Juxtacellular recordings from identified neurons in the mouse locus coeruleus

The locus coeruleus (LC) is the primary source of noradrenergic transmission in the mammalian central nervous system. This small pontine nucleus consists of a densely packed nuclear core—which contains the highest density of noradrenergic neurons—embedded within a heterogeneous surround of non‐noradrenergic cells. This local heterogeneity, together with the small size of the LC, has made it particularly difficult to infer noradrenergic cell identity based on extracellular sampling of in vivo spiking activity. Moreover, the relatively high cell density, background activity and synchronicity of LC neurons have made spike identification and unit isolation notoriously challenging. In this study, we aimed at bridging these gaps by performing juxtacellular recordings from single identified neurons within the mouse LC complex. We found that noradrenergic neurons (identified by tyrosine hydroxylase, TH, expression; TH‐positive) and intermingled putatively non‐noradrenergic (TH‐negative) cells displayed similar morphologies and responded to foot shock stimuli with excitatory responses; however, on average, TH‐positive neurons exhibited more prominent foot shock responses and post‐activation firing suppression. The two cell classes also displayed different spontaneous firing rates, spike waveforms and temporal spiking properties. A logistic regression classifier trained on spontaneous electrophysiological features could separate the two cell classes with 76% accuracy. Altogether, our results reveal in vivo electrophysiological correlates of TH‐positive neurons, which can be useful for refining current approaches for the classification of LC unit activity.


| INTRODUCTION
The locus coeruleus (LC) is the principal source of noradrenergic neurotransmission in the mammalian central nervous system (Swanson & Hartman, 1975).Noradrenergic LC neurons, through their widely divergent connections to virtually the entire central nervous system, are in a key position for regulating numerous brain functions, including attention, arousal and memory (for review, see Berridge & Waterhouse, 2003;Poe et al., 2020;Sara, 2009Sara, , 2015;;Sara & Bouret, 2012;Totah et al., 2019).LC neurons have been classically assumed to be a structurally and functionally homogenous population; however, over the years, diverse research lines have revealed a high degree of heterogeneity and modularity within local LC microcircuits (Chandler et al., 2019).Anatomically, the LC is a complex of noradrenergic neurons located within the rostral hindbrain.This complex has been classically partitioned into a 'nuclear core' region, which contains the highest density of noradrenergic neurons (especially in murid rodents, see Manger & Eschenko, 2021), and additional neighbouring subregions, where the organization of noradrenergic cells is more scattered and intermingled with the non-noradrenergic surround (Aston-Jones et al., 2004;Breton-Provencher & Sur, 2019;Gasparini et al., 2021;Gong et al., 2020;Grzanna & Molliver, 1980;Jin et al., 2016;Kuo et al., 2020;Luskin et al., 2022;McKinney et al., 2023;Shipley et al., 1996;Swanson, 1976; for review, see Manger & Eschenko, 2021).Early anatomical studies divided LC neurons into several types based on their distinct morphological characteristics (Cintra et al., 1982;Loughlin et al., 1986).More recent work indicates that LC neurons are also heterogeneous in terms of their molecular expression profiles, intrinsic electrophysiological properties and projection targets (Chandler et al., 2014(Chandler et al., , 2019;;Kalwani et al., 2014;Li et al., 2016;McKinney et al., 2023;Plummer et al., 2017;Schwarz et al., 2015;Su & Cohen, 2022;Tortorelli et al., 2022;Totah et al., 2018;Uematsu et al., 2017).However, whether and how this structural heterogeneity of LC neurons relates to in vivo activity has remained largely unresolved.
Extracellular recordings-the gold standard method in systems neuroscience, for review, see Buzs aki, 2004)have provided a massive amount of information about the in vivo activity of individual neurons in many areas of the mammalian central nervous system.These methods are particularly powerful when recordings can be confidently targeted to the brain region of interest, and cell identity can be inferred from spike signals alone.In the neocortex and hippocampus, for example, reliable electrophysiological signatures of anatomical boundaries have been identified, which allow targeting even specific subfields or layers within these structures (GoodSmith et al., 2017;Jung et al., 2019;Mizuseki et al., 2011;Sirota & Buzs aki, 2005).Moreover, 'ground-truth' datasets from anatomically identified neurons in these structures have made it possible to develop electrophysiological criteria that enable to infer the putative cell identity of the extracellularly recorded units with a high degree of confidence (Czurk o et al., 2011;Diamantaki et al., 2016;GoodSmith et al., 2017;Henze et al., 2000;Jung et al., 2019;Klausberger et al., 2003;Neto et al., 2016;Penttonen et al., 1997;Senzai & Buzs aki, 2017).
The relatively small size of the LC, together with the lack of clear electrophysiological signatures of anatomical boundaries, has made it particularly difficult to infer noradrenergic cell identity from blind extracellular recordings.Even if recordings are targeted to the LC 'nuclear core', which is known to be almost exclusively composed of noradrenergic cells, extracellular spike signals are de facto sampled from a larger area (Gold et al., 2006;Henze et al., 2000;Logothetis, 2003) that extends well beyond the narrow LC 'nuclear core'.Hence, to increase confidence in cell classification, it is necessary to first characterize the in vivo properties of unequivocally identified noradrenergic neurons and then test whether these features can be used for differentiating them from the local heterogeneous surround of non-noradrenergic cells.Elegant pharmacology-based approaches have been classically employed for inferring the putative noradrenergic identity of the recorded units (Aston-Jones & Bloom, 1981a;Manohar et al., 2017;Sara & Segal, 1991;Shea et al., 2008;Totah et al., 2018).More recent approaches involving high-density extracellular recordings, in combination with optogenetic tagging, have increased the efficiency of LC sampling and enabled the in vivo characterization of opto-identified LC units (Breton-Provencher et al., 2022;Carter et al., 2010;Hickey et al., 2014;Hirschberg et al., 2017;Ito et al., 2020;Martins & Froemke, 2015;McBurney-Lin et al., 2022;Megemont et al., 2022;Swift et al., 2018;Tortorelli et al., 2022;Uematsu et al., 2017;Xiang et al., 2019;Yamasaki & Takeuchi, 2017).When cell-type specific properties are very distinct and under low levels of recurrent connectivity, optotagging methods can achieve excellent separation of the extracellularly recorded units (as, e.g., for LC units from the medially located Barrigton's nucleus, Ito et al., 2020); however, if differences are more subtle and neural circuits more intermingled, extracellular optotagging becomes far from trivial and more critically dependent upon the performance of spike-sorting algorithms.The latter is known to be limited under conditions of high 'background' spiking activity, high cell densities and synchronous firing (as is the case in the LC) as well as during photostimulation (Franke et al., 2015;Roux et al., 2014;Royer et al., 2010;Stark et al., 2012).Distinguishing between directly and indirectly lightactivated neurons is often non-trivial, especially within circuits with recurrent connections (Beyeler et al., 2016;Roux et al., 2014;Zutshi et al., 2018), and direct validation is challenging because extracellular recording methods do not provide access to anatomic information of the recorded neurons.These limitations highlight the need for 'ground truth' data obtained from individual, anatomically identified LC neurons.
To bridge this gap, we employed the juxtacellular technique for recording and labelling single neurons in vivo in the mouse LC complex.Specifically, we aimed to characterize the in vivo activity of identified noradrenergic neurons and asked whether electrophysiological properties alone can be sufficient for distinguishing them from the heterogeneous surround of non-noradrenergic cells.Based on our dataset of identified neurons (n = 50), we describe in vivo electrophysiological correlates of noradrenergic neurons within the mouse LC complex.

| Surgical procedures
Surgical procedures were essentially performed as previously described (Balsamo et al., 2022;Diamantaki et al., 2018;Ding et al., 2022).Animals were anaesthetized with a mixture of ketamine and xylazine (i.p., 80-100 mg/kg and 10 mg/kg, respectively), and the anaesthesia was maintained with additional doses of ketamine (half of the initial dose).The animal's body temperature was kept at 37 C using a feedbackcontrolled heating pad (DC temperature controller, FHC).Ophthalmic ointment was applied to the eyes.Prior to the scalp incision, local anaesthesia was applied (s.c., Xylocaine, AstraZeneca, GmbH, Wedel, Germany).
Craniotomies were performed above the LC (for the specific stereotaxic coordinates, see Section 2.3).
Juxtacellular labelling was initiated with a brief, 'buzzlike' current pulse (Balsamo et al., 2022;Diamantaki et al., 2016) and carried out with positive square current pulses (200 ms, 50% duty cycle as in Pinault, 1996).During 'entrainment', the current intensity was manually adjusted, typically in the 1-20 nA range, to modulate spiking activity without damaging the cell.Following a post-labelling survival time of 1-3 h under deep general anaesthesia, histological analysis was performed (see Section 2.4).

| Histological analysis, morphological reconstruction and cell quantification
Animals were euthanized with an overdose of pentobarbital (i.p., >300 mg/kg; Narcoren, Boehringer Ingelheim Vetmedica GmbH, Ingelheim, Germany) and transcardially perfused with 0.1 M phosphate-buffered saline (PBS), followed by fixative (4% paraformaldehyde solution in 0.1 M PBS).The brain tissue was collected, incubated in fixative overnight and stored in 0.1 M phosphate buffer (PB) at 4 C. Following cryoprotection (successive overnight incubations in 15%, 30% sucrose [Cat# 4621.1,Carl Roth GmbH, Karlsruhe, Germany] in PBS) and freezing (À20 C, in blocks of Surgipath FSC 22 Clear, Leica Biosystems, Richmond, IL, USA), brain tissue was sliced on a cryostat (Cat#CM3050 S, Leica Microsystems, Germany) to obtain thin (70 μm for morphological reconstruction and neuronal tracing, 50 μm for cell quantification experiments) parasagittal sections (with the exception of four juxtacellularly labelled cells from two animals which were processed on coronal sections; see cells #1-4 in Figure S2).To visualize the Neurobiotinlabelled cells, brain slices were incubated overnight at 4 C with a fluorescent streptavidin-Alexa Fluor 546 conjugate (1:1000, Cat#S11225, ThermoFisher Scientific, Waltham, Massachusetts) after permeabilization in PBS with 0.1% Triton X-100.The tissue sections were then incubated for 15 min with a 1:1000 DAPI solution (Cat#D1306, Thermo-Fisher Scientific, Waltham, Massachusetts) at room temperature, followed by three washes with PBS.Immunostainings were performed on free-floating tissue sections, essentially as previously described (Balsamo et al., 2022;Ding et al., 2022).The primary antibodies used were rabbit anti-TH (1:1000, Cat#P40101-150, Pel Freez Biologicals, as in Kalinin et al., 2006), mouse anti-TH (1:1000, Cat#MAB318, Sigma-Aldrich), mouse anti-NET (Norepinephrine Transporter; 1:1000, Cat#NET05-2, Mab Technologies), rabbit anti-DBH (Dopamine-beta-Hydroxylase; 1:500, Cat#22806, ImmunoStar) and rabbit anti-NeuN (1:500, Cat#MAB377, Merck Millipore).Secondary antibodies (1:400, Goat anti-Rabbit Secondary Antibody Alexa Fluor 488, Cat#A-11034, Thermo Fisher Scientific, and Goat anti-Mouse Secondary Antibody Alexa Fluor 488, Cat#A-21424, Thermo Fisher Scientific) were incubated at room temperature for 4 h.The sensitivity and specificity of the two anti-TH antibodies were virtually identical, as indicated by the 100% colocalization between the two stainings (639/639 double-positive neurons; not shown).Fluorescence images were acquired via epifluorescence (5Â, 10Â, 20Â; Axio Imager Z1, Carl Zeiss, Oberkochen, Germany) or confocal microscopy (40Â, 63Â; LSM710, Carl Zeiss, Oberkochen, Germany).For each cell, the neurochemical assignment was assessed based on visual inspection of usually multiple serial z-stacks through the soma and confirmed by two independent experimenters.For colour-blind accessibility and display purposes, fluorescence images were converted into green and magenta, and intensities were linearly adjusted.To analyse cell morphologies, a horseradish peroxidase reaction was performed on Neurobiotin streptavidin-labelled sections, followed by conversion with a 3,3 0 -diaminobenzidine (DAB) substrate enhanced with Ni 2+ as previously described (Klausberger et al., 2003;Preston-Ferrer et al., 2016).Somatodendritic domains of neurons were manually reconstructed using Neurolucida (MBF Bioscience, Williston, Vermont).Long-range axons were only partially reconstructed (Figure S3F; n = 4 neurons) and typically consisted of one main branch that exited the LC complex towards the rostral direction.A post hoc adjustment accounted for tissue shrinkage in the Z dimension.The resulting dataset consisted of n = 24 (TH-positive n = 13, TH-negative n = 11) morphological reconstructions of neurons obtained from 17 animals.Quantitative morphological features of neuronal reconstructions were derived using the Neurolucida software (v.2022.2.3).For dendritic structures (Figure S3A-E), the number of nodes (branching points), Sholl intersection profiles (dendritic intersections quantified in concentric spheres of 10 μm radius centred at the soma) and dendritic complexity (see Pillai et al., 2012), defined as Sum of the terminal orders ð þ number of terminalsÞÁ total dendritic length number of primary dendrites , were calculated.Somatic shape and size (Figure S3D) were assessed using the minimum and maximum Feret diameters (respectively, the smallest and largest dimensions of the soma contour, as measured in the reconstruction slice where the contour area was maximal).Neuronal quantification and colocalization analysis (Figures 1b, S1B n = 5 sections from 2 LC nuclei, 2 animals) was performed on confocal microscopy images (40Â, 63Â magnification, z-stacks) using ImageJ (Schindelin et al., 2012) and the Colocalization Object Counter Plugin (Lunde & Glover, 2020).TH-positive cells were quantified using both TH and NeuN signals, as the latter tends to be less intense in TH-positive neurons (Figure 1b; Cannon & Greenamyre, 2009).

| Electrophysiology data analysis
Spikes were detected offline, essentially as previously described (Burgalossi et al., 2011).Briefly, a thresholdaided local maxima detection method was applied to high-pass filtered (cut-off frequency, 300 Hz) voltage traces to identify candidate spikes.Visual inspection of spike waveforms and the first two principal components allowed the spikes to be reliably isolated from artefacts.Parts of recordings displaying evidence of cellular damage (specifically, spike broadening, increased discharge frequency, prolonged negative DC shift; see Herfst et al., 2012;Pinault, 1996) were not included in further analysis.The resulting electrophysiology dataset consists of 50 recordings (in 32 animals) from anatomically identified LC neurons, 23 of which (in 14 animals) include foot shock stimulation.Multiple recordings were conducted (bilaterally: 21 recordings from 10 animals: cells #3-#4, #12-#13, #17-#19, #34-#35, #36-#37, #44-#45, #24-#47, #49-#50, #33-#48, #38; #39-#45; and unilaterally: 12 recordings from 5 animals: cells #1-#2, #5-#6-#7, #8-#9, #22-#23-#24, #38-#39; see Figure S2).In experiments where multiple labelling attempts were performed within the same LC complex, recordings were only included in the dataset if the respective neurons could be identified.Three experiments yielded two neurons (cells #1, #2; cells #8, #9; cells #38, #39; see Figure S2), and two experiments yielded three neurons (cells #5, #6, #7; cells #21, #22, #23; see Figure S2).In two cases, it was unclear which recordings corresponded to which neurons (cells #1, 2; cells #5, 6, 7, see Figure S2); however, since all neurons from each experiment were of the same class (i.e., TH-positive), the respective recordings were included in the analysis.Therefore, our electrophysiological dataset consists of 45 recordings from neurons identified both cytochemically and histologically and five recordings from neurons identified only cytochemically.The broadband juxtacellular voltage signal was used to extract spike waveforms (data snippets À2 to +6 ms aligned to the local maximum).After subtracting the DC offset for each spike waveform, the mean spike waveform was calculated.'Peak-to-trough' (temporal difference between the prominent electropositive local maximum and the subsequent electronegative local minimum of a spike waveform) and 'half-width' (temporal difference between two values of a spike waveform at the halfmaximal amplitude of the peak) durations were derived from the mean spike waveforms.The interspike interval (ISI) coefficient of variation (CV; calculated as the ratio of the standard deviation of the ISI over the mean ISI, as in Softky & Koch, 1993).Mean firing rates, mean spike waveforms, ISI, CV and autocorrelograms were computed based on baseline (spontaneous) firing, recorded prior to the initiation of juxtacellular labelling or foot shock stimulation.Temporal features of the spike waveforms, ISI distributions and autocorrelograms were extracted with sparse principal component analysis (sPCA; Zou et al., 2006) using the first four principal components (Figure S5A-C).Autocorrelograms were computed for the range ±0.5 s and ISI distributions (Figure S5B-C) for 0-1 s, normalized by z-scoring.THpositive and TH-negative neurons were compared using their projections on the first two principal components (Figure S5D-F).
To assess neuronal activity related to foot shock stimuli, we constructed peri-stimulus time histograms (À0.5 to +4.5 s with respect to stimulus onset; bin width, 30 ms).To assess the neuronal response reliability to foot shock stimulation (Figure 3c), we calculated the fraction of trials in which an evoked spike occurred within 100 ms of the foot shock onset.To determine the response latency to foot shock, we calculated the median latency of the first spikes occurring within 100 ms after stimulus onset.To quantify the strength of foot shock modulation, we computed a Response Modulation Index (Balsamo et al., 2022), defined as FRstimulus -FRbaseline FRitiþFRbaseline , where FR iti represents the mean firing rate during +1.5 to +4.5 s post-stimulus onset, and FR baseline represents the mean firing rate during spontaneous activity prior to the initiation of foot shock stimulation.
To further validate our conclusions, we compared the electrophysiological properties of TH-positive and THnegative cells in the following conditions: In all the above cases, the differences in firing rate, spike half-width and peak-to-trough duration between THpositive and TH-negative cells trended in the same direction as in our cumulative dataset (as in Figure 2a,c; data not shown), thus indicating that our conclusions are unlikely to be significantly biased by unintentional anatomical biases and/or cell type misclassification.We note that potential inadvertent misclassification would likely diminish the observed differences between THpositive and TH-negative neurons.

| PCA embedding
To embed neurons in two-dimensional space, we performed PCA on the standardized matrix that has the number of neurons (n = 50) as rows and the electrophysiological features as columns (Figure S5D).The arrows in the PCA biplot (Figure S5F) are obtained by VS= ffiffiffiffiffiffiffiffiffiffi ffi n À 1 p with V as the eigenvector, S as the eigenvalues and n as the number of data points.Showing the PCA loadings has the advantage of making the relations between features interpretable: the cosine of products between any two arrows approximates the covariance between them.

| Logistic regression
We used LASSO logistic regression (Tibshirani, 1996) to classify neurons into TH-positive and TH-negative classes.We included 17 features (Figure S5D, F, G) and used a cross-validated grid search to identify the optimal regularization (c = 1.21, inverse regularization strength).The data set was split into 80% training and 20% testing, and we used five folds for the cross-validation.'Accuracy' was defined as the fraction of correct predictions (including both true positives and true negatives) among the total number of cases examined, as implemented in the Python library scikit-learn (Pedregosa et al., 2011).We note that the classifier approach was primarily intended as a confirmatory approach to the physiological differences observed between the two neuronal classes.The classifier has overall limited accuracy (76%) and was trained on a limited number of observations (n = 30 THpositive, n = 20 TH-negative); hence, its applicability to other datasets and/or experimental preparations is likely to be limited.Foot shock response metrics were not included in this analysis, given the even more limited number of observations (n = 14 TH-positive, n = 9 THnegative).

| Statistical analysis
Statistical analysis of pairwise comparisons was performed with the two-sided Mann-Whitney-Wilcoxon test.P-values were not adjusted for multiple comparisons.This is an exploratory study, and p-values should be interpreted accordingly.Results are reported as mean ± sample standard deviation unless otherwise specified.

| RESULTS
To characterize the electrophysiological and morphological properties of LC neurons, we performed juxtacellular recording and labelling of single neurons in anaesthetized mice (Figure 1a).The cytochemical identity of the labelled neurons was assessed via immunoreactivity for tyrosine hydroxylase (TH)-the rate-limiting enzyme in catecholamine biosynthesis, and a classical marker for LC noradrenergic neurons (Baker et al., 1989;Carter et al., 2010;Ito et al., 2020;Manger & Eschenko, 2021;Pickel et al., 1975;Schmidt et al., 2019).In line with previous work, TH-positive (putatively noradrenergic) neurons were most abundant at posterior-dorsal levels of the LC (classically referred to as the LC 'nuclear core', Figure S1); however, outside of the LC 'core', TH-positive neurons were more intermingled with putatively non-noradrenergic (TH-negative) cells-the latter being identified by the lack of immunoreactivity for TH, and the presence of immunoreactivity for the pan-neuronal marker NeuN, and DAPI (Figures 1b and S1).These observations are in line with previous work on the rodent LC, which indicated that at the antero-ventral aspect of the LC complex, scattered noradrenergic neurons intermingle within the lateral dorsal tegmental nucleus (LDTg) and/or central grey (Grzanna & Molliver, 1980;Swanson, 1976).In this study, we did not restrict our analyses to the LC 'nuclear core' but instead focused on this extended LC complex (Figure S1; for review, see Manger & Eschenko, 2021).
Figure 1c-h shows a representative recording from an identified LC neuron that was positive for TH immunoreactivity (TH-positive; Figure 1c).This neuron displayed a multipolar morphology, and its dendritic tree sampled a broad territory outside the area containing TH-positive neuronal somatas (Figure 1d).Furthermore, this THpositive neuron fired at a relatively low rate (0.51 Hz) and displayed a broad spike waveform (Figure 1f) and a low tendency to fire spikes at short interspike intervals (Figure 1g)-in line with previous observations (Aston-Jones & Bloom, 1981a;Takahashi et al., 2010;Totah et al., 2018Totah et al., , 2019;;Williams et al., 1984).Furthermore, this TH-positive neuron exhibited a multiphasic excitatory response to foot shock stimulation, consisting of a short-latency component, followed by additional activity peaks that outlasted the stimulus itself (Chen & Sara, 2007;Hirata & Aston-Jones, 1994), followed by post-activation inhibition (Aghajanian et al., 1977;Cedarbaum & Aghajanian, 1978;Ennis & Aston-Jones, 1986) (Figure 1h).Notably, this neuron reliably responded to the stimulation, with 109 out of 109 stimulation trials displaying an evoked spike within a 100 ms window following stimulus onset (Response Reliability = 1, see Section 2).
Figure 1i-n shows a representative example from an identified TH-negative neuron.Like the TH-positive cell (Figure 1d), this neuron displayed a multipolar morphology (Figure 1j) and fired at a relatively low rate (0.73 Hz).However, compared with the TH-positive cell (Figure 1f), this neuron displayed a narrower spike waveform (Figure 1l) and distinct temporal firing patterns, characterized by a higher tendency to fire at short interspike intervals (Figure 1m).Notably, this TH-negative neuron also reliably responded to the foot shock stimulus (Figure 1n) with high trial-to-trial reliability (Response Reliability = 0.87, i.e., 87 out of 100 response trials with an evoked spike within a 100 ms window following stimulus onset, Figure 1n; see Section 2).
Altogether, we recorded 50 identified neurons in the extended LC complex of anaesthetized mice (see -Figure S2).Of these neurons, 60% (30 out of 50) were found to be positive and 40% (20 out of 50) negative for TH immunoreactivity, consistent with the relative abundance of these cell types from anatomical analysis (Figure S1C).A subset of neurons with high-quality filling (TH positive, n = 13; TH-negative, n = 11; see Section 2) was reconstructed for morphological analysis.On average, TH-negative and TH-positive neurons displayed largely overlapping morphometric features (Figure S3), indicating that neurochemical identity (i.e., TH expression) cannot be directly inferred from morphological criteria alone.Nevertheless, we acknowledge the possibility that due to the relatively small dataset of reconstructed neurons (n = 24), possible (subtle) morphological differences between the two cell classes might have gone undetected.
Previous studies differentiated heterogeneous LC neurons by their average spontaneous firing rate and spike width, which motivated us to compare these characteristics between TH-positive and TH-negative neurons (e.g., Tortorelli et al., 2022;Totah et al., 2018).On average, compared with TH-positive neurons, THnegative cells displayed lower spontaneous mean firing rates (TH-pos, 1.21 ± 0.78 Hz; TH-neg, 0.77 ± 0.62 Hz; p = 0.036; Figure 2a) and narrower spike waveforms (spike half-width durations; 0.83 ± 0.18 Hz; TH-neg, 0.69 ± 0.14 Hz; p = 0.008; Figure 2b,c).Moreover, THpositive neurons exhibited a higher tendency to fire spikes at short ISI (Figure 2d,e).Notably, in most neurons, we observed the presence of a 'shoulder' on the descending phase of the juxtacellular spike waveform (Figure S4).As shown in other cell types (Diamantaki et al., 2016), the presence of a shoulder resulted in relatively long peak-to-trough spike durations in our dataset (see Figure S4).These observations are consistent with prior extracellular recordings of LC units, which also showed 'broad' spike waveforms featuring a 'shoulder' in the descending phase of the spike, as well as a 'notch' in the ascending phase (Figure S4C; e.g., Aston-Jones et al., 1991;West et al., 2009;Williams et al., 1984).
To further investigate the differences in electrophysiological properties between TH-positive and THnegative neurons, we extracted temporal features of spike waveforms, interspike interval distributions and spike autocorrelograms using sPCA (Figure S5A-C; see Section 2).To validate differences in in vivo electrophysiological properties between TH-positive and TH-negative neurons, we undertook two independent approaches.First, we performed PCA on all LC neurons using spike-waveform and spike-train metrics (see Figure S5D-F and Section 2).Indeed, TH-positive and TH-negative cells were partially separated along the two first principal dimensions (Figure S5E).Second, a logistic-regression classifier trained on the same electrophysiological features was able to separate TH-positive and TH-negative neurons with reasonable accuracy (cross-validated accuracy: 76%), with the largest weights being assigned to spike waveform features and spontaneous mean firing rate (Figure S5G-H).Altogether, these data indicate that TH-positive neurons display distinct spontaneous electrophysiological features compared with putative non-adrenergic cells.
One limitation of the current dataset is that the identified neurons were classified according to the expression of a single marker (i.e., TH immunoreactivity).TH expression in the LC has been shown to be activitydependent, at least under certain experimental conditions (e.g., Osterhout et al., 2005;Zigmond et al., 1989), thereby raising the possibility of false-negative TH classification in our dataset.Moreover, the biological variability of TH expression levels among LC neurons might also reduce confidence in cell identification, especially in cells with relatively low TH expression.To address these concerns, we first stained the LC for additional noradrenergic markers, DBH and NET (Fan et al., 2014;Grzanna & Molliver, 1980;Loughlin et al., 1986;Shipley et al., 1996;Swanson, 1976;Tillage et al., 2020;Wagatsuma et al., 2018), and quantified the degree of colocalization with TH (Figure S7).The large majority of DBH-positive and NET-positive cells were also found to be immunoreactive for TH (TH and NET double-positive neurons; 98.15 ± 2.12%; n = 1489 neurons, n = 5 nuclei; TH and DBH double-positive neurons: 96.96 ± 1.83%; n = 1614 neurons, n = 5 nuclei, Figure S7), thus pointing to a relatively low probability of false TH-negative neurons in our dataset.Second, restricting the electrophysiological analysis to the cells with 'high confidence' of immunohistochemical classification (based on the overall quality and intensity of the fluorescence signal) led to qualitatively similar results regarding spontaneous electrophysiological properties of TH-positive and THnegative neurons (see Section 2; not shown).Altogether, this evidence indicates that our results are not significantly biased by potential misclassification of neuronal identities based on TH expression.

| DISCUSSION
Brainstem noradrenergic neurons, through their widespread projections to the rest of the central nervous system, are in a privileged position for controlling a wide range of physiological brain functions, including arousal, attention and memory (Aston-Jones & Cohen, 2005;Eschenko, 2018;Poe et al., 2020;Sara, 2009;Totah et al., 2019).Despite their crucial importance, the small size and deep location of this neuronal population and the lack of clear electrophysiological correlates of anatomical boundaries have made the in vivo characterization of noradrenergic singe-unit activity particularly challenging.
In the present study, we aimed at bridging this gap by taking advantage of the in vivo juxtacellular recordinglabelling method.Specifically, we applied this technique to a preparation that has been extensively used to study the in vivo activity of the LC, namely, foot shock stimulation in anaesthetized animals (e.g., Chen & Sara, 2007;Chiang & Aston-Jones, 1993;Neves et al., 2018;Safaai et al., 2015;Totah et al., 2018).This approach offers two main advantages compared with conventional extracellular techniques.Firstly, the high signal-to-noise ratio of juxtacellular spike signals enabled the precise identification of spikes, even during highly synchronous LC population activity, thus providing accurate ('ground truth') spiking patterns from individual LC cells.Secondly, by labelling recorded cells and subsequent immunohistochemical and morphological analysis, we could link in vivo recorded activity patterns to individual neurons.
The main goal of the study was two-fold: first, to characterize the in vivo activity of identified noradrenergic neurons within the mouse LC complex; second, to explore whether these properties can be used for separating noradrenergic cells from the heterogenous surround of non-noradrenergic neurons.We found that noradrenergic neurons (identified by the expression of TH, THpositive) displayed significantly different spontaneous electrophysiological features [i.e., mean firing rates (Figure 2a), spike waveforms (Figure 2b-c), temporal spiking properties (Figure 2d-e) and average foot shock responses (Figure 3)] compared with putative nonnoradrenergic (TH-negative) cells.Spontaneous in vivo electrophysiological features alone were sufficient for classifying TH-positive and TH-negative cells with 76% accuracy (see Figure S5G-H).These observations are consistent with recent studies, which also describe neuronal subpopulations with relatively narrow spike waveforms in the mouse, rat and primate LC (Su & Cohen, 2022;Tortorelli et al., 2022;Totah et al., 2018).Morphologically, TH-positive and TH-negative neurons displayed largely overlapping features (Figure S3), which prevented cell type classification based solely on morphological criteria.Notably, we observed a high degree of morphological variability even within the same class (Figure S3), pointing to further heterogeneity waiting to be resolved.
One limitation of the current study relates to the definition of noradrenergic and non-noradrenergic cell types based on TH immunohistochemistry.Although this approach has been validated and extensively employed in previous studies of the rodent and human LC complex (Baker et al., 1989;Carter et al., 2010;Ito et al., 2020;Manger & Eschenko, 2021;Pickel et al., 1975;Schmidt et al., 2019), it is known that TH expression can be modulated by neuronal activity (Osterhout et al., 2005;Zigmond et al., 1989).Therefore, since TH-negative neurons in our dataset were defined by the 'absence' of marker expression, false negatives arising from technical sensitivity and/or biological variability cannot be formally ruled out.However, colocalization analysis with additional noradrenergic markers (DBH and NET) revealed a high degree of consistency with TH expression (Figure S7), indicating that under our experimental conditions, the occurrence of false TH-negative cells is likely to be very low.Moreover, restricting the analysis to cells with 'high confidence' of immunohistochemical classification led to qualitatively similar results (see Section 2), indicating that our conclusions are unlikely to be significantly biased by inadvertent cell misclassification.
The relative distribution of TH-positive and THnegative cells within the mouse LC complex was consistent with previous results.Specifically, TH-positive neurons were most abundant within the densely packed 'nuclear core' of the LC, where TH-negative cells represented a small minority of all neurons (Figure S1B, C; Manger & Eschenko, 2021).On the other hand, noradrenergic cells were more scattered-and hence more intermingled with TH-negative neurons-outside of the 'nuclear core' region (e.g., at more antero-ventral levels, where intermingling of noradrenergic neurons with LDTg and/or central grey has been described (Gasparini et al., 2021;Grzanna & Molliver, 1980;Swanson, 1976).Since different nonnoradrenergic cell types with various neurotransmitter phenotypes (e.g., GABAergic, cholinergic and glutamatergic) have been described within (and especially around) this most antero-ventral aspect of the rodent LC complex (Aston-Jones et al., 2004;Breton-Provencher & Sur, 2019;Corteen et al., 2011;Gasparini et al., 2021;Gong et al., 2020;Jin et al., 2016;Kuo et al., 2020;Luskin et al., 2022), our TH-negative group likely includes heterogeneous subpopulations of neurons.For example, TH-negative neurons (Figure S2) could possibly also include LDTg neurons, which have been shown to exhibit broad action potentials and respond to sensory stimuli (Koyama et al., 1994).In the present study, these possibly diverse TH-negative neurons were grouped in order to test whether TH-positive neurons display unique electrophysiological features compared with the heterogenous surround of non-noradrenergic cell types.We also note that noradrenergic neurons have also been shown to be heterogeneous (Chandler et al., 2014;McKinney et al., 2022;Tortorelli et al., 2022;Totah et al., 2018), but our limited sample size prevented further within-class analysis.Future work will be required to resolve additional structure-function relationships within these heterogeneous TH-negative (and TH-positive) cell classes, which was beyond the scope of the present study.
In this study, in vivo recorded and identified THpositive neurons were sampled from a region extending beyond the densely packed LC 'nuclear core' (Figures S1A and S2).Whether noradrenergic subpopulations within this extended LC complex are functionally distinct remains to be demonstrated.Our relatively small dataset, together with the difficulty of unequivocally assigning neurons to specific subregions of the LC complex, particularly in parasagittal sections, prevented the analysis of possible relationships between the anatomical location of the recorded cells and in vivo activity patterns.
We envision that the registration of neuronal morphologies onto a standardized anatomical reference frame of the LC complex-following the approach being established for the neocortex (e.g., Lang et al., 2011;Oberlaender et al., 2012)-will enable the integration of data collected from different studies and laboratories and will be instrumental for resolving possible structurefunction relationships at the single cell level.
Another limitation of this study is the use of xylazine, an alpha-2 adrenergic receptor agonist, in the anaesthetic mixture (see Section 2; Aghajanian & VanderMaelen, 1982).However, in our preparation, several electrophysiological features of LC activity could be recapitulated.For example, spontaneous firing rates (Figure 2a), spike waveform (Figures 2b,c and S4), spike train features (Figure 2d,e) and foot shock responses (Figures 3 and S6) were consistent with estimates from prior work based on different anaesthetics or species (e.g., Chen & Sara, 2007;Takahashi et al., 2010;Totah et al., 2018).This indicates that several electrophysiological features classically associated with LC neurons were preserved in our preparation.We acknowledge, however, that despite similarities, a direct comparison between studies using different anaesthetics and/or animal species (e.g., rats vs. mice) is not entirely warranted because species-and/or anaesthetic-specific effects cannot be excluded a priori.For example, while spontaneous firing rates in our study were largely consistent with estimates from previous in vivo work (Aston-Jones & Bloom, 1981a;Chen & Sara, 2007;Totah et al., 2018Totah et al., , 2019)), Sugiyama et al. (2012) reported higher firing rates under a similar recording configuration (in vivo cellattached recordings) but in urethane anaesthetized rats.We also acknowledge that the use of xylazine, which acts on alpha2-adrenoceptors, might have suppressed the firing rates of noradrenergic cells in our study and, hence, narrowed the difference in excitability between the THpositive and TH-negative groups.Future in vivo recordings in drug-free animals will be required to extend our observations to the natural system during awake behaviour.
The advent of optotagging techniques in the study of LC circuits has provided an unprecedented level of spatial and temporal resolution (for review, see Tanguay et al., 2023;Totah et al., 2019).However, these systemslevel approaches are not optimally suited for resolving structure-function relationships, as they are typically restricted to targeting 'broad' cell types (i.e., defined by single marker expression) and do not provide access to morphological information.Additionally, known limitations of spike-sorting algorithms, particularly during epochs of highly synchronous activity (Franke et al., 2015;Garcia et al., 2022;Lewicki, 1998;Royer et al., 2010) such as the phasic response to foot shocks, have hindered the precise characterization of LC activity patterns.Here, we provide, to the best of our knowledge, the most extensive dataset of in vivo recorded and morphologically identified brainstem noradrenergic neurons.Our juxtacellular recordings from single, anatomically identified neurons provide 'ground-truth' firing patterns, which can be useful for benchmarking and/or validating current spike-sorting-based approaches for the classification of noradrenergic unit activity.
Altogether, our dataset of identified neurons reveals in vivo electrophysiological correlates of noradrenergic neurons in the mouse LC complex and contributes to the ongoing research efforts aiming at the multimodal characterization of brainstem noradrenergic neurons.
,C, and S7; TH-NeuN: n = 7 sections from 3 LC nuclei, 2 animals; TH-NET: n = 5 sections from 3 LC nuclei, 3 animals; TH-DBH: F I G U R E 1 In vivo recordings from anatomically identified TH-positive and TH-negative neurons within the mouse LC complex.(a) Schematic of the experimental preparation, showing juxtacellular recording in the mouse LC.The parasagittal sectioning plane for post hoc histological analysis is indicated with a red dotted line.D, dorsal; L, lateral; LC, locus coeruleus complex.(b) Fluorescence images of a parasagittal tissue section through the LC (as in a).Left, staining for tyrosine hydroxylase (TH; magenta).Scale bar, 100 μm.Right, highmagnification insets showing staining for NeuN (green), TH (magenta) and DAPI (white).TH-negative cells, positive for NeuN and DAPI, are indicated with dotted lines.Scale bar, 20 μm.D, dorsal; R, rostral.(c) Left, fluorescence image of a parasagittal section through the mouse LC, stained for TH (magenta), showing a representative identified TH-positive neuron (green; Neurobiotin, Nb) recorded in vivo.Scale bar, 50 μm.Right, high-magnification detail corresponding to the area indicated with a dotted line in the left image, showing the neuron's soma (green; Nb) colocalized with TH (magenta).Scale bar, 10 μm.(d) Morphological reconstruction of the neuron shown in a, superimposed on an outline of the TH fluorescence signal (magenta).Scale bar, 50 μm.(e) Top, representative high-pass filtered juxtacellular voltage trace from the neuron shown in c, d, recorded during foot shock stimulation (FS).Note the reliable spiking in response to the stimuli (indicated in grey).Bottom, high-magnification view on a single stimulation trial (dotted box).(f) Average spike waveform for the identified neuron shown in c, d.The spike half-width is indicated.(g) Cumulative distribution of interspike intervals (ISI) during spontaneous activity for the representative neuron shown in c, d.(h) Raster plot (top) and peristimulus time histogram (bottom) during foot shock stimulation (grey area) for the neuron shown in c, d. (i-n) Same as c-f, but for a representative TH-negative neuron.Scale bars in i, 50 μm (left), 10 μm (right).Scale bar in j, 50 μm.
FR baseline represents the mean firing rate during 1 s preceding the stimulus onset, FR stimulus represents the mean firing rate during stimulus delivery and Max (FR stimulus ) represents the peak response within 0-0.5 s of the peristimulus time histogram.To analyse the post-activation firing suppression by foot shock stimulation, a post-activation suppression index was calculated as FRitiÀFRbaseline ð Þ 1. Across different mediolateral levels of the LC (medial, n = 7 TH-positive, n = 4 TH-negative neurons; intermediate n = 19 TH-positive, n = 13 TH-negative neurons; lateral, n = 1 TH-positive, n = 2 THnegative neurons; n = 3 TH-positive and n = 1 THnegative neurons were not assigned to any mediolateral level due to histological assessment performed on coronal sections).2. Within versus outside the posterior-dorsal aspect of the LC complex (Figure S1B, C), which contains the highest density of noradrenergic cells ('inside,' n = 17 TH-positive, n = 13 TH-negative neurons; 'outside,' n = 13 TH-positive and n = 7 TH-negative neurons).3.By restricting the analysis to neurons with 'high confidence' of immunohistochemical classification (based on the overall quality and intensity of the fluorescence signal; 'high confidence,' n = 27 TH-positive and n = 12 TH-negative neurons; 'low confidence,' n = 3 TH-positive, n = 8 TH-negative neurons).

F
I G U R E 2 Electrophysiological properties of TH-positive and TH-negative neurons.(a) Average spontaneous firing rates of THpositive (n = 30; magenta) and TH-negative (n = 20; green) neurons.Lines indicate mean values.P-value is indicated (Mann-Whitney-Wilcoxon test, two-sided).(b) Superimposed average spike waveforms for TH-positive and TH-negative neurons (n and conventions as in a).Shaded regions indicate the standard error of means.(c) Spike half-width durations for TH-positive and TH-negative neurons (n and conventions as in a).(d) Representative juxtacellular spike traces from TH-positive (magenta) and TH-negative (green) neurons recorded during spontaneous activity.Insets show high magnification of a spike-double and -triplet from the TH-negative neuron.Scale bars, 1 mV, 1 s and 100 ms (insets).(e) Mean cumulative distributions of interspike intervals for TH-positive and TH-negative neurons (n and conventions as in a).

F
I G U R E 3 Foot shock responses of TH-positive and TH-negative neurons.(a) Top, z-scored peristimulus-time activity of TH-positive (n = 14) and TH-negative (n = 9) neurons during foot shock stimulation (FS; red).Bottom, average z-scored responses from the peristimulus activity shown above.Shadows indicate the standard deviation.(b) Foot shock Modulation Indices for TH-positive (n = 14, magenta) and TH-negative (n = 9, green) neurons.Mean values are denoted by lines.P-value is indicated (Mann-Whitney-Wilcoxon test, two-sided).(c) Foot shock response reliability for TH-positive and TH-negative neurons.Same n and conventions as in b.(d) Representative high-pass filtered juxtacellular voltage traces from a TH-positive neuron (top) and a TH-negative neuron (bottom) during FS (red).Note the longlasting post-activation suppression during inter-trial intervals (ITI) for the TH-positive neuron.Asterisks indicate artefacts.Scale bars, 2 mV, 10 s.(e) Average firing rates during ITIs for TH-positive and TH-negative neurons.Same n and conventions as in b.(f) Post-activation suppression index (see Section 2), which quantifies the degree of firing rate suppression in ITIs compared with baseline (spontaneous firing), for TH-positive and TH-negative neurons.Same n and conventions as in b.