Off-target effects of DREADD ligands revealed by an anesthesia emergence paradigm in mice

Summary

Designer receptors exclusively activated by designer drugs (DREADDs) enable reversible control of specific neural circuits, but the pharmacological neutrality of their ligands is increasingly questioned. Here, we introduce an anesthesia emergence paradigm to systematically assess the off-target effects of DREADD ligands in DREADD-naive mice. We show that intraperitoneal administration of clozapine N-oxide (CNO), compound 21 (C21), or deschloroclozapine (DCZ) delays motor recovery from isoflurane anesthesia. CNO produced the largest delay, likely due to its back-conversion to clozapine. DCZ showed the smallest effect magnitude, although its difference from C21 remained inconclusive. We then show that subcutaneous administration, which should reduce clozapine back-conversion, reduces the CNO-induced recovery delay to levels comparable to those of C21. Finally, we provide a freely available, deep-learning-based automated behavioral pipeline that integrates the anesthesia emergence paradigm with a reproducible analysis tool for future studies. Together, these results underscore the importance of accounting for ligand off-target effects through careful dose selection and DREADD-free, ligand-treated controls in chemogenetic experiments.

Graphical abstract

Highlights

• •DREADD ligands (CNO, C21, and DCZ) delay anesthesia emergence in mice
• •CNO causes the largest recovery delay
• •DCZ shows the smallest effect magnitude
• •Subcutaneous administration mitigates off-target effects of CNO to levels comparable to C21

Motivation

Chemogenetic tools, such as DREADDs, have profoundly influenced neuroscience. Their use rests on the critical assumption that DREADD ligands are pharmacologically inert. However, emerging evidence of off-target effects has challenged this premise, potentially confounding experimental interpretations. Standard behavioral assays are often labor intensive or lack sensitivity to subtle changes. Here, we develop and validate an anesthesia emergence paradigm that provides a simple, low-cost, and highly sensitive assay for assessing the off-target effects of chemogenetic ligands.

Moreno-Gomez et al. demonstrate that CNO, C21, and DCZ delay anesthesia emergence in mice. They show that subcutaneous administration mitigates CNO effects and provide an open-source deep learning pipeline to rigorously screen chemogenetic tools. Overall, they demonstrate that rigorous chemogenetic research requires mitigating off-target actions via careful dosing and DREADD-free controls.

Introduction

Designer receptors exclusively activated by designer drugs (DREADDs) are a powerful, widely used chemogenetic tool^1,2 that enables remote, reversible, and cell-type-specific manipulation of neuronal activity to establish causal links between neural circuits and behavior.^3,4,5,6 These engineered G protein-coupled receptors (GPCRs)^7,8,9 are designed to be unresponsive to endogenous ligands but are potently activated by a synthetic and pharmacologically inert ligand.^4,10 The validity of this premise of ligand inertness is important for correct experimental interpretation, as any pharmacological effect of the DREADD ligand itself could confound the conclusions about circuit-level manipulations.^11,12

The history of DREADDs has been a continuous search for truly inert ligands. The first-generation ligand, clozapine N-oxide (CNO), was found to undergo in vivo back-conversion to the atypical antipsychotic clozapine, a compound with a broad receptor-binding profile.^11,13,14,15 Consequently, CNO administration in DREADD-naive rodents produces a range of off-target effects, including changes in glutamate levels,¹⁶ alterations in sleep,¹⁷ locomotion,¹⁸ breathing dynamics,¹⁹ and even somatosensory processing and anxiety-like behaviors.²⁰ This spurred the development of new ligands.^{21,22,23,24,25} Yet, subsequent ligands like compound 21 (C21), while not metabolized to clozapine,^22,24 were also found to possess off-target activity, likely through direct engagement with endogenous receptors such as the histamine H1 receptor, affecting processes including sleep architecture,^17,26 neuronal firing rates,²⁷ and diuresis.²⁸ More recently, deschloroclozapine (DCZ) was introduced as a superior alternative, engineered for high potency and exquisite selectivity. Structural biology studies confirmed that DCZ binds to muscarinic DREADDs with nanomolar affinity, orders of magnitude higher than its affinity for any endogenous receptor.^23,29 Initial studies in wild-type rodents and non-human primates reported no off-target effects at standard, effective doses (0.1 mg kg⁻¹) on locomotor or cognitive tasks.^23,30 Evidence from non-human primates and rats, however, suggests that this inertness may be lost at higher concentrations: a higher dose of DCZ (0.2–0.3 mg kg⁻¹)—but not the standard dose (0.1 mg kg⁻¹)—produces off-target effects on brain functional connectivity and cognitive performance.^31,32 As masked off-target effects may exist even in well-controlled studies,¹⁹ a sufficiently sensitive behavioral paradigm could reveal subtle off-target effects that are missed by less demanding assays.

In this study, we introduce an anesthesia emergence paradigm to systematically assess the off-target effects of DREADD ligands by measuring time to motor recovery from isoflurane anesthesia in DREADD-naive mice. Emergence from anesthesia is a stereotyped yet active neurobiological process that requires coordinated reactivation of ascending and descending arousal systems to overcome the anesthetic-induced state,³³ offering a sensitive and finely graded quantitative readout of the underlying neurophysiological state.³⁴ First, we systematically quantified the effects of three common DREADD ligands (CNO, C21, and DCZ) on the behavioral dynamics of emergence from anesthesia, administering the ligands intraperitoneally (i.p.) as in the vast majority of previous off-target characterization studies.^{14,18,20,26,27,30,35} Second, we investigated whether subcutaneous (s.c.) administration could mitigate the off-target effects of CNO by reducing its back-conversion to clozapine in the liver.^14,36,37 Finally, we developed and validated a high-throughput, automated behavioral analysis pipeline with deep-learning-based tracking to integrate the anesthesia emergence paradigm with a rigorous and reproducible analysis tool for future studies.

Results

Intraperitoneal DREADD ligands produce robust off-target effects on anesthesia emergence

We first aimed to determine the acute effects of intraperitoneal administration of the DREADD ligands CNO, C21, and DCZ on the behavioral dynamics of anesthesia induction and emergence by means of manual, real-time evaluation. To this end, we scored several motor events during the induction of and emergence from anesthesia (Table 1).

Table 1.

Measures of motor events during anesthesia induction and emergence phases

Induction phase measures
Onset of macroscopic motor effects of anesthesia	the time of the first discernible alteration in voluntary motor activity
Cessation of whisker movement	the time of the last discernible movement of the whiskers
Complete cessation of movement	the time of the last discernible voluntary movement of the torso or head
Onset of rhythmic breathing	the time of the initiation of a slower regular respiratory pattern
Emergence phase measures
Onset of whisker movement	the time of the first discernible reappearance of whisker movement
Stability on hindlimbs	the first time when the animal places its hindlimbs on the floor
Stability on forelimbs	the first time when the animal supports its full body weight on its forelimbs
Onset of head movement	the time of the first discernible voluntary movement of the head
Onset of forelimb movement	the time of the first coordinated locomotion involving forelimbs only
Onset of walking	the time of the first coordinated forward locomotion involving all four limbs

Each measurement parameter (left column) is conveniently defined (right column) to allow for reproducibility and repeatability of the anesthesia paradigm.

The overall analysis of the anesthesia induction events revealed moderate evidence that the DREADD ligands did not alter the dynamics of entering anesthesia (overall three-way ANOVA, real/control effect: BF_incl = 0.202; Figure 1A; Table S1; Figure S1).

Figure 1.

Summary of DREADD ligand off-target effects

(A–D) Temporal evolution of all manually recorded motor events for each DREADD ligand and its respective control, shown in four graphs: (A) intraperitoneal administration of all DREADD ligands with their control; (B) subcutaneous one, (C) intraperitoneal and subcutaneous administration of CNO, and (D) intraperitoneal and subcutaneous administration of C21. Graphs show the mean time of occurrence (min) for each event; error bars represent 95% credibility intervals. Meaning of the acronyms under the x axis: O.m.m.e. for “onset of motor macroscopic motor effects,” C.c.m. for “complete cessation of movement,” C.w.m. for “cessation of whisker movement,” O.r.b. for “onset of rhythmic breathing,” O.w.m. for “onset of whisker movement,” S.h. for “stability on hindlimbs,” O.h.m. for “onset of head movement,” O.f.m. for “onset of forelimb movement,” and O.w. for “onset of walking.”

Conversely, the overall analysis of the anesthesia emergence events revealed extreme evidence that the DREADD ligands delayed recovery in a ligand-specific manner (overall three-way ANOVA, real/control effect: BF_incl > 10³; real/control × DREADD_ligand interaction: BF_incl > 10³; Figure 1A; Table S1). Furthermore, we found extreme evidence that the recovery delay was more pronounced for later motor events (overall three-way ANOVA, real/control × measure interaction: BF_incl = 167.6).

When we examined each ligand individually, we found robust evidence of delay for all three DREADD ligands: CNO, C21, and DCZ (follow-up two-way ANOVA, real/control effect: CNO BF_incl > 10³; C21 BF_incl > 10³; DCZ BF_incl > 10³). Further analysis of the delta values (i.e., real minus control) provided extreme evidence for a difference between ligands (follow-up two-way ANOVA on deltas, DREADD_ligand effect: BF_incl > 10³). Specifically, the off-target effects of CNO were larger, with strong-to-extreme evidence, compared to both C21 (post hoc t test, BF_10,U = 16.3) and DCZ (BF_10,U > 10³). The magnitude of the delay was smallest for DCZ, but the evidence for a difference between C21 and DCZ remained inconclusive (BF_10,U = 1.1). When excluding CNO from the analysis, the evidence for lower off-target effects of DCZ compared to C21 remained anecdotal (follow-up two-way ANOVA, main effect of DREADD_ligand on deltas: BF_incl = 2.3). Analyses of individual motor events are presented in Figure 2.

Figure 2.

Manual behavioral measurements during emergence from anesthesia following intraperitoneal administration

(A, C, E, G, I, and K) Survival curves showing the cumulative percentage of animals reaching each motor event over time (min after isoflurane off) for the ligands CNO, C21, DCZ, and their pooled control. Bayesian t tests provided moderate evidence against any difference between the controls for each DREADD ligand, justifying their combination into a single control group for visualization.

(B, D, F, H, J, and L) Distribution of the raw timing values for each real DREADD ligand and control. Paired dot plots (left) show the statistical evidence for this effect shown directly above each pair. Symbols above boxplots (right) indicate the statistical evidence from the comparison of delta values (time on real ligand − time on control) between the different DREADD ligands (^oBF₋₀ <1/3, ∗BF₋₀ > 3, ^∗∗BF₋₀ > 10, ^∗∗∗BF₋₀ > 30, ^∗∗∗∗BF₋₀ > 100).

These results establish that i.p. administered DREADD ligands robustly and differentially delay anesthesia emergence, identifying "onset of walking" as a sensitive motor event for quantifying this effect.

S.c. administration route mitigates the off-target effects of CNO but not C21

The off-target effects observed with intraperitoneal administration prompted a second experiment to determine whether s.c. administration would mitigate the off-target effects of CNO. The DREADD ligand C21 was also included to assess whether the route of administration differentially affected the two ligands.

Replicating the findings from the intraperitoneal administration, the overall analysis of the induction events revealed that s.c. administration of the ligands had no discernible impact on the induction of anesthesia, with anecdotal evidence for the absence of an effect (overall three-way ANOVA, real/control effect: BF_incl = 0.430; Figure 1B; Table S1; Figure S2).

The overall analysis of the anesthesia emergence events also replicated the presence of off-target effects. Specifically, s.c. administration of DREADD ligands still prolonged the recovery period with extreme evidence (overall three-way ANOVA, real/control effect: BF_incl > 10³; Figure 1B). However, this time there was anecdotal evidence for the absence of a difference between ligands (overall three-way ANOVA, real/control × DREADD_ligand interaction: BF_incl = 0.429). A closer examination of the individual DREADD ligands revealed extreme evidence of an effect for C21 and almost moderate evidence for CNO (follow-up two-way ANOVA, real/control effect: C21 BF_incl = 160.0; CNO BF_incl = 2.9; Table S1). Analyses of individual motor events are presented in Figure 3.

Figure 3.

Manual behavioral measurements during emergence from anesthesia following subcutaneous administration

(A, C, E, G, I, and K) Survival curves showing the cumulative percentage of animals reaching each motor event over time (min after isoflurane off) for the ligands CNO, C21, and their pooled control. Bayesian t tests provided moderate evidence against any difference between the controls for each DREADD ligand, justifying their combination into a single control group for visualization.

(B, D, F, H, J, and L) Distribution of the raw timing values for each real DREADD ligand and control. Paired dot plots (left) show the statistical evidence for this effect shown directly above each pair. Symbols above boxplots (right) indicate the statistical evidence from the comparison of delta values (time on real ligand − time on control) between the different DREADD ligands (^oBF₋₀ <1/3, ^∗BF₋₀ > 3, ^∗∗BF₋₀ > 10).

A direct comparison between intraperitoneal and s.c. administration confirmed with very strong evidence that the delayed recovery induced by CNO was attenuated by s.c. administration, while the delayed recovery induced by C21 remained unchanged with strong evidence (secondary two-way ANOVA, i.p./s.c. × measure interaction: CNO BF_incl > 10³; C21 BF_incl = 0.065; Figures 1C and 1D). These findings indicate that s.c. administration selectively mitigates the delayed motor recovery induced by CNO but not by C21.

Automated analysis replicates manual analysis

We established that i.p. administered DREADD ligands (CNO, C21, and DCZ) all cause off-target effects and that these effects are mitigated by s.c. administration in the case of CNO. We next validated our automated tool by testing its ability to replicate these results. This tool employs deep-learning-based markerless video tracking to quantify “walking onset,” the most sensitive event identified manually.

The automated measure was an excellent proxy for its manual counterpart, as the two measures were almost perfectly positively correlated for both the intraperitoneal (r = 0.996, BF₊₀ > 10³; Figure 4A) and s.c. (r = 0.997, BF₊₀ > 10³; Figure 4D) datasets. Analyses based on the automated measure of walking onset reproduced all key findings obtained with manual scoring (Figures 2K, 2L, 3K, and 3L) for both intraperitoneal and s.c. administrations (Figures 4B, 4C, 4E, and 4F; Table 2).

Figure 4.

Automated analysis of "walking onset" following intraperitoneal and subcutaneous administration

(A and B) Correlation between the automatic walking onset and its manual counterpart. (B and E) Survival curves showing the cumulative percentage of animals walking over time (min after isoflurane off) for the real DREADD ligands and their pooled control. Bayesian t tests provided evidence against any difference between the controls for each DREADD ligand, justifying their combination into a single control group for visualization.

(C–F) Distribution of the raw timing values for each real DREADD ligand and control. Paired dot plots (left) show the statistical evidence for this effect shown directly above each pair. Symbols above boxplots (right) indicate the statistical evidence from the comparison of delta values (time on real ligand − time on control) between the different DREADD ligands (^ooBF₋₀ < 1/10, ^∗BF₋₀ > 3, ^∗∗BF₋₀ > 10, ^∗∗∗BF₋₀ > 30, ^∗∗∗∗BF₋₀ > 100).

Table 2.

Automatic vs. manual results

Experiment	Statistical analysis	Effect	Measure	dF	F	BFincl
i.p. administered	two-way ANOVA	real/control	automatic walking onset	1, 94.63	35.835	>103
			manual walking onset	1, 94.62	39.287	>103
	two-way ANOVA	real/control ∗ DREADD_ligand interaction	automatic walking onset	2, 94.63	5.332	4.0
			manual walking onset	2, 94.62	5.236	6.5
s.c. administered	two-way ANOVA	real/control	automatic walking onset	1, 56.16	9.566	32.203
			manual walking onset	1, 56.15	9.930	22.454
	two-way ANOVA	real/control ∗ DREADD_ligand interaction	automatic walking onset	1, 56.16	0.534	2.088
			manual walking onset	1, 56.15	0.660	1.607

Experiment	Bayesian t test	Measure	95% CI	$\tilde{\mathrm{x}}$	BF+0
i.p. administered	CNO vs. control	automatic walking onset	[0.412, 1.544]	0.964	289
		manual walking onset	[0.768, 2.129]	1.431	>103
	C21 vs. control	automatic walking onset	[0.323, 1.334]	0.817	124
		manual walking onset	[0.547, 1.682]	1.103	>103
	DCZ vs. control	automatic walking onset	[0.058, 0.849]	0.413	2.700
		manual walking onset	[0.073, 0.892]	0.449	3.712
	ΔCNO vs. ΔC21	automatic walking onset	[0.102, 1.008]	0.525	6.049
		manual walking onset	[0.204, 1.204]	0.687	24.633
	ΔCNO vs. ΔDCZ	automatic walking onset	[0.066, 0.912]	0.447	3.130
		manual walking onset	[0.141, 1.090]	0.593	10.792
	ΔC21 vs. ΔDCZ	automatic walking onset	[0.011, 0.555]	0.193	0.396
		manual walking onset	[0.016, 0.623]	0.592	0.238
s.c. administered	C21 vs. control	automatic walking onset	[0.204, 1.203]	0.686	24.377
		manual walking onset	[0.128, 1.036]	0.558	10.337
	CNO vs. control	automatic walking onset	[0.041, 0.800]	0.366	2.088
		manual walking onset	[0.039, 0.793]	0.360	1.607
	ΔCNO vs. ΔC21	automatic walking onset	[0.003, 0.313]	0.073	0.102
		manual walking onset	[0.004, 0.373]	0.093	0.150
i.p. vs. s.c. administered	CNO	automatic walking onset	[0.077, 1.206]	0.572	3.552
		manual walking onset	[0.086, 1.235]	0.596	4.112
	C21	automatic walking onset	[0.004, 0.435]	0.104	0.147
		manual walking onset	[0.004, 0.446]	0.107	0.154

Top: recompilation of the corresponding Bayes factor (BF_incl) associated with each mean comparison of the two-way ANOVA reported in this study. The F-statistics (F) and degrees of freedom (dF) shown were computed by re-analyzing all two-way ANOVAs through a classical linear mixed-model framework. Bottom: recompilation of the corresponding credibility interval (95% CI), effect size ($\tilde{\mathrm{x}}$), and Bayes factor (BF₊₀ for first > second) associated with each post hoc comparison reported in this study.

Together, these automated data validate the automated pipeline as a robust tool for quantifying off-target effects of DREADD ligands across different administration routes using the anesthesia emergence paradigm (Figure 5).

Figure 5.

Anesthesia paradigm and identification of walking onset

(A and B) Anesthesia paradigm protocol and (B) representative video frame showing manually defined labels required for training the DeepLabCut deep neural network.⁵⁶

(C and D) The main images illustrate two representative examples of the instantaneous speed profiles over time for the head center (red), body center (blue), and tail base (green). Dashed lines indicate the dynamic movement threshold of each body part (mean + 8 × SD of baseline sleep from 400 to 600 s). Discrete periods of motor activity (bout) that surpass detection thresholds are highlighted in gray. Inset images display the spatial trajectory (continuous lines) of the bouts exhibiting the maximum net displacement (dashed lines). Key kinetic metrics (duration, net displacement, and path length) are indicated for mean trajectories (black). (C) Example of bout refinement. Immediately after movement threshold, the only initial bout (980.68–1,313.96 s) containing multiple consecutive movements is automatically split into separate bouts when the speed drops below 5% of the bout’s mean peak speed (48.82 px/s) for at least 3s. The right inset (1,283.73 s) represents the locomotion bout, as it is the only candidate to surpass the minimum net displacement threshold of 50 pixels, thereby satisfying the full four-step filtering process. The start time of the locomotion bout is assigned as the automatic walking onset. (D) Example of bout selection. In this session, all three inset images represent candidate bouts, as they all surpass the net displacement threshold. The final automatic walking onset is assigned to the start time of the locomotion bout that maximizes the product of net displacement × path length (i.e., the bout of the right inset image has the maximum $netdisplacement\times pathlength$, so its start time is assigned to the walking onset event: 1,416.92 s). Scale bars are shown in each inset.

Discussion

We assessed the off-target effects of three widely used DREADD ligands—CNO, C21, and DCZ—in DREADD-naive mice by employing a sensitive anesthesia emergence paradigm. Our principal findings are 3-fold. First, intraperitoneal administration of CNO, C21, and, most notably, the highly potent and selective DCZ²⁹ administered at standard effective doses induces sedative effects, delaying motor recovery from anesthesia. The magnitude of the effects was smallest for DCZ, but the evidence for differences between C21 and DCZ remained inconclusive. Second, switching from intraperitoneal to s.c. administration moderately mitigates the delay in motor recovery from anesthesia induced by CNO, rendering it comparable to the delay induced by C21. Third, we offer a high-throughput, automated behavioral analysis pipeline that integrates the anesthesia emergence paradigm with a rigorous and reproducible analysis tool. Collectively, these results provide new insight into the off-target effects of existing chemogenetic tools, offer a practical strategy to mitigate some of these issues, and deliver accessible methods to improve the rigor and reproducibility of future chemogenetic research.

Our first finding is that intraperitoneal administration of CNO (5 mg/kg), C21 (3 mg/kg), and DCZ (0.1 mg/kg) induced sedative off-target effects in mice. This result is consistent with the range of off-target effects previously reported for CNO and C21^{17,18,20,26,27,28,38} but is less expected for DCZ at the dose used here, which produced an absence of detectable off-target effects in previous studies.^30,31 However, the absence of evidence cannot be interpreted as evidence of absence across all contexts. While the affinity of DCZ for DREADDs is nanomolar, the original characterization reported weak but non-negligible binding affinities (K_i) for endogenous receptors, including the serotonin 5-HT2A (K_i = 87 nM) and muscarinic hM1 (K_i = 83 nM) and hM5 (K_i = 55 nM) receptors.²³ These affinities may become relevant in sensitive behavioral paradigms such as our emergence from anesthesia, which is not a passive decay of drug concentration but a complex neurobiological process involving the reactivation of specific ascending/descending arousal systems to overcome what is known as neural inertia.^33,39,40,41 The systems responsible for emergence—including dopaminergic, histaminergic, and noradrenergic pathways^{42,43,44,45,46}—are precisely the systems that CNO, C21, and DCZ are known or suspected to modulate,^13,47,48 providing a possible mechanistic link between their off-target pharmacology and the observed delay in motor recovery. Importantly, the off-target effects of DREADD ligands by no means invalidate their utility, but they do highlight the need to include proper controls—administration of the DREADD ligand to DREADD-free animals—in chemogenetic studies.¹²

We further show that the sedative off-target effects of CNO were mitigated by altering the route of administration from intraperitoneal to s.c. Intraperitoneal administration typically produces a rapid, high peak plasma concentration with substantial hepatic exposure,^36,49 a critical factor because CNO-clozapine metabolism occurs in the liver.^14,37 This leads to a high degree of conversion to clozapine, which, unlike CNO, can readily cross the blood-brain barrier and exert its centrally active pharmacological effects.^11,50,51 Conversely, s.c. administration results in slower, more sustained absorption and, more importantly, a lower degree of hepatic exposure and metabolism.^36,52 This lower exposure of the liver could explain the smaller off-target effects of CNO with s.c. administration. In contrast, the off-target effects of C21 were not sensitive to the route of administration. The off-target effects of C21 are not mediated by clozapine, suggesting a direct interaction with endogenous receptors,^17,27 which aligns with the known affinity of C21 for the histamine H1 receptor and the critical role of the histaminergic system in promoting emergence from anesthesia.^13,17,24 These results highlight the importance of pharmacokinetic context—such as route of administration—when interpreting chemogenetic experiments and assessing off-target effects.

Finally, we established and validated a robust and accessible methodological framework. We first introduced the anesthesia emergence paradigm as a simple, low-cost, and highly sensitive assay for assessing the off-target effects of chemogenetic ligands. Unlike complex cognitive tasks or chronic sleep studies that require specialized equipment, surgical procedures, and extensive animal handling, this protocol can be implemented in most standard animal facilities with minimal setup.^34,53,54,55 A key limitation of the emergence from anesthesia assay is its inability to tie any evidence for ligand effects to a specific pharmacological system. Yet, its power to unmask off-target effects in ligands previously deemed inert confirms its value as a validation pipeline for both existing and future compounds. Furthermore, we paired this paradigm with the development and validation of an automated behavioral analysis tool—which is made freely available—using the open-source, deep-learning-based software DeepLabCut.⁵⁶ Automated, machine-learning-based analysis can complement or even replace manual scoring, promoting objectivity, standardization, and reproducibility in behavioral neuroscience.⁵⁷ The high correlation between our automated and manual measures (r > 0.99) and the successful replication of all key manual findings validate this pipeline. Importantly, however, we focused the automatization only on the last behavioral event in the emergence from anesthesia, i.e., walking onset, which was highly sensitive and clearly measurable from the video recordings. One-measure automated scoring requires larger sample sizes than multiple-measure manual scoring to achieve the same level of evidence. This can be optimized with a flexible, Bayesian approach, in which an initial cohort is analyzed (e.g., n = 20) and new subjects are incrementally added until the target level of evidence for the alternative or the null hypothesis is reached.⁵⁸ By lowering the barrier to entry for robust off-target screening, this framework makes DREADD ligand validation accessible, efficient, and objective.

In conclusion, this study provides compelling evidence that three widely used DREADD ligands—CNO, C21, and DCZ—cause measurable off-target effects at standard effective doses in mice. We further show that s.c. administration of CNO is a viable and effective strategy for mitigating its off-target effects. Finally, we offer the neuroscience community a validated, simple yet powerful methodological framework—combining the anesthesia emergence paradigm with an automated analysis pipeline—to facilitate the rigorous screening of current and future chemogenetic tools. Together, these results underscore the importance of accounting for the off-target effects of DREADD ligands—through careful dose selection and the inclusion of DREADD-free ligand-treated controls—in chemogenetic experiments.

Limitations of the study

This study was designed as a sensitive and practical screening approach for detecting off-target effects of DREADD ligands under experimentally relevant conditions, rather than as a full pharmacological characterization of each compound. We therefore focused on commonly used doses and on isoflurane anesthesia, a standard anesthetic context in systems neuroscience and chemogenetic experiments. While this choice maximizes experimental relevance, future work will be required to define dose-response relationships and establish the thresholds at which each ligand can be considered behaviorally neutral. In addition, the anesthesia emergence paradigm provides a sensitive behavioral readout of ligand interference but does not identify the molecular receptors or neural circuits responsible for the observed effects. The relative contribution of candidate mechanisms, including histaminergic, muscarinic, serotonergic, dopaminergic, or noradrenergic pathways, will require dedicated mechanistic studies.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Guglielmo Foffani (gfoffani@fundacioncien.es).

Materials availability

This study did not generate new unique reagents or mouse lines.

Data and code availability

• •All raw behavioral datasets associated with DeepLabCut results reported in this study have been deposited at Zenodo (https://doi.org/10.5281/zenodo.19541529).
• •The trained DeepLabCut model (v.3.0.0rc6) used for tracking is available at Zenodo (https://doi.org/10.5281/zenodo.19538650). The original MATLAB scripts for the automated behavioral analysis pipeline, including anesthesia induction and emergence monitoring, are available on GitHub (https://doi.org/10.5281/zenodo.20664516).
• •Any additional information needed to reanalyze the data presented in this working paper is available upon request from the primary contact.

Acknowledgments

The project leading to these results was supported by MCIN/AEI/10.13039/501100011033 with co-funding by the European Regional Development Fund (ERDF) of the European Union (grant PID2021-128623OB-I00 to G.F. and fellowship FPU20/01061 to M.M.-G.).

Author contributions

Conceptualization, G.F. and D.H.-V.; investigation, experiment, and analysis, M.M.-G.; writing – original draft, M.M.-G.; writing – review and editing, G.F. and D.H.-V.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Chemicals, peptides, and recombinant proteins
Clozapine-N-oxide	Tocris, Bristol, UK	Cat# 4936; RRID:SCR_003689
Compound 21 (DREADD agonist 21 dihydrochloride)	Tocris, Bristol, UK	Cat# 6422; RRID:SCR_003689
Deschloroclozapine	Tocris, Bristol, UK	Cat# 7193; RRID:SCR_003689
Isoflurane (ISOFLO 100% w/w liquid for steam inhalation)	Zoetis Spain, Madrid, Spain	QN01AB06
Deposited data
Raw behavioral datsets associated with Deeplabcut results reported in this study	This paper	https://doi.org/10.5281/zenodo.19541529
Anesthesia paradigm	This paper	N/A
Experimental models: Organisms/strains
Mouse: C57BL6. Wild-type strain	The Jackson Laboratory	RRID:IMSR_JAX:000664
Software and algorithms
DeepLabCut (version 3.0.0rc6)	Mathis et al.56	https://deeplabcut.github.io/DeepLabCut/README.html
JASP (version 0.19.3.0; JASP Team, 2025)	Love et al.59	https://jasp-stats.org/
MATLAB (version R2018a & R2021a)	MathWorks, Natick, MA, USA	N/A
Trained Deeplabcut algorithm used for tracking	This paper	https://doi.org/10.5281/zenodo.19538650
MATLAB script for automated behavioral analysis pipeline	This paper	https://doi.org/10.5281/zenodo.20664516

Experimental model and study participant details

A total of 42 wild-type C57BL6 mice (20 females) aged between 6 and 11 months were used for this study. All procedures were conducted in strict accordance with the European Union regulation 2010/63/EU and the International Council for Laboratory Animal Science guidelines. The study protocol was reviewed and approved by the Animal Care Committee of HM-CINAC (Madrid, Spain; PROEX 330.4/23). Animals were group-housed under a controlled 12/12h light/dark cycle with ad libitum access to food and water. Animals were not involved in previous procedures.

Method details

Animals were randomly assigned to receive different DREADD ligands in a within-subject, randomised, cross-over design. The study involved two distinct sets of animals, differentiated by the route of ligand administration.

In the first group (n = 20, 0 deaths), animals received intraperitoneal administrations. The DREADD ligands tested were CNO (5 mg kg⁻¹; Tocris, Bristol, UK), C21 (3 mg kg⁻¹; Tocris), and DCZ (0.1 mg kg⁻¹; Tocris),^23,26,30 each dissolved in 1% DMSO in saline. For each behavioral test, animals received either a DREADD ligand or its corresponding control solution (1% DMSO in saline). Following intraperitoneal administration, animals were immediately placed back in their home cage for either 15 min (CNO, C21 and corresponding controls)¹⁷ or 5 min (DCZ and corresponding control)²³ before the anesthesia paradigm.¹⁷

To control for potential order effects, each animal received all DREADD ligands and their respective controls in a fully counterbalanced, within-subject design. The procedure was as follows: first, the animal cohort was randomized in two subgroups.⁶⁰ One subgroup received the real DREADD ligand and the other received its control, after which the first anesthesia paradigm was conducted. Following a 72-h washout period,^{3,13,17,23,61,62} a second test session was performed in which the administrations were crossed over: the subgroup that had initially received the real ligand was given the control, and vice versa. This entire two-session crossover was then repeated for the subsequent DREADD ligands in the subsequent two weeks, with the cohort being re-randomized at the beginning of each new ligand (C21 and then DCZ). This ensures that each animal served as its own control for every DREADD ligand tested.

A second group of animals (n = 22, 1 death following the anesthesia paradigm) was tested using subcutaneous administrations. The DREADD ligands for this part of the study were CNO (5 mg kg⁻¹) and C21 (3 mg kg⁻¹), administered alongside their respective controls as described above. For this subcutaneous administration, the anesthesia paradigm was performed 60 min post-administration for all conditions.

Anesthesia Paradigm. Following the specified post-administration waiting periods, the anesthesia paradigm was systematically evaluated through behavioral test. Animals were placed in a transparent acrylic anesthesia chamber (27 × 12.1 × 12.5 cm) where they were continuously video-monitored throughout three consecutive phases: a 2-min free-moving baseline period, a 10-min period of isoflurane inflow (1.75%, 0.8 L min−1 O₂; Zoetis) and a subsequent recovery period initiated upon cessation of the isoflurane inflow [mean ± SD duration (range): 12.1 ± 5.1 min (5.6–34.9 min)] (Figure 5A). Throughout the paradigm, a thermal blanket set at 35⁰C was placed under the chamber to maintain the mouse’s body temperature during anesthesia phases. Animals were transferred to their house-cage after completing the motor recovery phase (i.e., after the onset of walking).

Quantification and statistical analysis

Manual Behavioral Analysis. Manual recording of motor events was performed by an experimenter observing each animal in real time throughout the entire experimental session. Using a digital event timer, the experimenter logged the precise timestamp of each motor event at the moment the behavioral criterion was met (Table 1). The initiation of isoflurane inflow was defined as the starting time for the anesthesia induction phase. Conversely, the cessation of isoflurane inflow determined the starting time for the anesthesia emergence phase.

Automated Behavioral Analysis. To develop an automated method for quantifying motor events, an analysis pipeline was created using video recordings from all experimental sessions. These sessions were video-recorded from the top, at a distance of 66.5 cm from the base of the anesthesia chamber, with a resolution of 1920 × 1080 pixels and a frame rate of 25 Hz.

The videos underwent objective offline analysis using DeepLabCut (version 3.0.0rc6; an automatic markerless tracking software leveraging deep learning. The deep neural network was trained on a custom-labelled dataset of 1845 frames from 75 different recordings, using anatomical labels including the snout, head center, both ears, neck, body center, left and right body sides, tail base, and tail tip (Figure 5B). Training included 95% of the labeled data over 200 epochs (batch size 8). During pose estimation from new videos, coordinates with a confidence likelihood below 0.95 were excluded and imputed via linear interpolation from temporally adjacent, high-confidence values.

We aimed to automatically determine “walking onset” after anesthesia. From the resulting trajectories for each tracked body part, the instantaneous speeds of the head center, body center, and tail base were calculated, irrespective of the direction of movement. To differentiate true locomotor events from other preliminary or non-locomotor movements, these speed profiles were then used to precisely identify the “walking onset” through a four-step process based on identifying and filtering movement bouts, defined as discrete continuous periods of motor activity (Figures 5C and 5D). First, bouts were detected using a dynamic movement threshold defined for each of the three body parts as the mean +8 Standard Deviations of the speed during a 3-min baseline anesthesia period (minutes 2–5 before isoflurane withdrawal). Second, to correct for multiple consecutive movements, any single bout containing a period of low activity (speed below 5% of the bout’s peak for at least 3s) was split into separate events. Third, candidate bouts had to surpass a kinetic threshold of 50 pixels for minimum net displacement distance – calculated as the straight-line distance from the movement’s start to its end. Finally, to exclude large non-locomotion movements that could appear before locomotion ones, the final automatic "Walking onset"' was assigned to the start time of the single movement bout with the maximum net displacement x path length. This entire identification process first required concurrent movement in all three key points; if this failed, the criterion was relaxed to require movement in at least two of the three points (17 out of 203 sessions, with 3 sessions being lost due to the subtle nature of their initial movement).

Statistical Analysis. Data were analyzed using Bayesian statistics in JASP (version 0.19.3.0; JASP Team, 2025). The effects of DREADD ligands on the timing of motor events were evaluated through a hierarchical series of analyses separately for intraperitoneal or subcutaneous administration.

First, to determine if DREADD ligands produced an overall effect (Figures 1, 2 and 3, S1, and S2), we conducted two separate Bayesian ANOVAs over all ligands and all manual measures, either of the anesthesia induction or emergence phase, implemented as Bayesian mixed-models ANOVA to account for possible violations of sphericity and/or missing values. The model included 'Real/control' (real ligand vs. control), 'DREADD_ligand' (CNO, C21 or DCZ) and "Measure" (Table 1) as fixed factors, with ‘Animal’ specified as a random factor. This constituted the primary overall three-way ANOVA. Follow-up Bayesian two-way ANOVAs were performed separately for each measure to identify the most robust event for automatic analysis; for each DREADD ligand (within ligand comparisons) to explore the effect of each DREADD ligand individually; for all delta values (time on real ligand less time on control; between ligands comparisons) to explore their relative efficacy. Bayesian paired t-tests were conducted as post hoc analyses. Additionally, to compare the effects of the intraperitoneal versus subcutaneous administration for delta CNO and C21 values (within ligand comparison), secondary Bayesian two-way ANOVAs were conducted through a Bayesian mixed-models ANOVA with 'Measure', and "I.p./s.c." (intraperitoneal vs. subcutaneous administration) as fixed factors, with "Animal" specified as a random factor. Bayesian independent t-tests were conducted as post hoc analyses.

An analogous analysis was conducted for the automated ‘Walking onset’ measure (Figure 4). To validate the automated pipeline, Bayesian Pearson’s correlation analyses were performed between the automated and manual measures of ‘Walking onset’. Overall effects were assessed with a Bayesian two-way ANOVA, implemented as a Bayesian mixed models ANOVA for ensuring coherence with previous analysis. It included ‘Real/control’ and ‘DREADD_ligand’ as fixed factors, and ‘Animal’ as random factor. The post hoc comparison between intraperitoneal and subcutaneous administration was performed using paired t-tests on delta values (time on real ligand minus time on control) for each DREADD ligand.

For all Bayesian analyses, default JASP priors were used. Evidence in favor of the alternative hypothesis (BF > 1) or the null hypothesis (BF < 1) is reported in “Results” section and Table 2 as BF_incl for ANOVA effects or BF₊₀ for pairwise comparisons, and described according to standard levels: anecdotal (1/3 < BF < 3), moderate (3 ≤ BF < 10 or 1/10 < BF ≤ 1/3), strong (10 ≤ BF < 30 or 1/30 < BF ≤ 1/10), very strong (30 ≤ BF < 100 or 1/100 < BF ≤ 1/30), and extreme (BF ≥ 100 or BF ≤ 1/100). F-values and degrees of freedom for all analyses are detailed in Table S1. Full statistical details – including sample size (n), statistical test applied and effect analyzed – are provided throughout the “Results” section. Graphs show the measures of central tendency and dispersion, expressed as mean and 95% credibility intervals.

Published: July 17, 2026