Neocortical long-range inhibition promotes cortical synchrony and sleep

Sleep and rest are essential for mammalian physiology. Their disruption leads to widespread physiological and cognitive impairments and is linked to a wide range of disorders, including insomnia, sleep apnoea, narcolepsy, depression, attention deficit hyperactivity disorder and schizophrenia, and can even lead to death⁵. Transitions between alertness and low-arousal states, such as quiet wakefulness and sleep, occur on a timescale of seconds and are accompanied by pronounced changes in neocortical activity patterns (oscillatory activity or cortical states)¹. Low-arousal states, including slow-wave sleep (SWS) and quiet immobility, are characterized by synchronous low-frequency cortical fluctuations and coordinated spiking activity, often referred to as synchronized cortical states. These alternate with desynchronized states associated with periods of high arousal and active behaviour, during which low-frequency cortical activity is suppressed¹. GABAergic inhibitory neurons (INs) have been implicated as key regulators of arousal-dependent neocortical activity⁴. Although fast-spiking parvalbumin-expressing INs have been shown to promote desynchronized states—which dominate periods of alertness—by enhancing gamma-band activity^2,3, much less is known about the identity and function of IN populations that regulate the synchronized cortical states of low-arousal periods.

The diversity of cortical INs has been recognized for nearly a century⁶, but assigning specific functions to individual classes of IN has proved challenging, in part because established IN classes are themselves highly heterogeneous⁷. For example, somatostatin (Sst)-expressing INs, which are often treated as a monolithic group, can be subdivided into more than ten subclasses with distinct morpho-electric and transcriptomic properties⁷. In this study, we focus on a transcriptionally homogeneous and evolutionarily conserved subtype of Sst-expressing INs characterized by co-expression of Sst, chondrolectin (Chodl), neuronal nitric oxide synthase (Nos1) and the neurokinin-1 receptor (Tacr1). These cells, hereafter referred to as Sst-Chodl cells, correspond to the Nos1 (or nNOS for immunoreactivity)-expressing cortical IN population described in previous studies^7,8,9. Sst-Chodl cells are GABAergic INs that differ considerably from canonical, locally projecting neocortical interneurons in that they have long-range projection axons that can extend across millimetres in the mouse cortex and centimetres in larger brains^10,11. Despite their extreme sparsity—comprising less than 1% of cortical GABAergic neurons⁹—Sst-Chodl cells are highly conserved across vertebrate species, from salamanders to humans¹². Previous ex vivo studies have suggested that putative Sst-Chodl cells are likely to be active during SWS^13,14,15, raising the possibility that they contribute to synchronized states. However, their activity and functional roles in vivo have remained difficult to investigate owing to a lack of tools for their specific targeting.

Here, using an intersectional genetic strategy to selectively target Sst-Chodl cells, we show that these neurons form dense regional and inter-areal intracortical connections and are preferentially active during periods of cortical synchrony, including quiet wakefulness and SWS. Activation of Sst-Chodl cells promotes synchronized cortical states and sleep-related behaviours. Together, these findings identify Sst-Chodl cells as a cortical-circuit element that links neocortical synchrony to sleep regulation, consistent with emerging evidence that cortical activity contributes directly to the control of sleep^{16,17,18,19,20,21,22,23,24,25}.

We used intersectional genetics to selectively target Sst-Chodl cells (Fig. 1a and Extended Data Fig. 1). To evaluate the extent of their projections, we used the Sst^flp;Chodl^cre intersection to sparsely label individual neurons, enabling single-cell anatomical reconstruction of individual Sst-Chodl cells across the neocortex. We reconstructed the full morphology of 16 Sst-Chodl cells across all layers in the primary visual cortex (V1) and higher visual areas (Fig. 1b), and compared their projection pattern with those of parvalbumin (Pvalb) and non-Chodl Sst (other Sst cells) reconstructed cells (Fig. 1c–g). Whereas Pvalb and other Sst cells had local arborization, rarely crossing area boundaries, we found that Sst-Chodl cells had both dense local arborization within visual areas and long-range axons travelling millimetres from the soma across broad neocortical areas (Fig. 1b–g and Supplementary Tables 1 and 2). The projections of Sst-Chodl cells were exclusively intracortical and ipsilateral, broadly targeting nearly all cortical areas, with the notable exception of far-anterior (for example, frontal pole and prelimbic area) and ventral (for example, gustatory area) regions (Fig. 1e). Of note, projection patterns were highly similar between Sst-Chodl cells that originated in V1 and those that originated in higher visual areas (Extended Data Fig. 1g,h and Supplementary Tables 1 and 2), and between medial and lateral Sst-Chodl cells (Extended Data Fig. 1i–l and Supplementary Tables 1 and 2). We encountered local and regional projections across all layers, whereas the long-range projections were preferentially found in upper layers (Fig. 1h).

We next quantified the density of the Sst-Chodl projections from V1 to other brain regions by injecting the intersectional AAV8-Ef1a-CreOn/FlpOn-oScarlet vector into V1 of Sst^flp and Nos1^creER double-transgenic mice (Extended Data Fig. 2), driving robust cytosolic labelling of Sst-Chodl cells (Fig. 1i). At the population level, we annotated the oScarlet-labelled arbor of Sst-Chodl cells across the brain and found broadly ipsilateral neocortical projections (with small contralateral V1 projections), with dense innervation of visual areas (200–600 mm of arbor per mm³ of tissue; Fig. 1j–l and Extended Data Fig. 2a–i) and extensive inter-areal projections that targeted most of the neocortex, including retrosplenial, auditory, somatosensory and frontal motor areas (Fig. 1l). Notably, only a limited set of far-anterior and dorsal neocortical regions lacked detectable innervation (agranular insular area, frontal pole, gustatory areas, infralimbic area and orbital area). Consistent with the single-cell reconstruction data (Fig. 1b–e), we also observed small but reproducible projections from V1 Sst-Chodl cells beyond the neocortex, targeting ipsilateral para-hippocampal structures, including the subiculum and the entorhinal cortex (Fig. 1m and Supplementary Table 3).

Using whole-brain tissue clearing and light-sheet imaging, we compared Sst-Chodl arborization across mice labelled in V1 or in the primary somatosensory cortex (S1) (Extended Data Fig. 2j–o and Supplementary Video 1). Similarly to cells in V1, S1 Sst-Chodl cells exhibited dense local arbors across layers and long-range ipsilateral projections confined to the neocortex and preferentially travelling through superficial layers (Extended Data Fig. 2j–n). Whereas the inter-areal projections from V1 Sst-Chodl cells extensively targeted retrosplenial areas, the long-range projections of S1 Sst-Chodl cells more densely innervated areas surrounding the injection site, such as frontal, motor and visual areas (Extended Data Fig. 2j–o).

Together, these data show that Sst-Chodl cells originating from distinct neocortical areas share a common feature: they are long-range projecting neurons that broadly target the ipsilateral neocortex.

To identify the presynaptic partners of Sst-Chodl cells, we used an intersectional genetic, monosynaptic retrograde rabies tracing strategy from Sst-Chodl cells in V1 (Extended Data Fig. 3a and Supplementary Table 4). Presynaptic neurons were located almost exclusively within neocortical areas, with only a small fraction detected in the thalamus (less than 4% of all labelled cells; Extended Data Fig. 3b,c). Within the neocortex, labelled neurons were restricted to the ipsilateral hemisphere, distributed mainly across intermediate layers (2–5) and most densely represented in V1 (Extended Data Fig. 3d–i). Immunohistochemical analyses revealed that presynaptic neurons comprised both glutamatergic (CTIP2⁺, putative pyramidal tract; SATB2⁺, putative intratelencephalic) and GABAergic (PV⁺, VIP⁺ and SST⁺) populations, with the majority co-expressing SATB2 and VIP, and the fewest co-expressing SST and nNOS (Extended Data Fig. 3j,k).

Together, these data indicate that Sst-Chodl cells integrate a broad diversity of regional excitatory and inhibitory inputs from all of the major glutamatergic and GABAergic classes that we probed.

To test whether the activity of Sst-Chodl cells varies across sleep and wake states, as suggested by previous studies of FOS expression^14,15, we expressed a GCaMP calcium indicator using either viral vectors or transgenic lines (see Methods) in Sst^flp;Nos1^creER or Sst^flp;Chodl^cre mice. We then used two-photon calcium imaging to measure the activity of labelled cells across arousal and network synchronization states, using pupillometry, locomotion, contralateral cortical local field potentials (LFPs; delta band, 1–4 Hz), electromyography (EMG) and facial movements (Fig. 2a,b and Supplementary Video 2). Because these state measurements were strongly correlated (Extended Data Fig. 4a), we combined them into a one-dimensional ‘arousal score’, corresponding to the first principal component of these state variables (Fig. 2b and Extended Data Fig. 4b). Behavioural state was classified into four categories: wakefulness, which was subdivided into movement and quiet wakefulness (quiet wake; QW); SWS; and rapid eye movement (REM) sleep (Fig. 2b; see Methods).

We found that a large proportion of labelled cells in superficial layers (L1–2/3) from both intersectional mouse lines exhibited intense activity during low-arousal, highly synchronous states such as SWS and QW, and were strongly suppressed during REM sleep and movement, when the neocortex is desynchronized (95 out of 111 cells; Fig. 2b,c and Extended Data Fig. 4c). A minor subset of labelled cells showed the opposite pattern of activation, being most active during periods of high movement (16 out of 111 cells; Extended Data Fig. 4c,d). This subset is consistent with our sequencing and immunohistochemistry analyses indicating the presence of a minor population of non-Sst-Chodl cells in both intersectional crosses, as well as with previously described^8,26 small heterogenous populations of SST⁺nNOS⁺ neurons that do not express Chodl (Extended Data Fig. 4e).

We next examined the consistency of Sst-Chodl cell activity across cortical layers using microprism-based imaging (Extended Data Fig. 4f). Similar to superficial-layer neurons, Sst-Chodl cells recorded in deep layers were strongly active during SWS and QW and suppressed during movement and REM sleep (14 out of 16 cells; Extended Data Fig. 4g). Across all layers, Sst-Chodl cells that were active during low-arousal states exhibited strong positive correlations with low-arousal metrics, including delta-band spectral power and time elapsed since last locomotion (that is, stillness duration), and negative correlations with high-arousal measures such as EMG, facial motion, locomotion and pupil diameter (Fig. 2d). Activity among these cells was also highly correlated and dependent on the mouse’s arousal level, with stronger correlations observed during low-arousal states (Fig. 2d and Extended Data Fig. 4h). Consistent with this relationship, separating LFP epochs on the basis of periods of high versus low Sst-Chodl cell activity revealed that these neurons were most active when cortical networks were highly synchronized, as indicated by increased (high) delta power (Extended Data Fig. 4i–k).

To examine the temporal relationship between Sst-Chodl cell activity and cortical dynamics, we implanted flexible 32-channel probes beneath the imaging window, to enable two-photon imaging of Sst-Chodl cell activity with high temporal resolution alongside electrophysiological recordings (Fig. 2e–k; see Methods). Sst-Chodl cells exhibited state-dependent changes in activity that closely tracked fluctuations in LFP delta power (Fig. 2f and Extended Data Fig. 4l). Sst-Chodl cell activity closely co-varied with delta power, rather than anticipating future changes in cortical synchrony (Fig. 2g,h and Extended Data Fig. 4m). In addition, neither the mean activity of Sst-Chodl cells during SWS bouts nor other examined features were correlated with the duration of SWS bouts (Extended Data Fig. 4n,o). Together, these results suggest that the activity of Sst-Chodl cells reflects ongoing sleep-related cortical dynamics, rather than anticipating forthcoming state transitions.

We next examined Sst-Chodl cells on a shorter timescale during SWS, which alternates between phases characterized by a near-complete suppression of spiking during upward deflections of the cortical LFP (‘DOWN’ states) and phases of high spiking activity (‘UP’ states)²⁷ (Fig. 2i). Aligning Sst-Chodl cell activity with DOWN state transitions revealed an increase in activity before the onset of the DOWN state, whereas other neurons, simultaneously recorded as multi-unit spiking activity from the same recording site, exhibited prominent rebound firing after DOWN-to-UP transitions (UP onset; Fig. 2j,k). Notably, Sst-Chodl cell activity was not correlated with either the duration or the amplitude of the subsequent DOWN state (Extended Data Fig. 4p–q), suggesting that these neurons are preferentially engaged near the termination of cortical UP states rather than during the quiescent DOWN phase itself.

Finally, to evaluate the specificity of patterns of Sst-Chodl cell activity, we compared their state modulation with that of the broader Sst population and other neuronal classes using two publicly available datasets. First, using the Allen Brain Observatory dataset²⁸, we examined the correlations between Sst cell activity and arousal metrics, including locomotion and pupil area state (Extended Data Fig. 5a). In this dataset, the whole Sst population was correlated predominantly with increased arousal (Extended Data Fig. 5b,c). Next, using an independent dataset²⁹, we compared the activity of Sst-Chodl cells to that of other simultaneously recorded neuronal types (Extended Data Fig. 5d–f). In contrast to arousal-activated interneuron subtypes, including regular-spiking Sst cells, Sst-Chodl cells showed negative correlations with arousal (Extended Data Fig. 5d–f). Among IN populations, Sst-Chodl cells exhibited their strongest positive correlation with fast-spiking Sst subtypes, particularly a numerically minor Tac1-expressing population (Extended Data Fig. 5e,f).

Together, our results suggest that, unlike other Sst cell populations, Sst-Chodl cells closely mirror changes in arousal states in terms of their activity, becoming most active during highly synchronized cortical states.

To test whether the activity of Sst-Chodl cells causally promotes neocortical synchrony, we acutely inserted 64-channel linear silicon probes equipped with tapered optical fibres into the V1 of head-fixed mice virally expressing channelrhodopsin (ChR2) in Sst-Chodl cells (Fig. 3a,b). We quantified multiple complementary metrics of cortical synchrony derived from LFP and spiking activity to compare network states with and without optogenetic stimulation of Sst-Chodl cells (Fig. 3c–n and Extended Data Figs. 6 and 7).

We first examined changes in LFP power across cortical depth during stimulation relative to non-stimulation (spontaneous) periods. Optogenetic activation of Sst-Chodl cells led to a rapid and robust increase in delta band power, across all layers, with the largest effects observed in deeper layers (Fig. 3e,f). We next assessed spiking synchrony among well-isolated single units (putative single neurons) by computing spike train cross-correlations. Sst-Chodl cells activation increased the magnitude of spiking synchrony across all layers (Fig. 3g,h and Extended Data Fig. 6a), independently of stimulation frequencies (Extended Data Fig. 6b) or unit class (regular spiking versus fast spiking; Extended Data Fig. 6c). Although Sst-Chodl cell activation produced a slight reduction in the overall firing rate, spike-train structure and burst properties were mostly preserved (Extended Data Fig. 6d–h), indicating that enhanced synchrony did not result from increased firing but instead reflected changes in temporal coordination across neurons. Moreover, single-shank stimulation using a 6-shank (200-µm spacing), 64-channel silicon probe reliably increased delta power and spiking synchrony at the stimulated site. Notably, these effects propagated across all shanks, spanning around 1 mm of cortex (Extended Data Fig. 7a–c). This spatial spread indicates that activation of Sst-Chodl cells induces distributed synchronization across visual cortical networks.

Given the close relationship between LFP fluctuations and spiking activity, we next assessed spike–field coupling during stimulation by quantifying spike phase locking values. Under baseline conditions, spikes were preferentially locked to the delta phase of the LFP (Fig. 3i), and Sst-Chodl cell stimulation strongly enhanced this coupling (Fig. 3i,j and Extended Data Fig. 6i). As an extreme manifestation of phase locking, DOWN states became more frequent and longer-lasting during stimulation, with event rate and duration both significantly increased (Fig. 3k and Extended Data Fig. 7d–f).

We then asked whether Sst-Chodl cell activation modulates cortical synchrony across different states. Comparing stimulation epochs with matched spontaneous periods revealed significant increases in delta-band power during SWS, QW and movement (Fig. 3l and Extended Data Fig. 7i,j; REM was excluded owing to insufficient bout duration). Spiking synchrony was similarly enhanced across all three states (Fig. 3m and Extended Data Fig. 6j–l). Of note, synchrony increased even during movement, when Sst-Chodl cells are typically not active endogenously (Fig. 3l–n). When comparing the magnitude of stimulation-induced synchrony to physiological levels, we found that Sst-Chodl cell activation during QW increased delta power and spiking synchrony to levels comparable with those seen in spontaneous SWS, whereas stimulation during movement produced a delta power comparable with that observed in spontaneous QW (Fig. 3n and Extended Data Figs. 6j,k and 7i,j).

Together, these findings demonstrate that activation of Sst-Chodl cells robustly enhances neocortical synchrony, recapitulating key features of endogenous synchronized cortical states observed during SWS.

We next asked whether Sst-Chodl cells influence neuronal activity at more-distant neocortical sites, by determining the spatial extent of their synaptic targets. We expressed ChR2 in Sst-Chodl cells in V1 and performed whole-cell patch-clamp recordings from putative postsynaptic neurons in acute neocortical slices up to about 2.4 mm from the injection site (Fig. 4a,b and Extended Data Fig. 8). Optogenetically evoked inhibitory postsynaptic currents (oIPSCs) were observed in both pyramidal neurons and interneurons, with the highest response rates in V1 (the injection site), particularly in layers 1, 5 and 6 (Fig. 4c,d and Extended Data Fig. 8b–h). Although the proportion of responsive cells decreased with increasing distance (Chi-squared including layer 1 cells, P < 0.001; Chi-squared excluding layer 1 cells, P = 0.009), oIPSCs remained detectable in nearly one-third of neurons recorded around 2 mm from the injection site (Fig. 4d). Consistent with anatomical tracing (Fig. 1h and Extended Data Fig. 2c), layer 1 exhibited the highest proportion of responsive cells across all recording sites (Fig. 4d and Extended Data Fig. 8b), whereas layer 6 exhibited the largest response amplitudes (Fig. 4d and Extended Data Fig. 8b–c). Finally, in a subset of experiments, we confirmed that these oIPSCs were GABAergic and monosynaptic (Fig. 4e and Extended Data Fig. 8e,f). Together, these results align with our anatomical data and show that Sst-Chodl cells provide direct, long-range inhibition across neocortical layers.

Given the broad ipsilateral intracortical projections of Sst-Chodl cells, we next asked how far the synchronization induced by their activation propagates across the neocortex. To address this, we expressed ChR2 in Sst-Chodl cells in V1 and recorded LFPs from eight ipsilateral neocortical locations positioned well beyond the spatial extent sampled in our multishank recordings (Extended Data Fig. 7a–c), spanning distances of approximately 2–6 mm from the stimulation site (Fig. 4f). Compared with inter-trial intervals (ITIs), optogenetic activation of Sst-Chodl cells increased low-frequency LFP power across most of the recording sites, extending from the visual cortex to frontal motor areas (Fig. 4g–i). This effect was strongest at sites closest to the stimulation region, and declined progressively with increasing distance (Fig. 4i). Although we cannot exclude contributions from polysynaptic corticocortical propagation, these results suggest that Sst-Chodl cell activation can drive neocortical synchronization over millimetre-scale distances, with a graded decay consistent with long-range intracortical influence.

Previous work has suggested that during periods of intense SWS, putative Sst-Chodl cells are active throughout the neocortex¹⁵. Therefore, we hypothesized that widespread neocortical activation of these cells might not only enhance cortical synchrony (Figs. 3 and 4 and Extended Data Figs. 6 and 7), but also promote behavioural features of sleep. To test this, we used an INTRSECT-based chemogenetic approach to express an excitatory DREADD (designer receptors exclusively activated by designer drugs) in Sst-Chodl cells throughout the neocortex using 22 injected sites (11 per hemisphere) (Fig. 5a and Extended Data Figs. 9 and 10). Using two-photon imaging, we confirmed that systemic administration of a low dose of the DREADD agonist clozapine-N-oxide (CNO; 0.5 mg kg⁻¹) increased the activity of Sst-Chodl cells, mostly independently of the arousal state of the mice (Extended Data Fig. 9b,c and Supplementary Videos 2 and 3). To test the effect of pan-neocortical Sst-Chodl cell activation on sleep behaviour, we performed counterbalanced systemic injections of vehicle or CNO and monitored LFP activity and movement in freely behaving mice in their home cage for 2 h during the light (inactive) phase of their circadian cycle (Extended Data Fig. 9d). Chemogenetic activation of Sst-Chodl cells increased the duration of both SWS and REM sleep, reduced sleep latency and increased the number of SWS bouts without affecting bout duration (Fig. 5b–g and Extended Data Fig. 9e). These changes were accompanied by decreased locomotion and increased nest occupancy (Fig. 5h,i and Extended Data Fig. 9f), consistent with a reduced proportion of time spent in movement relative to quiet wakefulness (Extended Data Fig. 9g,h) and with the observed increase in sleep in CNO-treated mice (Fig. 5b–g).

Similarly to optogenetic manipulation (Fig. 3), pan-neocortical activation of Sst-Chodl cells increased neocortical LFP power predominantly in low-frequency bands, and independently of behavioural state (Fig. 5j,k and Extended Data Fig. 9j). The effects were prominent in the neocortex but minimal in the hippocampus (Extended Data Fig. 9i–k), indicating a preferential neocortical engagement. Notably, treating control mice lacking DREADD expression with CNO did not alter their sleep–wake architecture (Extended Data Fig. 9l–o).

Because the original behavioural experiments were performed during the light (inactive) phase, when mice naturally spend more time asleep, we next tested whether activation of Sst-Chodl cells would have similar effects during the dark (active) phase (Fig. 5l and Extended Data Fig. 10). Indeed, mice that were treated with CNO during the dark (active) phase exhibited increased SWS duration and reduced sleep latency, compared with vehicle-treated mice (Extended Data Fig. 10b–g). Notably, the amounts of sleep induced by Sst-Chodl cell activation during the dark (active) phase were comparable to—and in some cases exceeded—baseline levels of sleep observed during the light (inactive) phase of vehicle-treated mice (Fig. 5m), indicating a robust sleep-promoting effect.

Together, these results indicate that, independent of circadian phase, global activation of neocortical Sst-Chodl cells is sufficient to promote cortical synchrony and sleep.

Here we show that long-range, neocortical Sst-Chodl cells—a unique neuronal subtype^7,9 that is highly conserved across evolution¹²—are selectively active during low-arousal, synchronized states and promote patterns of synchronous neocortical activity. We propose that Sst-Chodl cells synchronize neocortical networks through their extensive local and long-range projections, positioning them as a key cortical component in the regulation of sleep.

The prevailing framework for sleep regulation emphasizes subcortical structures as primary drivers of sleep initiation and maintenance, with the cortex viewed mainly as a passive downstream recipient³⁰. However, accumulating evidence supports a more active role for the neocortex in sleep–wake regulation. Our findings extend this framework by identifying Sst-Chodl cells as a distinct cortical population capable of shaping neocortical synchrony and sleep-related dynamics. In this context, we find that Sst-Chodl cells exhibit exceptionally dense projections and high synaptic connectivity across all cortical layers in the cortical region in which their cell bodies are located, making them well suited to promoting local cortical synchrony. This feature could be particularly relevant for forms of local sleep, in which circumscribed cortical regions exhibit sleep-like patterns in the absence of overt behavioural sleep³¹. In addition, we show that Sst-Chodl cells extend long-range axons that inhibit neurons millimetres away from their somata, suggesting a mechanism by which these cells contribute to synchronization across distant neocortical regions³². Notably, we find that Sst-Chodl neurons provide particularly strong synaptic inhibition and LFP modulation in deep cortical layers, in which pyramidal neurons associated with synchronized states are concentrated^33,34. Pyramidal neurons in these layers have been shown to contribute to sleep regulation²², raising the possibility that Sst-Chodl cells influence sleep homeostasis by shaping the activity of deep-layer cortical circuits. Future studies will be required to determine whether Sst-Chodl cells outside sensory cortices receive inputs from, or project to, canonical sleep-promoting regions in the hypothalamus or midbrain, or whether their influence on sleep is mediated mainly through intracortical mechanisms.

Although cortical Sst-expressing INs as a group have been implicated in the regulation of cortical states and linked to sleep-related processes^20,21, their functional roles have been difficult to reconcile across studies, with reported effects spanning gamma^35,36, beta³⁷, theta³⁸ and delta²⁰ rhythms. Single-cell transcriptomic profiling has revealed substantial heterogeneity in the Sst population, indicating that distinct Sst subclasses subserve divergent, and potentially opposing, functions^8,29. In this context, differences in experimental approaches might preferentially engage different Sst subtypes, contributing to variability in reported outcomes. Consistent with this interpretation, we identified a minor population of cells within the Sst-Nos1 intersection that were anti-correlated with arousal and probably correspond to non-Sst-Chodl, type II Nos1 cells expressing low levels of SST and nNOS (refs. ^8,26). These cells might participate in neurovascular control during movement^39,40, consistent with a functional role distinct from the promotion of synchronized cortical states. Our results highlight the importance of transcriptomically informed targeting strategies for dissecting IN subclass function and underscore the utility of this approach for understanding cortical-circuit mechanisms.

Beyond their synaptic inhibitory effects, Sst-Chodl cells express other signalling molecules that could contribute to their function. Notably, these neurons are likely to represent the largest neuronal source of nitric oxide in the neocortex^39,41. Mice knockout for Nos1 exhibit disrupted sleep architecture, including a diminished ability to sustain prolonged SWS bouts and an impaired homeostatic response to sleep deprivation¹⁴. Similarly, selective deletion of Nos1 from Sst cells results in deficits in delta rhythms⁴². Potential targets of nitric oxide signalling from Sst-Chodl cells include parvalbumin-expressing INs^41,43 and components of the neurovascular system³⁹. The slow, synchronized electrical activity promoted by Sst-Chodl cell activation might therefore be mirrored in vascular dynamics, potentially supporting rhythmic fluid movement during glymphatic clearance in sleep⁴⁴.

The mechanisms that govern the state-dependent activation and suppression of Sst-Chodl cells remain to be elucidated. Although thalamocortical circuits are well known to regulate cortical UP and DOWN state dynamics^1,30, our monosynaptic tracing experiments indicate that Sst-Chodl cells receive predominantly neocortical inputs, with minimal thalamic contributions. Our findings show that the activity of Sst-Chodl cells increases before DOWN state onset, which suggests that these neurons participate in regulating UP–DOWN transitions, similar to other IN classes^45,46,47. Activation of Sst-Chodl cells could arise through disinhibition mediated by VIP-expressing INs, consistent with our rabies tracing data, or through changes in arousal-related neuromodulatory tone, such as acetylcholine⁴⁸. This interpretation is consistent with previous studies showing that Sst-Chodl cells receive extensive neuromodulatory inputs^7,29,41,48 which might not be captured by our rabies-based tracing approaches^7,41,48,49. Another possibility is that sleep-promoting substances (somnogens), such as adenosine, could activate these cells during synchronized cortical states³⁰. In addition, substance P has been shown to activate Sst-Chodl cells, and local application of substance P promotes cortical slow-wave activity and electrophysiological signatures of SWS^50,51, although its precise role in sleep regulation remains unclear. In this context, we find that Sst-Chodl cells receive input from both PV- and SST-expressing INs, which represent potential sources of substance P. Identifying additional G-protein-coupled receptors expressed by Sst-Chodl neurons will be important to elucidate the signalling pathways through which sleep- and arousal-related factors engage this cell population.

Overall, this work identifies Sst-Chodl cells as a key cortical circuit element associated with low-arousal neocortical states and sleep. The properties of this genetically distinct IN subtype indicate that cortical circuits make an active contribution to sleep-related dynamics, complementing the established roles of subcortical structures in sleep regulation. Together with previous studies^{16,17,18,19,20,21,22,23,24,25}, our findings highlight an underappreciated role of the cortex in shaping sleep states. Given that sleep disturbances are a prominent feature of many neuropsychiatric, neurodevelopmental and neurological disorders, this highly conserved cell population could represent a useful entry point for future mechanistic and translational investigations²⁵.

Mice

All mouse handling and maintenance was performed according to the regulations of the Institutional Animal Care and Use Committee of the Albert Einstein College of Medicine (protocol 00001393). Sst^flp+/+; Nos1^creER+/− and Sst^flp+/−;Nos1^creER+/− mice were used interchangeably in this study, with no differences detected between the two groups (Nos1^creER, Jax 014541; Sst^flp, Jax 031629). Nos1^creER+/− mice were crossed with Sst^flp+/+ mice to obtain Sst^flp+/−;Nos1^creER+/− mice and Sst^flp+/−;Nos1^creER+/+ were crossed with Sst^flp+/+ mice to obtain Sst^flp+/+;Nos1^creER+/− mice. Sst^flp+/+;Nos1^creER+/− mice were crossed with Ai210^+/+ mice (a gift from the Allen Institute)⁵² to obtain mice for in vivo imaging experiments. Chodl^cre+/+ mice (a gift from the Allen Institute)⁵² were crossed with Sst^flp+/+ or Sst^flp+/− mice. Adult male and female mice aged 50 days or older were used in this study. Mice were kept under a 12 h light–dark cycle (lights on 07:00 or lights on 16:00 for ZT18 chemogenetic experiments) under standard housing conditions (20–24 °C and 30–70% humidity).

Cell type classification and sequencing

Data were retrieved from a previous study⁵².

Abbreviations of brain regions

ACA, anterior cingulate area (d, dorsal); AIp, agranular insular area, posterior; APr, area prostriata; Aud, auditory cortical area (d, dorsal; p, primary; po, posterior; v, ventral); AuT, auditory and temporal cortical area; BS, brainstem; CB, cerebellum; CLA, claustrum; ECT, ectorhinal area; ENT, entorhinal area (l, lateral; m, medial dorsal); EPd, endopiriform nucleus, dorsal; FrM, frontomotor cortex; HPF, hippocampal formation; HVAs, higher visual areas; Mo, motor cortical area (p, primary; s, secondary); NC, neocortex; OB, olfactory bulb; Olf, olfactory areas; PAR, parasubiculum; PERl, perirhinal area; PIR, piriform area; POST, postsubiculum; PRE, presubiculum; RSP, retrosplenial cortical area (agl, lateral agranular; d, dorsal; v, ventral); RSPagl, retrosplenial area, lateral agranular; RSPd, retrosplenial area, dorsal; S1, primary somatosensory cortex; SS, somatosensory cortical area (bfd, barrel field; ll, lower limb; m, mouth; p, primary; ul, upper limb; tr, trunk; un, unassigned; s, supplemental); StP, striatum–pallidum; SUB, subiculum; TEa, temporal association area; V1, primary visual cortex; VIS, visual area (a, anterior; al, anterolateral; am, anteromedial; l, lateral; li, laterointermediate; p, primary; pl, posterolateral; pm, posteromedial; por, postrhinal; rl, rostrolateral); VISC, visceral area.

Surgical procedures

Mice were anaesthetized with isoflurane (5% by volume for induction and 1–2% for maintenance), placed on a stereotaxic frame and kept warm with a closed-loop heating pad. For pain management, mice were given intraperitoneal (i.p.) meloxicam at 2.5 mg kg⁻¹ and local lidocaine on the scalp before skin incision. The following coordinates are distance (in mm) from bregma and dorsoventral (DV) values refer to brain surface.

For viral injections, we minimized brain damage by performing burr holes, keeping a thin layer of the bone intact where the glass micropipettes could penetrate.

For wide-labelling morphological reconstructions, 50 nl viral vector was injected into a burr hole split across two levels (DV −0.25 and DV −0.55) at anteroposterior (AP) −3.3, mediolateral (ML) ±2.7 to target primary visual cortex (V1); or at AP −1, ML ±3 to target primary somatosensory cortex (S1).

For silicon probe-coupled optogenetic experiments, four burr holes were drilled centred over V1 (AP −3.3, ML ±2.7) spaced around 1 mm apart, and positioned to avoid blood vessels visible through the skull. Virus (500 nl) was injected into each burr hole split across two levels (DV −0.25 and DV −0.55) for a total of 2 µl virus.

For calcium imaging-related surgeries, mice were also administered dexamethasone at 4 mg kg⁻¹ 1 h before craniotomy. A 3-mm craniotomy was made over V1, and, if applicable, virus was injected into the centre of this craniotomy while the brain was kept moist with hydrated gelatin surgical foam. Then a cranial window was placed over the opening and sealed to the skull with cyanoacrylate glue. For imaging Sst-Chodl cells in superficial layers, the cranial window consisted of a stack of three coverslips: two 3-mm #1 glass coverslips (Warner Instruments) attached with Norland Optical Adhesive 71 to a 5-mm glass coverslip, allowing the 3-mm glass to sit against the dura to discourage bone regrowth⁵⁴. For imaging Sst-Chodl cells in deep layers, durotomy was performed before implantation of a 1.5-mm microprism (A1 coated, 4531-0023, Tower Optical) attached with Norland Optical Adhesive to a single 3-mm glass coverslip.

For pan-neocortex viral delivery, 22 burr holes were split bilaterally and 400 nl of viral vectors was injected at the following coordinates: AP +1, ML ±3, DV −1; AP +1, ML ±1.8, DV −0.4; AP −0.5, ML ±3.8, DV −1.5; AP −0.5, ML ±2.6, DV −0.4; AP −2, ML ±4, DV −1.2; AP −2, ML ±3.2, DV −0.3; AP −2, ML ±1.2, DV −0.3; AP −3.5, ML ±3.5, DV −0.3; and AP −3.5, ML ±2, DV −0.3; and 800 nl of viral vectors was injected at the following coordinates: AP 2.5, ML ±1.2, DV −1; and AP 0, ML ±1, DV −0.5.

For rabies tracing experiments, a total volume of 500 nl of the helper mixture (2:1 ratio optimized glycoprotein over avian tumour virus receptor A (TVA)) split across two levels (DV −0.25 and DV −0.55) was injected into a single burr hole at AP −3.3, ML ±2.7. After a minimum of four weeks to allow robust expression of the helper constructs, rabies vector was delivered as a single injection of 150 nl at the same AP–ML coordinates and DV −0.4. Mice were perfused 10 days after rabies delivery to capture the full extent of monosynaptic retrograde labelling while minimizing cytotoxicity.

For calcium imaging and whole-neocortex chemogenetic experiments, a separate burr hole was made in contralateral above V1 and a made-in-house microwire array connected to a 16-channel omnetic (A79038-001, Omnetics Connector), used for measuring LFPs, was inserted. The LFP wires consisted of five to eight tungsten wires, 50 µm in diameter, spanning from the middle layers of V1 area into cornu ammonis area 1 (CA1). For calcium imaging combined with flexible 32-channel probes (tetrode configuration, NeuralThread⁵⁵), an electrode attached to a stainless steel pole with adhesive polyethylene glycol was inserted perpendicular to the brain at the border of the craniotomy. Then, the polyethylene glycol was dissolved by applying saline continuously for 10 min on top of the pole to ensure its disengagement from the electrode. After pulling the pole straight up out of the brain, the cranial window was positioned above the electrode.

For multi-site LFP and optogenetic experiments, eight burr holes were made at the following coordinates: anterior: AP+ 1.35, ML −2.8 and AP +0.7, ML −1.2; distal: AP +0.05, ML −2.8 and AP −0.6, ML −1.2; medial: AP −1.25, ML −2.8 and AP −1.9, ML −1.2; proximal: −2.55, ML −2.8 and −3.2, ML −1.2. A made-in-house microwire array consisting of eight tungsten wires of 200 µm in diameter, were aligned to the mouse skull, connected to a 16-channel omnetic and inserted in the surface of the brain. Virus (500 nl) was injected ipsilateral of the multi-LFPs at AP −2.7, ML −2.7 split across two levels (DV −0.25 and DV −0.55), and a fibre-optic cannulae of 400-µm diameter and 6-mm length was positioned at the entrance of the burr hole.

Injections were performed using a Nanoject III system at a rate of 1–2 nl s⁻¹ through glass micropipettes that were pulled and then ground to bevel with 40-µm diameter using a Naragishe diamond wheel. For LFPs and flexible probes, a tungsten ground wire of 200 µm diameter attached to a gold pin was inserted into the cerebellum, and a headpost was implanted as previously described⁵⁶.

All implants were secured to the skull using Optibond or Super-Bond followed by dental cement. When Nos1^creER+/− mice were used, tamoxifen (Thermo Fisher Scientific) was administered at least 2 weeks after surgery. A stock solution of 20 mg ml⁻¹ tamoxifen in corn oil was injected i.p. at 0.1 mg tamoxifen per g body weight for 5 consecutive days. Mice were left for a minimum of 4 weeks for CreER-mediated recombination before experimental use.

The following AAV viral vectors were used in this study at concentrations of 10¹²–10¹³ vg ml^–1 (written in the format AAV vector, serotype, source):

• Ef1a-CreOn/FlpOn-oScarlet, 8, provided by the laboratory of K.D.

• Ef1a-CreOn/FlpOn-GCaMP6m, 8, provided by the laboratory of K.D.

• hSyn-CreOn/FlpOn-ChR2-EYFP, DJ, UNC Vector Core

• EF1a-fDIO-mCherry, 5, Addgene (114471)

• nEF-CreOn/FlpOn-hM3Dq-mCherry, 8, provided by the laboratory of K.D.

• nEF-CreOn/FlpOn-TVA-mCherry WPRE, 8, Gene Vector and Virus Core (GVVC; AAV-197), Wu Tsai Neurosciences Institute, Stanford University

• Ef1a-CreOn/FlpOn-oG WPRE, 8, GVVC (AAV-198)

• EnvA-DG-Rabies-EGFP, not applicable (N/A), Salk Vector Core

• iSSTe4-ChR2-mCherry, 1, provided by the laboratory of G. Fishell⁵⁷

Immunohistochemistry

Mice were perfused transcardially with ice-cold 4% paraformaldehyde (PFA) and brains were dissected and post-fixed in 4% PFA for 1 h at 4 °C. After sucrose cryoprotection, brains were embedded in OCT, cryosectioned at 20 µm and adhered to glass slides. For staining, sections were incubated in a blocking solution containing 1.5% donkey serum, 1% Triton-x-100 for 1 h at room temperature. Primary antibodies were applied overnight at 4 °C, followed by washes and 1-h incubation in secondary antibodies. Tissue sections were mounted in Prolong Gold with DAPI (Thermo Fisher Scientific) and imaged on a Zeiss Axioscan slide scanner and a Zeiss LSM 880 confocal microscope for high-resolution imaging.

The following antibodies were used in this study (written in the format antigen target, concentration, vendor, product number):

• Somatostatin (SST), 1:250, Millipore, MAB354

• Somatostatin-14 (SST), 1:1,000, Peninsula, T-4103

• Parvalbumin (PV), 1:1,000, SYSY, 195004

• Vasoactive intestinal peptide (VIP), 1:250, SYSY, 443005

• Neuronal nitric oxide synthase (nNOS), 1:500, Abcam, ab1376

• CTIP2, 1:500, Abcam, ab18465

• SATB2, 1:500, SYSY, 327004

• GFP (reactive against EYFP), 1:500, Invitrogen, A-111222

• RFP (reactive against mCherry), 1:250, Chromotek, 5f8

Tissue clearing and light-sheet imaging

Mice were prepared as described above for morphological reconstruction. Mice were then transcardially perfused with ice-cold 1× PBS followed by 4% PFA. Brains were dissected and post-fixed in 4% PFA at 4 °C for 24 h. Brains were processed with SHIELD reagents (LifeCanvas Technologies) before beginning active tissue clearing using the SmartBatch+ system (LifeCanvas Technologies) using manufacturer protocols. Tissues were refractive index-matched using EasyIndex (LifeCanvas Technologies) and imaged on a SmartSPIM light-sheet microscope (LifeCanvas Technologies) with a 3.9× magnification objective (NA: 0.2, manufactured by Thorlabs and modified by LifeCanvas Technologies). Data were aligned to the Allen Mouse Brain Common Coordination Framework (CCF) using NeuroInfo (MBF Bioscience).

Whole-brain sparse labelling and reconstructions

Whole-neuron morphology data were generated using HortaCloud⁵⁸, an open-source, cloud-based neuron reconstruction platform. Morphological reconstructions were performed on gene-defined, sparsely labelled individual neurons imaged across the entire mouse brain using two-photon fluorescent micro-optical sectioning tomography (fMOST)^59,60. For each neuron, the full local and long-range axonal arbor and dendritic trees were manually identified and traced by an experienced annotator. A second independent annotator then proofread the initial reconstruction and added missed branches and removed incorrectly identified segments. All branch points and terminal endings were verified to ensure complete reconstruction of each neurite. Any discrepancies between annotators were resolved by consensus before the reconstructed cell was deemed final.

Slice electrophysiology and oIPSCs

Acute coronal slices of the neocortex were prepared from adult Sst^flp+/+;Nos1^creER+/− mice of either sex injected with AAV-CreOn/FlpOn-ChR2-EYFP (1 µl split across two depths) (DV −0.25 and DV −0.55) at AP −3.3, ML ±2.7, 5–8 weeks before recording. Mice were anaesthetized with isoflurane and transcardially perfused with ice-cold NMDG-based cutting solution containing (in mM) 92 NMDG, 2.5 KCl, 1.25 NaH₂PO₄, 30 NaHCO₃, 20 HEPES, 25 glucose, 2 thiourea, 5 sodium ascorbate, 3 sodium pyruvate, 0.5 CaCl₂·2H₂O and 10 MgSO₄·7 H₂O, adjusted to pH 7.3–7.4 with 5 M hydrochloric acid. The brain was dissected rapidly and sectioned at 250–260 µm with a vibratome (VT1200S, Leica), and allowed to recover in the same solution for 5-10 min at 34 °C before transfer to warm artificial cerebrospinal fluid (ACSF) containing (in mM) 120 NaCl, 25 NaHCO₃, 1.25 NaH₂PO₄, 3 KCl, 1 MgCl₂, 1.5 CaCl₂, 11 glucose, 3 sodium pyruvate and 1 sodium ascorbate equilibrated with O₂ and 5% CO₂. Slices were incubated for another 30 min and then kept at room temperature until recording for up to 8 h. For a subset of experiments, we used a sucrose-based cutting solution containing (in mM) 90 sucrose, 60 NaCl, 5 MgCl₂, 2.75 KCl, 1.25 NaH₂PO₄, 1.1 CaCl₂, 9 glucose, 26.5 NaHCO₃, 3 sodium pyruvate, 1 sodium ascorbate, equilibrated with 95% O₂ and 5% CO₂. Slices were transferred directly to warm ACSF after cutting under these conditions.

We verified the presence of ChR2-EYFP-expressing fibres in each slice at various distances from the injection site before recording. Putative postsynaptic neurons were recorded across all cortical layers and at various distances from the injection site at near-physiological temperature (around 32 °C) with an internal solution containing (in mM) 15 CsCl, 120 caesium gluconate, 8 NaCl, 10 HEPES, 2 MgATP, 0.3 NaGTP, 0.2 EGTA and 2 mg ml⁻¹ biocytin for post-hoc anatomical analysis (pH adjusted to 7.2 with CsOH; osmolarity adjusted to 290 mOsm). The chloride reversal potential was around −44 mV, and GABAergic currents were recorded at a holding potential of 0 mV. Visually guided whole-cell recordings were obtained with patch pipettes of around 3-MΩ resistance pulled from borosilicate capillary glass (BF150-86-10, Sutter Instrument). Electrophysiology data were acquired using a Sutter dPATCH amplifier (Sutter Instrument), digitized at 10 kHz and filtered at 5 kHz. oIPSCs were evoked with 0.5-ms whole-field pulses of blue light (CoolLED) at a frequency of 0.2 Hz. To isolate inhibitory currents in voltage-clamp, the following receptor antagonists were added to the bath solution: 2 µM (R)-CPP, 5  µM NBQX and 1.5 µM CGP to block NMDA, AMPA and GABA_B receptors, respectively. To confirm GABAergic identity, oIPSCs were blocked with 50 µM gabazine SR95531 in some experiments. To confirm that oIPSCs were monosynaptic, 0.5 µM TTX and 100 µM 4-AP were included in the bath solution in a subset of experiments. All drugs were purchased from Abcam, Tocris or HelloBio. After recording, the patch pipette was withdrawn slowly to allow resealing of the membrane and slices were fixed in 4% PFA overnight. Biocytin-filled neurons were labelled with streptavidin–Alexa 647 using standard protocols. Confocal z-stacks of streptavidin-labelled neurons and EYFP⁺ Sst-Chodl fibres were taken on a Zeiss LSM 880 microscope at 1–2-µm increments. z-stacks were processed in ImageJ and their positions in the cortex were recorded.

Habituation and head restraint

Mice were briefly habituated to handling for several days before surgery. Mice were allowed to recover for 2 days before continuing habituation to handling. After a minimum of 1 week after implantation, mice were gradually exposed to the head-fixation apparatus (https://github.com/janelia-experimental-technology/Rodent-Belt-Treadmill), which consists of a low-friction rodent-driven belt treadmill. The head-fixation duration was gradually increased over 2 weeks until the mice were comfortable with multi-hour head fixation sessions.

To promote sleep while head-fixed, mice were given multiple habituation sessions on the recording rig before data collection. We ensured that the treadmill was oriented horizontally with no slope, with the distance between the head fixation level and the treadmill properly adjusted. The set-up was cleaned thoroughly to minimize odours from other mice, and a radiant heat lamp was provided.

Videography

Videos of mice were acquired using FlyCapture 2 software, at 30 Hz using Blackfly machine-vision ethernet-enabled cameras (FLIR) equipped with a Basler lens (model c125-1620-5m) for head-fixed experiments and a Computar lens (TG4Z2813FCS-IR 0.33-Inch) for freely moving experiments. Rigs were illuminated with infrared LED arrays, similar to methods described previously⁶¹. The camera was configured to send transistor–transistor logic (TTL) pulses to synchronize videography with other data streams.

In vivo two-photon calcium imaging

GCaMP was expressed either using an AAV-CreOn/FlpOn-GCaMP6m viral vector or through transgenic expression in the Ai210 mouse line (a Cre- and Flp-dependent GCaMP7f). No substantial differences were observed between the two approaches. Mice were imaged on a custom Bergamo two-photon microscope coupled to a Ti:Sapphire laser (Mira 700, Coherent). Emitted light was collected through a 10× 0.5-NA long-working-distance objective (TL10X-2P, Thorlabs). Images were typically acquired with ThorImage software at 1.4 frames per second with a resolution of 512 × 512 pixels and, for experiments coupled with flexible probes, at 14.6 frames per second with a resolution of 128 × 128 pixels. For microprism data, cell positions were estimated from the distance of the soma relative to the visible-damage layer produced by microprism insertion, which was used as a reference for the cortical surface (estimation relative to the Allen Brain Atlas). Data streams were synchronized using TTL pulses collected on an RHD USB interface board (Intan Technologies) at 20 kHz. Before imaging, novel objects were placed in the mouse’s home cage, one per hour for up to 4 hours, to promote exploration and subsequently facilitate sleep during imaging.

Acute in vivo electrophysiology and optogenetics

Mice were injected with AAV-CreOn/FlpOn-ChR2-EYFP and implanted with headposts as described above. After viral expression, tamoxifen induction and treadmill habituation, mice were administered dexamethasone (4 mg kg⁻¹, i.p.) 1 h before craniotomy. An approximately 3-mm craniotomy was made over the injection site (V1).

A 64-channel linear silicon probe (H3 probe, Cambridge NeuroTech) physically coupled to a tapered optical fibre was slowly (1 µm s⁻¹, over around 20 min) inserted into V1. The craniotomy was kept moisturized during the recording by placing a small volume of silicon oil on the surface of the brain. State measurements were made as described below. Between recording days, the craniotomy was protected with KWIK-CAST silicon elastomer (World Precision Instruments). Mice were recorded once per day for 3–4 days.

Recordings were split into blocks of around 30-min spontaneous periods (without stimulation) and around 30-minute periods of stimulation. Several types of stimulation were used, including sinusoids (1 Hz, 4 Hz, 10 Hz and 40 Hz), flat pulses and white noise, each delivered for 30 s followed by a 30-s ITI. Blue light was provided by a fibre-coupled LED (MF470F4, Thorlabs). Signals to control LED light intensity were generated using custom-written MATLAB software and delivered to an NI DAQ card (NI-PCIe-6323) to be output to the LED control box (LEDD1B, Thorlabs). No light artefacts were detected in our recordings.

Across more than 700 recorded single units, plus additional multi-units from 10 mice, no optotagged units were found, underscoring the extremely low abundancy of this neuronal population and the difficulty of capturing the cells directly in silicon probe recordings.

Data were acquired using an Intan RHD2000 interface board at 20 kHz.

Multi-LFPs and optogenetics

Mice were injected with AAV-CreOn/FlpOn-ChR2-EYFP and implanted with headposts as described above. Recordings were split into blocks of around 30-min spontaneous periods (without stimulation) and around 30-min periods of stimulation using a 10-Hz sinusoid frequency. Similar to the acute recordings, the stimulation periods consisted of 30 s of blue light followed by a 30-s ITI. The optical and acquisition hardware used was the same as that described above.

Freely moving behavioural assay

Mice were injected with AAV nEF-CreOn/FlpOn-hM3Dq-mCherry or control EF1a-fDIO-mCherry, implanted with LFP wires as described above and single-housed after surgery. Mice were habituated to wired tethering via a 16-channel digital headstage with an accelerometer (Intan Technologies) and connected to a motorized commutator (Doric Lenses or Neurotek). To minimize stress and noise distraction, mice were kept in their home cage in an acoustic foam box for 2 h per day for 3–5 days. Mice were also habituated to receiving systemic saline injections at the beginning of the session for at least 2 days before monitoring their behavioural state. A consistent time of day was used for each mouse during the light (inactive) phase (ZT3–ZT7) or during the dark (active) phase (ZT18), and this was maintained throughout the experiment. No differences in sleep behaviour were detected across specific ZT windows used within the light (inactive) phase, but sessions during the light (inactive) and the dark (active) phase were analysed separately.

After habituation, mice received i.p. injections of either vehicle (saline) or CNO (HelloBio) at 0.5 mg kg⁻¹. This dose has been shown to have no effect on the sleep of control mice without DREADD treatment^62,63, and we confirmed this in our control experiment (12 sessions from 6 mice during the light phase, Extended Data Fig. 9l–o). The initial treatment was counterbalanced across mice, and the alternative treatment was administered the following day. Thus, each session was paired with its opposite treatment for analysis (vehicle versus CNO or CNO versus vehicle). Mice completed two vehicle and two CNO sessions, and data for each treatment were averaged across the two sessions. To ensure sufficient wakefulness, only mice that remained awake for at least 60% of the time during vehicle sessions were included in the analysis within the light or dark phase. We obtained a total of 28 sessions during the light phase from 14 mice and 12 sessions during the dark phase from 6 mice.

LFP data were acquired at 1,250 Hz using an Intan RHD2000 interface board.

Behavioural state scoring

Behavioural state was divided on the basis of a combination of the acquired state measurements. First, to score sleep states we used previously published and validated methods implemented in the Buzcode toolbox from the Buzsaki lab^64,65. In brief, this method provides automatic state scoring by distinguishing SWS from wake using the slope of the power spectrum (a large slope occurs during high delta periods of sleep). REM sleep is identified as periods of high theta/delta ratio with low EMG. The automatic scoring is then visualized and manually refined by experts. For optogenetic and chemogenetic experiments, state was scored using LFPs that were recorded outside the neocortex, in the hippocampus, to provide an accurate measure of state. All sleep scoring was done blinded to stimulation conditions (optogenetic or chemogenetic).

The remaining periods of wakefulness were then divided into movement and quiet wake (QW) states. In head-fixed recordings, movement intervals were defined as periods of 5 s or longer in which facial motion energy (extracted using Facemap, binned at 1 s) exceeded the 90th percentile of the session’s distribution. QW was defined as the remaining wake intervals, with a minimum duration of 5 s. A similar procedure was applied in freely moving experiments, but accelerometer signals were used instead of the facial motion to classify movement and QW states. We noted that head-fixed mice exhibited substantially longer QW periods than did freely moving mice, and we confirm, consistent with previous reports, that head-fixed mice sleep with their eyes open, and exhibit typical LFP signatures (high delta power during SWS and high theta/delta ratio during REM)⁶⁶. Although episodes of QW, REM and SWS in head-fixed mice might differ in meaningful ways from those recorded in freely sleeping mice, we used the same state terminology across head-fixed and freely moving experiments for consistency.

Data analysis

Data were analysed using open-source software packages and custom-written MATLAB (2018b–2023b) and Python code.

Whole-neuron morphology projection matrix

SWC files were merged and resampled so that spacing between nodes was uniform. Merged files were registered to the average mouse brain template of the Allen Mouse Brain CCF (v.3)⁶⁷. CCF-registered reconstructions were translated so that all somas were positioned on the left hemisphere. To generate a projection matrix, we used two strategies: one included all axons present in an area (Fig. 1, Supplementary Table 1 and Extended Data Fig. 1), and the other required that a targeted structure contain at least one branch tip and one node (Supplementary Table 2). All projections were ipsilateral, and targeted structures are reported only for the ipsilateral hemisphere.

Whole-brain fibre density and rabies

Mice were cryosectioned as described above, with uniform sampling throughout the entire brain. Imaged brain sections were aligned to the Allen Mouse Brain CCF using NeuroInfo (MBF Bioscience), with the BrainMaker workflow. Labelled neuronal arbor in aligned tissue was then reconstructed within single sections using Neurolucida 360 (MBF Bioscience). Density measurements for individual areas were made by calculating the total path length of reconstructed arbor within that area divided by the volume of the area (the two-dimensional area within section multiplied by the section thickness). The coordinates of each identified rabies-positive neuron were compiled to yield a complete whole-brain map for each mouse. To account for variability in labelling efficiency between mice, data from each brain were normalized to the total number of rabies-positive cells identified in that mouse. These normalized values were then used for quantification of layer distributions, regional localization and subregional visual cortex analyses.

Ex vivo patch-clamp recordings

Electrophysiology data were analysed using SutterPatch v.2.4 (written in IgorPro, Wavemetrics) and AxoGraph v.1.7.6. oIPSCs were aligned to the light stimulus and 20–60 traces were averaged to calculate oIPSC amplitude and latency. Amplitude was measured as the maximum positive peak of the baseline-subtracted oIPSC, and latency was calculated as the time between the light stimulus and crossing a 2 s.d. threshold of the oIPSC from baseline.

Extraction of calcium activity

Acquired images were processed using Suite2p (ref. ⁶⁸) to extract regions of interest and fluorescence traces. ΔF/F₀ values were computed by normalizing fluorescence to a moving 10th percentile within 10-min windows as the baseline (F₀). Deconvolved calcium signals were obtained from Suite2p, and, owing to their higher temporal resolution, were used to examine the temporal relationships between calcium activity, delta power, state transition and DOWN states. To investigate state transition, calcium activity was evaluated only during states lasting a minimum of 10 s. The latency of calcium activation during SWS was defined as the earliest time point at which the deconvolved calcium signal exceeded 2 s.d. above baseline activity. When measuring calcium activity around DOWN states, calcium traces were corrected by subtracting chance-level calcium activity generated by circularly shifting the calcium signal (circshift in Matlab) 100 times within the 1-s window around the onset of DOWN states.

Arousal score

Arousal score (Fig. 2b) was calculated using the pca function in MATLAB. It corresponds to the time-varying score for the first principal component of the measured state metrics. Principal component loadings were calculated for each individual recording to provide robustness against measurement error for any singular state metric.

Classification of low-arousal- and high-arousal-correlated cells

Cell types were determined on the basis of the correlation of ΔF/F traces with state metrics. k-means clustering (n = 2) was performed on these correlation coefficients. Separately, we calculated the correlation coefficient of the ΔF/F signal with the arousal score. Cells with negative correlation coefficients (low-arousal-correlated cells) overlapped completely with the k-means low-arousal cluster.

High- versus low-Sst-Chodl cell activity epochs

ΔF/F traces were averaged within each recording and then z-scored. High-activity epochs were defined as periods in which activity was more than 2 s.d. from the mean, and low-activity epochs were defined as those in which activity was less than 0.5 s.d.

Single units and LFP extraction

Single units were isolated using Kilosort2 (ref. ⁶⁹). Figures contain data from both broad-spiking and narrow-spiking units, because no differences were found between these groups. LFP signals were collected from the microwire array or from silicon probes and were downsampled to 1,250 Hz. EMG signals were obtained by band-pass-filtering the raw signal (more than 200 Hz) followed by a Hilbert transform.

Calculation of power spectra

Power spectra were calculated using a wavelet-based spectrogram method (Morlet wavelets, five-cycle width). These wavelets were generated for 100 log-spaced frequencies from 1 Hz to 128 Hz. For freely moving optogenetic experiments, changes in power spectra were assessed by comparing the 30-s stimulation periods to the 30-s ITIs. To investigate changes in power spectra across the neocortex as a function of distance from the injection site, we selected, at each distance (proximal to anterior), the channel that showed the largest delta band power increase. This approach was implemented to minimize the confounding effects of asymmetrical projections in which Sst-Chodl fibres might be absent on one side but present on the other at a single distance, given the substantial variation in individual projection patterns (see Fig. 1).

Determination of cortical depth

As shown previously, high-frequency power (more than 500 Hz) peaks in mid-layer 5 (ref. ⁷⁰). We interpolated between this layer 5 channel and the first channel outside of the brain (1.3 mm probe inserted 1.2 mm into the brain), assuming the remaining distance, without the mechanical deformation of the brain caused by insertion of the probe, to be 700 µm. With these interpolated pseudo-depths, we then defined layer boundaries at 100, 300, 450, 650 and 900 µm.

Spiking synchrony

The cross-correlation was calculated for each unit pair using default parameters of the CCG.m function from the Buzcode package. The spike times of these units were shuffled, and the cross-correlation was recalculated 100 times to estimate a chance level. The real cross-correlation was then normalized by this chance level to quantify spiking co-activation.

Phase locking

The Fourier spectrum of the LFP was calculated for each spike from each unit using the ft_spiketriggeredspectrum function from the FieldTrip toolbox. The circular mean of the phase values at each frequency was calculated as the phase locking value. Although this measure is sensitive to spike count, with higher spikes leading to higher phase locking values, we do not see this as an issue, because stimulation leads to slightly lower firing rates compared with baseline.

DOWN state detection

DOWN states were detected using a previously published method²⁷. In brief, the method detects a confluence of a large positive deflection in LFP, a drop in gamma-band power and a sharp drop in the multi-unit firing rate.

Linear regression model

Deconvolved calcium signals were used to predict the z-scored delta amplitude after constructing a time-lagged design matrix spanning a 20-s temporal window, sampled every 0.2 s. The kernel was centred on the present time point, allowing both past and future lags to contribute to the prediction. Ordinary least-squares regression was fitted in a fivefold cross-validation scheme, and model performance was quantified by the cross-validated coefficient of determination (R²). The resulting regression weights were averaged across folds to obtain a mean temporal kernel describing the influence of calcium activity at different lags on the predicted delta signal.

Mouse tracking

Cage position and mouse nest area were delimited manually using a compilation of frames extracted every 10 min from each video. Mouse position was detected from the body centroid of the mouse using the DeepLabCut open-source system⁷¹. The distance moved by the mouse was extracted according to the registered position of the cage. The duration spent in the nest was estimated by calculating the time that the mouse was present in the nest according to the total duration of the experiment.

Allen Brain Observatory data analysis

Sst-expressing cell state correlation data were retrieved using the Allen SDK and querying the database for all experiments performed with Sst^cre and Cre-dependent GCaMP mice. Pearson’s correlations were calculated between activity traces and state metrics after binning both signals into 3-s windows. Data on the proportions of Sst-Chodl cells within the Sst population were retrieved from a previously published dataset⁸ from the web-based Brain Knowledge Platform (https://knowledge.brain-map.org/data).

Analysis of previously published data

Data were retrieved from https://doi.org/10.6084/m9.figshare.19448531 (ref. ²⁹), and analysed from spontaneous recordings without visual stimulation.

The following software packages were used to analyse the data presented in this paper (written in the format package name, URL, publication):

• AllenSDK, https://github.com/AllenInstitute/AllenSDK, N/A

• Buzcode, https://github.com/buzsakilab/buzcode, N/A

• Facemap, https://github.com/MouseLand/facemap, ref. ⁷²

• Suite2p, https://github.com/MouseLand/suite2p, ref. ⁶⁸

• Kilosort2, https://github.com/MouseLand/Kilosort, ref. ⁶⁹

• FieldTrip, https://www.fieldtriptoolbox.org/, ref. ⁷³

• DeepLabCut, https://github.com/DeepLabCut/DeepLabCut, ref. ⁷¹

• Phy v.2.0 beta, https://github.com/cortex-lab/phy, N/A

• SutterPatch v.2.4, https://www.sutter.com/amplifiers/sutterpatch, N/A

• AxoGraph v.1.7.6, https://axograph.com/, N/A

• HortaCloud, https://doi.org/10.1101/2025.03.13.642887, N/A

• MBF Bioscience, https://www.mbfbioscience.com, N/A

• MATLAB v.2018b–2023b, https://www.mathworks.com, N/A

Statistical analysis

MATLAB (MathWorks, v.2018b–2023b) was used for statistical analysis. No power calculations were used to predetermine sample sizes or to formally assess normality. The sample size (mice, sessions, cells and units) is consistent with similar studies in the field^21,22,39 and reflects the technical complexity of the manipulations and recordings, and their yield. Chemogenetic experimental mice were randomly assigned to receive CNO or vehicle injection in the first session. Other experimental conditions were defined by recording site, genetic targeting or stimulation protocols independent of the group assignment. Comparisons were performed using two-tailed parametric t-test or one-way, two-way (anova1) or repeated-measures ANOVA, with post-hoc Bonferroni corrections for multiple comparisons (unless stated otherwise). Differences in proportions were assessed using Chi-squared tests (chi2cdf). For optogenetic experiments (linear probes, multi-LFPs and patch clamp) and comparison of SWS duration between light and dark sessions, a linear mixed-effects model was implemented in MATLAB (fitlme) and the model significance was assessed by ANOVA, which tests for the contribution of each fixed factor and their interaction while accounting for unequal sample sizes. Values and statistical tests used are reported in the text and data are represented as mean ± s.e.m. unless stated otherwise. Significance was set with α = 0.05 and is represented on graphs as *P < 0.05, **P < 0.01 and ***P < 0.001.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

1. Harris, K. D. & Thiele, A. Cortical state and attention. Nat. Rev. Neurosci. 12, 509–523 (2011).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

2. Cardin, J. A. et al. Driving fast-spiking cells induces gamma rhythm and controls sensory responses. Nature 459, 663–667 (2009).

   Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

3. Sohal, V. S., Zhang, F., Yizhar, O. & Deisseroth, K. Parvalbumin neurons and gamma rhythms enhance cortical circuit performance. Nature 459, 698–702 (2009).

   Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

4. Batista-Brito, R., Zagha, E., Ratliff, J. M. & Vinck, M. Modulation of cortical circuits by top-down processing and arousal state in health and disease. Curr. Opin. Neurobiol. 52, 172–181 (2018).

   Article  CAS  PubMed  Google Scholar 

5. Freeman, D., Sheaves, B., Waite, F., Harvey, A. G. & Harrison, P. J. Sleep disturbance and psychiatric disorders. Lancet Psychiatry 7, 628–637 (2020).

   Article  PubMed  Google Scholar 

6. Ramón y Cajal, S. Recollections of My Life (MIT Press, 1937).

7. Tasic, B. et al. Adult mouse cortical cell taxonomy revealed by single cell transcriptomics. Nat. Neurosci. 19, 335–346 (2016).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

8. Gouwens, N. W. et al. Integrated morphoelectric and transcriptomic classification of cortical GABAergic cells. Cell 183, 935–953 (2020).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

9. Yao, Z. et al. A taxonomy of transcriptomic cell types across the isocortex and hippocampal formation. Cell 184, 3222–3241 (2021).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

10. Tomioka, R. et al. Demonstration of long-range GABAergic connections distributed throughout the mouse neocortex. Eur. J. Neurosci. 21, 1587–1600 (2005).

    Article  PubMed  Google Scholar 

11. Tomioka, R. & Rockland, K. S. Long-distance corticocortical GABAergic neurons in the adult monkey white and gray matter. J. Comp. Neurol. 505, 526–538 (2007).

    Article  PubMed  Google Scholar 

12. Woych, J. et al. Cell-type profiling in salamanders identifies innovations in vertebrate forebrain evolution. Science 377, eabp9186 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

13. Dittrich, L., Morairty, S. R., Warrier, D. R. & Kilduff, T. S. Homeostatic sleep pressure is the primary factor for activation of cortical nNOS/NK1 neurons. Neuropsychopharmacology 40, 632–639 (2015).

    Article  CAS  PubMed  Google Scholar 

14. Morairty, S. R. et al. A role for cortical nNOS/NK1 neurons in coupling homeostatic sleep drive to EEG slow wave activity. Proc. Natl Acad. Sci. USA 110, 20272–20277 (2013).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

15. Gerashchenko, D. et al. Identification of a population of sleep-active cerebral cortex neurons. Proc. Natl Acad. Sci. USA 105, 10227–10232 (2008).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

16. Groenhout, T. et al. Tetrodotoxin-mediated inactivation of medial prefrontal cortex decreases wakefulness and rapid eye movement sleep, and increases slow-wave sleep in rat. Sleep 48, zsae260 (2025).

    Article  PubMed  PubMed Central  Google Scholar 

17. Sawada, T. et al. Prefrontal synaptic regulation of homeostatic sleep pressure revealed through synaptic chemogenetics. Science 385, 1459–1465 (2024).

    Article  ADS  CAS  PubMed  Google Scholar 

18. Kon, K. et al. Cortical parvalbumin neurons are responsible for homeostatic sleep rebound through CaMKII activation. Nat. Commun. 15, 6054 (2024).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

19. Hong, J., Lozano, D. E., Beier, K. T., Chung, S. & Weber, F. Prefrontal cortical regulation of REM sleep. Nat. Neurosci. 26, 1820–1832 (2023).

    Article  CAS  PubMed  Google Scholar 

20. Funk, C. M. et al. Role of somatostatin-positive cortical interneurons in the generation of sleep slow waves. J. Neurosci. 37, 1303–1317 (2017).

    Article  Google Scholar 

21. Tossell, K. et al. Somatostatin neurons in prefrontal cortex initiate sleep-preparatory behavior and sleep via the preoptic and lateral hypothalamus. Nat. Neurosci. 26, 1805–1819 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

22. Krone, L. B. et al. A role for the cortex in sleep–wake regulation. Nat. Neurosci. 24, 1210–1215 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

23. Krone, L. B. et al. Contrasting effects of global and local cortical activity on the regulation of sleep. Preprint at bioRxiv https://doi.org/10.1101/2025.09.07.673922 (2025).

24. Wasilczuk, A. Z. et al. Convergent control of NREM sleep and anesthesia by prefrontal layer 5 extratelencephalic neurons. Preprint at bioRxiv https://doi.org/10.1101/2025.10.10.681639 (2025).

25. Asenjo-Martinez, A. et al. Dysfunction of cortical GABAergic projection neurons as a major hallmark in a model of neuropsychiatric syndrome. Neuron 113, 3204–3223 (2025).

    Article  CAS  PubMed  Google Scholar 

26. Perrenoud, Q. et al. Characterization of type I and type II nNOS-expressing interneurons in the barrel cortex of mouse. Front. Neural Circuits 6, 36 (2012).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

27. Levenstein, D., Buzsáki, G. & Rinzel, J. NREM sleep in the rodent neocortex and hippocampus reflects excitable dynamics. Nat. Commun. 10, 2478 (2019).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

28. de Vries, S. E. J. et al. A large-scale standardized physiological survey reveals functional organization of the mouse visual cortex. Nat. Neurosci. 23, 138–151 (2020).

    Article  PubMed  Google Scholar 

29. Bugeon, S. et al. A transcriptomic axis predicts state modulation of cortical interneurons. Nature 607, 330–338 (2022).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

30. Sulaman, B. A., Wang, S., Tyan, J. & Eban-Rothschild, A. Neuro-orchestration of sleep and wakefulness. Nat. Neurosci. 26, 196–212 (2023).

    Article  CAS  PubMed  Google Scholar 

31. Vyazovskiy, V. V. et al. Local sleep in awake rats. Nature 472, 443–447 (2011).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

32. Volgushev, M., Mukovski, C. & Timofeev, I. Precise long-range synchronization of activity and silence in neocortical neurons during slow-wave sleep. J. Neurosci. 26, 5665–5672 (2006).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

33. Lőrincz, M. L. et al. A distinct class of slow (∼0.2–2 Hz) intrinsically bursting layer 5 pyramidal neurons determines UP/DOWN state dynamics in the neocortex. J. Neurosci. 35, 5442–5458 (2015).

    Article  PubMed  PubMed Central  Google Scholar 

34. Beltramo, R. et al. Layer-specific excitatory circuits differentially control recurrent network dynamics in the neocortex. Nat. Neurosci. 16, 227–234 (2013).

    Article  CAS  PubMed  Google Scholar 

35. Chen, N., Sugihara, H. & Sur, M. An acetylcholine-activated microcircuit drives temporal dynamics of cortical activity. Nat. Neurosci. 18, 892–902 (2015).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

36. Veit, J., Hakim, R., Jadi, M. P., Sejnowski, T. J. & Adesnik, H. Cortical gamma band synchronization through somatostatin interneurons. Nat. Neurosci. 20, 951–959 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

37. Chen, G. et al. Distinct inhibitory circuits orchestrate cortical beta and gamma band oscillations. Neuron 96, 1403–1418 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

38. Huang, P., Xiang, X., Chen, X. & Li, H. Somatostatin neurons govern theta oscillations induced by salient visual signals. Cell Rep. 33, 108415 (2020).

    Article  CAS  PubMed  Google Scholar 

39. Ruff, C. F. et al. Long-range inhibitory neurons mediate cortical neurovascular coupling. Cell Rep. 43, 113970 (2024).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

40. Echagarruga, C. T., Gheres, K. W., Norwood, J. N. & Drew, P. J. nNOS-expressing interneurons control basal and behaviorally evoked arterial dilation in somatosensory cortex of mice. eLife 9, e60533 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

41. Paul, A. et al. Transcriptional architecture of synaptic communication delineates GABAergic neuron identity. Cell 171, 522–539 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

42. Zielinski, M. R. et al. Somatostatin⁺/nNOS⁺ neurons are involved in delta electroencephalogram activity and corticaldependent recognition memory. Sleep 42, zsz143 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

43. Vruwink, M., Schmidt, H. H., Weinberg, R. J. & Burette, A. Substance P and nitric oxide signaling in cerebral cortex: anatomical evidence for reciprocal signaling between two classes of interneurons. J. Comp. Neurol. 441, 288–301 (2001).

    Article  CAS  PubMed  Google Scholar 

44. Xie, L. et al. Sleep drives metabolite clearance from the adult brain. Science 342, 373–377 (2013).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

45. Niethard, N., Ngo, H.-V. V., Ehrlich, I. & Born, J. Cortical circuit activity underlying sleep slow oscillations and spindles. Proc. Natl Acad. Sci. USA 115, E9220–E9229 (2018).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

46. Zucca, S. et al. An inhibitory gate for state transition in cortex. eLife 6, e26177 (2017).

    Article  PubMed  PubMed Central  Google Scholar 

47. Valero, M. et al. Sleep down state-active ID2/Nkx2.1 interneurons in the neocortex. Nat. Neurosci. 24, 401–411 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

48. Williams, R. H. et al. Cortical nNOS/NK1 receptor neurons are regulated by cholinergic projections from the basal forebrain. Cereb. Cortex 28, 1959–1979 (2017).

    Article  Google Scholar 

49. Bocchio, M. et al. Sleep and serotonin modulate paracapsular nitric oxide synthase expressing neurons of the amygdala. eNeuro 3, ENEURO.0177-16.2016 (2016).

    Article  PubMed  PubMed Central  Google Scholar 

50. Dittrich, L. et al. Cortical nNOS neurons co-express the NK1 receptor and are depolarized by substance P in multiple mammalian species. Front. Neural Circuits 6, 31 (2012).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

51. Juarez Anaya, F., Kim, J., Ross, S. E. & Vazquez, A. L. Substance P regulates Tacr1 neurons, which control nitric oxide-mediated neurovascular coupling in the mouse cortex. Sci. Adv. 12, eadx5109 https://doi.org/10.1126/sciadv.adx5109 (2026).

52. Ben-Simon, Y. et al. A suite of enhancer AAVs and transgenic mouse lines for genetic access to cortical cell types. Cell 188, 3045–3064 (2025).

    Article  CAS  PubMed  Google Scholar 

53. Batista-Brito, R. et al. Developmental dysfunction of VIP interneurons impairs cortical circuits. Neuron 95, 884–895 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

54. Goldey, G. J. et al. Removable cranial windows for long-term imaging in awake mice. Nat. Protoc. 9, 2515–2538 (2014).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

55. Luan, L. et al. Ultraflexible nanoelectronic probes form reliable, glial scar-free neural integration. Sci. Adv. 3, e1601966 (2017).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

56. Osborne, J. E. & Dudman, J. T. RIVETS: a mechanical system for in vivo and in vitro electrophysiology and imaging. PLoS ONE 9, e89007 (2014).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

57. Furlanis, E. et al. An enhancer-AAV toolbox to target and manipulate distinct interneuron subtypes. Neuron 113, 1525–1547 (2025).

    Article  CAS  PubMed  Google Scholar 

58. Rokicki, K. et al. HortaCloud: an open and collaborative platform for whole brain neuronal reconstructions. Preprint at bioRxiv https://doi.org/10.1101/2025.03.13.642887 (2025).

59. Peng, H. et al. Morphological diversity of single neurons in molecularly defined cell types. Nature 598, 174–181 (2021).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

60. Wang, X. et al. Chemical sectioning fluorescence tomography: high-throughput, high-contrast, multicolor, whole-brain imaging at subcellular resolution. Cell Rep. 34, 108709 (2021).

    Article  CAS  PubMed  Google Scholar 

61. Stringer, C. et al. Spontaneous behaviors drive multidimensional, brainwide activity. Science 364, eaav7893 (2019).

    Article  CAS  Google Scholar 

62. Varin, C., Luppi, P.-H. & Fort, P. Melanin-concentrating hormone-expressing neurons adjust slow-wave sleep dynamics to catalyze paradoxical (REM) sleep. Sleep 41, zsy068 (2018).

    Article  Google Scholar 

63. Traut, J. et al. Effects of clozapine-N-oxide and compound 21 on sleep in laboratory mice. eLife 12, e84740 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

64. Grosmark, A. D., Mizuseki, K., Pastalkova, E., Diba, K. & Buzsáki, G. REM sleep reorganizes hippocampal excitability. Neuron 75, 1001–1007 (2012).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

65. Watson, B. O., Levenstein, D., Greene, J. P., Gelinas, J. N. & Buzsáki, G. Network homeostasis and state dynamics of neocortical sleep. Neuron 90, 839–852 (2016).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

66. Yüzgeç, Ö., Prsa, M., Zimmermann, R. & Huber, D. Pupil size coupling to cortical states protects the stability of deep sleep via parasympathetic modulation. Curr. Biol. 28, 392–400 (2018).

    Article  PubMed  PubMed Central  Google Scholar 

67. Sorensen, S. A. et al. Connecting single-cell transcriptomes to projectomes in the mouse visual cortex. Nature https://doi.org/10.1038/s41586-026-10424-8 (2026).

    Article  PubMed  PubMed Central  Google Scholar 

68. Pachitariu, M. et al. Suite2p: beyond 10,000 neurons with standard two-photon microscopy. Preprint at bioRxiv https://doi.org/10.1101/061507 (2017).

69. Pachitariu, M., Sridhar, S. & Stringer, C. Solving the spike sorting problem with Kilosort. Preprint at bioRxiv https://doi.org/10.1101/2023.01.07.523036 (2023).

70. Senzai, Y., Fernandez-Ruiz, A. & Buzsáki, G. Layer-specific physiological features and interlaminar interactions in the primary visual cortex of the mouse. Neuron 101, 500–513 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

71. Mathis, A. et al. DeepLabCut: markerless pose estimation of user-defined body parts with deep learning. Nat. Neurosci. 21, 1281–1289 (2018).

    Article  CAS  PubMed  Google Scholar 

72. Syeda, A. et al. Facemap: a framework for modeling neural activity based on orofacial tracking. Nat. Neurosci. 27, 187–195 (2024).

    Article  CAS  PubMed  Google Scholar 

73. Oostenveld, R., Fries, P., Maris, E. & Schoffelen, J.-M. FieldTrip: open source software for advanced analysis of MEG, EEG, and invasive electrophysiological data. Comput. Intell. Neurosci. 2011, 156869 (2011).

    Article  PubMed  Google Scholar 

Download references

Author notes

1. Jacob M. Ratliff

   Present address: UCL Queen Square Institute of Neurology, University College London, London, UK

2. Geoffrey Terral, Gianna Mattessich & Renata Batista-Brito

   Present address: Nash Family Department of Neuroscience, Icahn School of Medicine at Mount Sinai, New York, NY, USA

3. These authors contributed equally: Jacob M. Ratliff, Geoffrey Terral

Authors and Affiliations

1. Dominick P. Purpura Department of Neuroscience, Albert Einstein College of Medicine, New York, NY, USA

   Jacob M. Ratliff, Geoffrey Terral, Arenski Vazquez, Stefano Lutzu, Nelson Perez-Catalan, Gabriela Neubert da Silva, Soyoun Kim, Julie Mota, Bianca Stith, Gianna Mattessich, Lucas Sjulson, Stephanie Rudolph & Renata Batista-Brito

2. Allen Institute for Brain Science, Seattle, WA, USA

   Arena Manning, Matt Mallory, Tanya Daigle, Hongkui Zeng, Bosiljka Tasic & Staci Sorensen

3. Department of Bioengineering, Stanford University, Stanford, CA, USA

   Charu Ramakrishnan & Karl Deisseroth

4. Department of Neuroscience, University of Texas at Austin, Austin, TX, USA

   Lief E. Fenno

5. Department of Psychiatry and Behavioral Sciences, University of Texas at Austin, Austin, TX, USA

   Lief E. Fenno

6. Department of Molecular and Cellular Biology, University of California, Berkeley, Berkeley, CA, USA

   David A. Stafford

7. Department of Psychiatry and Behavioral Sciences, Stanford University, Stanford, CA, USA

   Karl Deisseroth

8. Howard Hughes Medical Institute, Stanford University, Stanford, CA, USA

   Karl Deisseroth

9. National Institute of Neurological Disorders and Stroke, National Institutes of Health, Bethesda, MD, USA

   John Ngai

10. Biosciences Division, SRI International, Menlo Park, CA, USA

    Thomas S. Kilduff

11. Department of Psychiatry and Behavioral Sciences, Albert Einstein College of Medicine, New York, NY, USA

    Lucas Sjulson, Stephanie Rudolph & Renata Batista-Brito

12. Department of Genetics, Albert Einstein College of Medicine, New York, NY, USA

    Renata Batista-Brito

Contributions

Conceptualization: J.M.R., G.T. and R.B.-B. In vivo imaging: J.M.R. and G.T. Single-cell reconstruction and analysis: A.M., M.M., B.S., S.S. and J.M.R. Bulk morphological reconstructions: J.M.R., J.M. and B.S. Ex vivo patch clamp: A.V., S.L. and S.R. Head-fixed optogenetics of Sst-Chodl cells: J.M.R. and G.T. Head-fixed optogenetics of Martinotti cells: A.V., S.K., J.M.R. and L.S. Freely moving chemogenetics: G.T. and J.M.R. Rabies staining and quantification: N.P.-C. and G.M. Surgery and brain processing: J.M.R., G.T., A.V., G.N.d.S. and J.M. Reagents: C.R., L.E.F., B.T., T.D., D.A.S., H.Z., K.D. and J.N. Scientific advising: T.S.K. Supervision: R.B.-B. Initial draft: J.M.R., G.T. and R.B.-B. All authors edited and approved the manuscript.

Corresponding authors

Correspondence to Geoffrey Terral or Renata Batista-Brito.

Competing interests

The authors declare no competing interests.

Peer review information

Nature thanks Antonio Fernandez-Ruiz, William Wisden and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.

Reporting Summary (download PDF )

Peer Review File (download PDF )

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions