Tone vs. Quiet, P = 0.0013; Lick vs. Tone, P = 1.0e-4; Lick vs. Quiet, P = 1.4e-8; Stim vs. Tone, P = 0.0058; Stim vs. Quiet, P = 6.7e-6.