Author response:

The following is the authors’ response to the original reviews.

Public Reviews:

Reviewer #2 (Public review):

(1) Vglut2 isn't a very selective promoter for the STN. Did the authors verify every injection across brain slices to ensure the para-subthalamic nucleus, thalamus, lateral hypothalamus, and other Vglut2-positive structures were never infected?

The STN is anatomically well-confined, with its borders and the overlying zona incerta (composed of GABAergic neurons) providing protection against off-target expression in most neighboring forebrain regions. All viral injections were histologically verified and did not into extend into thalamic or hypothalamic areas. As described in the Methods, we employed an app we developed (Brain Atlas Analyzer, available on OriginLab) that aligns serial histological sections with the Allen Brain Atlas to precisely assess viral spread and confirm targeting accuracy. The experiments included in the revised manuscript now focus on optogenetic inhibition and irreversible lesion approaches—three complementary methods that consistently targeted the STN and yielded similar behavioral effects.

(2) The authors say in the methods that the high vs low power laser activation for optogenetic experiments was defined by the behavioral output. This is misleading, and the high vs low power should be objectively stated and the behavioral results divided according to the power used, not according to the behavioral outcome.

Optogenetic excitation is no longer part of the study.

(3) In the fiber photometry experiments exposing mice to the range of tones, it is impossible to separate the STN response to the tone from the STN response to the movement evoked by the tone. The authors should expose the mouse to the tones in a condition that prevents movement, such as anesthetized or restrained, to separate out the two components.

The new mixed-effects modeling approach clearly differentiates sensory (auditory) from motor contributions during tone-evoked STN activation. In prior work (see Hormigo et al, 2023, eLife), we explored experimental methods such as head restraint or anesthesia to reduce movement, but we concluded that these approaches are unsuitable for addressing this question. Mice exhibit substantial residual movement even when head-fixed, and anesthesia profoundly alters neural excitability and behavioral state, introducing major confounds. To fully eliminate movement would require paralysis and artificial ventilation, which would again disrupt physiological network dynamics and raise ethical concerns. Therefore, the current modeling approach—incorporating window-specific covariates for movement—is the most appropriate and rigorous way to dissociate tone-evoked sensory activity from motor activity in behaving animals.

(4) The claim 'STN activation is ideally suited to drive active avoids' needs more explanation. This claim comes after the fiber photometry experiments during active avoidance tasks, so there has been no causality established yet.

Text adjusted.

(5) The statistical comparisons in Figure 7E need some justification and/or clarification. The 9 neuron types are originally categorized based on their response during avoids, then statistics are run showing that they respond differently during avoids. It is no surprise that they would have significantly different responses, since that is how they were classified in the first place. The authors must explain this further and show that this is not a case of circular reasoning.

Statistically verifying the clustering is useful to ensure that the selected number of clusters reflects distinct classes. It is also necessary when different measurements are used to classify (movement time series classified the avoids) and to compare neuronal types within each avoid mode/class (know called “mode”). Moreover, the new modeling approach goes beyond the prior statistical limitations related to considering movement and neuronal variables separately.

(6) The authors show that neurons that have strong responses to orientation show reduced activity during avoidance. What are the implications of this? The author should explain why this is interesting and important.

The new modeling approach goes beyond the prior analysis limitations. For instance, it shows that most of the prior orienting related activations closely reflect the orienting movement, and only in a few cases (noted and discussed in the results) orienting activations are related to the behavioral contingencies or behavioral outcomes in the task.

(8) The experiments in Figure 10 are used to say that STN stimulation is not aversive, but they only show that STN stimulation cannot be used as punishment in place of a shock. This doesn't mean that it is not aversive; it just means it is not as aversive as a shock. The authors should do a simpler aversion test, such as conditioned or real-time place preference, to claim that STN stimulation is not aversive. This is particularly surprising as previous work (Serra et al., 2023) does show that STN stimulation is aversive.

Optogenetic excitation is no longer part of the study.

(7) It is not clear which conditions each mouse experienced in which order. This is critical to the interpretation of Figure 9 and the reduction of passive avoids during STN stimulation. Did these mice have the CS1+STN stimulation pairing or the STN+US pairing prior to this experiment? If they did, the stimulation of the STN could be strongly associated with either punishment or with the CS1 that predicts punishment. If that is the case, stimulating the STN during CS2 could be like presenting CS1+CS2 at the same time and could be confusing.

Optogenetic excitation is no longer part of the study.

(8) The experiments in Figure 10 are used to say that STN stimulation is not aversive, but they only show that STN stimulation cannot be used as punishment in place of a shock. This doesn't mean that it is not aversive; it just means it is not as aversive as a shock. The authors should do a simpler aversion test, such as conditioned or real-time place preference, to claim that STN stimulation is not aversive. This is particularly surprising as previous work (Serra et al., 2023) does show that STN stimulation is aversive.

Optogenetic excitation is no longer part of the study.

(9) In the discussion, the idea that the STN encodes 'moving away' from contralateral space is pretty vague and unsupported. It is puzzling that the STN activates more strongly to contraversive turns, but when stimulated, it evokes ipsiversive turns; however, it seems a stretch to speculate that this is related to avoidance. In the last experiments of the paper, the axons from the STN to the GPe and to the midbrain are selectively stimulated. Do these evoke ipsiversive turns similarly?

Optogenetic excitation is no longer part of the study.

(10) In the discussion, the authors claim that the STN is essential for modulating action timing in response to demands, but their data really only show this in one direction. The STN stimulation reliably increases the speed of response in all conditions (except maximum speed conditions such as escapes). It seems to be over-interpreting the data to say this is an inability to modulate the speed of the task, especially as clear learning and speed modulation do occur under STN lesion conditions, as shown in Figure 12B. The mice learn to avoid and increase their latency in AA2 vs AA1, though the overall avoids and latency are different from controls. The more parsimonious conclusion would be that STN stimulation biases movement speed (increasing it) and that this is true in many different conditions.

Optogenetic excitation is no longer part of the study.

(11) In the discussion, the authors claim that the STN projections to the midbrain tegmentum directly affect the active avoidance behavior, while the STN projections to the SNr do not affect it. This seems counter to their results, which show STN projections to either area can alter active avoidance behavior. What is the laser power used in these terminal experiments? If it is high (3mW), the authors may be causing antidromic action potentials in the STN somas, resulting in glutamate release in many brain areas, even when terminals are only stimulated in one area. The authors could use low (0.25mW) laser power in the terminals to reduce the chance of antidromic activation and spatially restrict the optical stimulation.

Optogenetic excitation is no longer part of the study.

(12) Was normality tested for data prior to statistical testing?

Yes, although now we use mixed models

(13) Why are there no error bars on Figure 5B, black circles and orange triangles?

When error bars are not visible, they are smaller than the trace thickness or bar line—for example, in Figure 5B, the black circles and orange triangles include error bars, but they are smaller than the symbol size.

Reviewer #3 (Public review):

(1) I really don't understand or accept this idea that delayed movement is necessarily indicative of cautious movements. Is the distribution of responses multi-modal in a way that might support this idea, or do the authors simply take a normal distribution and assert that the slower responses represent 'caution'? Even if responses are multi-modal and clearly distinguished by 'type', why should readers think this that delayed responses imply cautious responding instead of say: habituation or sensitization to cue/shock, variability in attention, motivation, or stress; or merely uncertainty which seems plausible given what I understand of the task design where the same mice are repeatedly tested in changing conditions. This relates to a major claim (i.e., in the work's title).

In our study, “caution” is defined operationally as the tendency to delay initiation of an avoidance response in demanding situations (e.g., taking more time or care before crossing a busy street). The increase in avoidance latency with task difficulty is highly robust, as we have shown previously through detailed analyses of timing distributions and direct comparisons with appetitive behaviors (e.g., Zhou et al., 2022 JNeurosci). Moreover, we used the tracked movement time series to statistically classify responses into cautious modes, which is likely novel. This definition can dissociate cautious responding from broader constructs listed by a reviewer, such as attention, motivation, or stress, which must be explicitly defined to be rigorously considered in this context, including the likelihood that they covary with caution without being equivalent to it.

Cue-evoked orienting responses at CS onset are directly measured, and their habituation and sensitization have been characterized in our prior work (e.g., Zhou et al., 2023 JNeurosci). US-evoked escapes are also measured in the present study and directly compared with avoidance responses. Together, these analyses provide a rigorous and consistent framework for defining and quantifying caution within our behavioral procedures.

Importantly, mice exhibit cautious responding as defined here across different tasks, making it more informative to classify avoidance responses by behavioral mode rather than by task alone. Accordingly, in the miniscope, single-neuron, and mixed-effects model analyses, we classified active avoids into distinct modes reflecting varying levels of caution. Although these modes covary with task contingencies, their explicit classification improves model predictability and interpretability with respect to cautious responding.

(2) Related to the last, I'm struggling to understand the rationale for dividing cells into 'types' based the their physiological responses in some experiments (e.g., Figure 7).

This section has now been expanded into 3 figures (Fig. 7-9) with new modeling approaches that should make the rationale more straight forward.

By emphasizing the mixed-effects modeling results and integrating these analyses directly into the figures, the revised manuscript now more clearly delineates what is encoded at the population and single-neuron levels. Including movement and baseline covariates allowed us to dissociate motor-related modulation from other neural signals, substantially clarifying the distinction between movement encoding and other task-related variables, which we focus on in the paper. These analyses confirm the strong role of the STN in representing movement while revealing additional signals related to aversive stimulation and cautious responding that persist after accounting for motor effects. These signals arise from distinct neuronal populations that can be differentiated by their movement sensitivity and activation patterns across avoidance modes, reflecting varying levels of caution. At the same time, several effects that initially reflected orienting-related activity at CS-onset (note that our movement tracking captures both head position and orientation as a directional vector) dissipated once movement and baseline covariates were included in the models, emphasizing the utility of the analytical improvements in the revision.

(3)The description and discussion of orienting head movements were not well supported, but were much discussed in the avoidance datasets. The initial speed peaks to cue seem to be the supporting data upon which these claims rest, but nothing here suggests head movement or orientation responses.

As described in the methods (and noted above), we track the head and decompose the movement into rotational and translational components. With the new approach, several effects that initially reflected orienting-related activity at CS-onset (note that our movement tracking captures both head position and orientation as a directional vector) dissipated once movement and baseline covariates were included in the models, emphasizing the utility of the analytical improvements in the revision.

(4) Similar to the last, the authors note in several places, including abstract, the importance of STN in response timing, i.e., particularly when there must be careful or precise timing, but I don't think their data or task design provides a strong basis for this claim.

The avoidance modes and the measured latencies directly support the relation to action timing, but now the portion of the previous paper about optogenetic excitation and apparently the main source of criticism is no longer in the present study.

(5) I think that other reports show that STN calcium activity is recruited by inescapable foot shock as well. What do these authors see? Is shock, independent of movement, contributing to sharp signals during escapes?

The question, “Is shock, independent of movement, contributing to sharp signals during escapes?” is now directly addressed in the revised analyses. By incorporating movement and baseline covariates into the mixed-effects models, we dissociate STN activity related to aversive stimulation from that associated with motor output. The results show that shock-evoked STN activation persists even after controlling for movement within defined neuronal populations, supporting a specific nociceptive contribution independent of motor dynamics—a dissociation that appears to be new in this field.

(6) In particular, and related to the last point, the following work is very relevant and should be cited: Note that the focus of this other paper is on a subset of VGLUT2+ Tac1 neurons in paraSTN, but using VGLUT2-Cre to target STN will target both STN and paraSTN.

We appreciate the reviewer’s reference to the recent preprint highlighting the role of the para-subthalamic nucleus in avoidance learning. However, our study focused specifically on performance in well-trained mice rather than on learning processes. Behavioral learning is inherently more variable and can be disrupted by less specific manipulations, whereas our experiments targeted the stable execution of learned avoidance behaviors. Future work will extend these findings to the learning phase and examine potential contributions of subthalamic subdivisions, which our current Vglut2-based manipulations do not dissociate. We will consider this and related work more closely in those studies.

(7) In multiple other instances, claims that were more tangential to the main claims were made without clearly supporting data or statistics. E.g., claim that STN activation is related to translational more than rotational movement; claim that GCaMP and movement responses to auditory cues were small; claims that 'some animals' responded differently without showing individual data.

We have adjusted the text accordingly.

(8) In several figures, the number of subjects used was not described. This is necessary. Also necessary is some assessment of the variability across subjects. The only measure of error shown in many figures relates to trial-to-trial or event variability, which is minimal because, in many cases, it appears that hundreds of trials may have been averaged per animal, but this doesn't provide a strong view of biological variability. When bar/line plots are used to display data, I recommend showing individual animals where feasible.

All experiments report number of mice and sessions. Wherever feasible, we display individual data points (e.g., Figures 1 and 2) to convey variability directly. However, in cases where figures depict hundreds of paired (repeated-measures) data points, showing all points without connecting them would not be appropriate, while linking them would make the figures visually cluttered and uninterpretable. All plots and traces include measures of variability (SEM), and the raw data will be shared on Dryad. When error bars are not visible, they are smaller than the trace thickness or bar line—for example, in Figure 5B, the black circles and orange triangles include error bars, but they are smaller than the symbol size.

Also, to minimize visual clutter, only a subset of relevant comparisons is highlighted with asterisks, whereas all relevant statistical results, comparisons, and mouse/session numbers are fully reported in the Results section, with statistical analyses accounting for the clustering of data within subjects and sessions.

(9) Can the authors consider the extent to which calcium imaging may be better suited to identify increases compared to decreases and how this may affect the results, particularly related to the GRIN data when similar numbers of cells show responses in both directions (e.g., Figure 3)?

This is an interesting issue related to a widely used technique beyond the scope of our study.

(10) Raw example traces are not provided.

We do not think raw traces are useful here. All figures contain average traces to reflect the activity of the estimated population.

(11) The timeline of the spontaneous movement and avoidance sessions was not clear, nor was the number of events or sessions per animal nor how this was set. It is not clear if there was pre-training or habituation, if many or variable sessions were combined per animal, or what the time gaps between sessions were, or if or how any of these parameters might influence interpretation of the results.

We have enhanced the description of the sessions, including the number of animals and sessions, which are daily and always equal per animals in each group of experiments. As noted, the sessions are part of the random effects in the model.

(12) It is not clear if or how the spread of expression outside of the target STN was evaluated, and if or how many mice were excluded due to spread or fiber placements.

Recommendations for the authors:

Reviewing Editor Comments:

The primary feedback agreed upon by all the reviewers was that the manuscript requires significant streamlining as it is currently overly long and convoluted.

We thank the reviewers and editors for their thoughtful and constructive feedback. In response to the primary comment that “the manuscript requires significant streamlining as it is currently overly long and convoluted,” we have substantially revised and refocused the paper. Specifically, we streamlined the included data and enhanced the analyses to emphasize the central findings: the encoding of movement, cautious responding, and punishment in the STN during avoidance behavior. We also focused the causal component of the study by including only the loss-of-function experiments—both optogenetic inhibition and irreversible viral/electrolytic lesions—that establish the critical role of STN circuits in generating active avoidance. Together, these revisions enhance clarity, tighten the narrative focus, and align the manuscript more closely with the reviewers’ recommendations.

Major revisions include the addition of mixed-effects modeling to dissociate the contributions of movement from other STN-encoded signals related to caution and punishment. This modeling approach allowed us to reveal that these components are statistically separable, demonstrating that movement, cautious responding, and aversive input are encoded by neuronal subsets. To streamline the manuscript and address reviewer concerns, we removed the optogenetic excitation experiments. As revised, the paper presents a more concise and cohesive narrative showing that STN neurons differentially encode movement, caution, and aversive stimuli, and that this circuitry is essential for generating active avoidance behavior.

Many of the specific points raised by reviewers now fall outside the scope of the revised manuscript. This is primarily because the revised version omits data and analyses related to optogenetic excitation and associated control experiments. By removing these components, the paper now presents a streamlined and internally consistent dataset focused on how the STN encodes movement, cautious responding, and aversive outcomes during avoidance behavior, as well as on loss-of-function experiments demonstrating its necessity for generating active avoidance. Below, we address the points that remain relevant across reviews.

Following extensive revisions, the current manuscript differs in several important ways from what the assessment describes:

The description that the study “uses fiber photometry, implantable lenses, and optogenetics” is more accurately represented as using both fiber photometry and singleneuron calcium imaging with miniscopes, combined with optogenetic and irreversible lesion approaches.

The phrase stating that “active but not passive avoidance depends in part on STN projections to substantia nigra” is better characterized as “STN projections to the midbrain,” since our data show that optogenetic inhibition of STN terminals in both the mesencephalic reticular tegmentum (MRT) and substantia nigra pars reticulata (SNr) produce equivalent effects, and thus these sites are combined in the study.

Finally, the original concern that evidence for STN involvement in cautious responding or avoidance speed was incomplete no longer applies. The revised focus on encoding, through the inclusion of mixed-effects modeling, now dissociates movement-related, cautious, and aversive components of STN activity. By removing the optogenetic excitation data, we no longer claim that the STN controls caution but rather that it encodes cautious responding, alongside movement and punishment signals. Furthermore, loss-of-function experiments demonstrate that silencing STN output abolishes active avoidance entirely, supporting an essential role for the STN in generating goal-directed avoidance behavior—a behavioral domain that, unlike appetitive responding, is fundamentally defined by caution and the need to balance action timing under threat.

Reviewer #2 (Recommendations for the authors):

(1) Show individual data points on bar plots.

Wherever feasible, we display individual data points (e.g., Figures 1 and 2) to convey variability directly. However, in cases where figures depict hundreds of paired (repeatedmeasures) data points, showing all points without connecting them would not be appropriate, while linking them would make the figures visually cluttered and uninterpretable. All plots and traces include measures of variability (SEM), and the raw data will be shared on Dryad. When error bars are not visible, they are smaller than the trace thickness or bar line—for example, in Figure 5B, the black circles and orange triangles include error bars, but they are smaller than the symbol size.

(2) The active avoidance experiments are confusing when they are introduced in the results section. More explanation of what paradigms were used and what each CS means at the time these are introduced would add clarity. For example, AA1, AA2, etc, are explained only with references to other papers, but a brief description of each protocol and a schematic figure would really help.

The avoidance protocols (AA1–4) are now described briefly but clearly in the Results section (second paragraph of “STN neurons activate during goal-directed avoidance contingencies”) and in greater detail in the Methods section. As stated, these tasks were conducted sequentially, and mice underwent the same number of sessions per procedure, which are indicated. All relevant procedural information has been included in these sections. Mice underwent daily sessions and learnt these tasks within 1-2 sessions, progressing sequentially across tasks with an equal number of sessions per task (7 per task), and the resulting data were combined and clustered by mouse/session in the statistical models.

(3) How do the Class 1, 2, 3 avoids relate to Class 1, 2, 3 neural types established in Figure 3? It seems like they are not related, and if that is the case, they should be named something different from each other to avoid confusion. (4) Similarly, having 3 different cell types (a,b,c) in the active avoidance seems unrelated to the original classification of cell types (1,2,3), and these are different for each class of avoid. This is very confusing, and it is unclear how any of these types relate to each other. Presumably, the same mouse has all three classes of avoids, so there are recordings from each cell during each type of avoid.

The terms class, mode, and type are now clearly distinguished throughout the manuscript. Modes refer to distinct patterns of avoidance behavior that differ in the level of cautious responding (Mode 3 is most cautious). Within each mode, types denote subgroups of neurons identified based on their ΔF/F activity profiles. In contrast, classes categorize neurons according to their relationship to movement, determined by cross-correlation analyses between ΔF/F and head speed (Class1-4; Fig. 7 is a new analysis) or head turns (ClassA-C, renamed from 1-3). This updated terminology clarifies the analytic structure, highlighting distinct neuronal populations within each analysis. For example, during avoidance behaviors, these classifications distinguish neurons encoding movement-, caution-, and outcome-related signals. Comparisons are conducted within each analytical set, within classes (A-C or 1-4 separately), within avoidance modes, or within modespecific neuronal types.

…So the authors could compare one cell during each avoid and determine whether it relates to movement or sound, or something else. It is interesting that types a,b, and c have the exact same proportions in each class of avoid, and makes it important to investigate if these are the exact same cells or not.

That previous table with the a,b,c % in the three figure panels was a placeholder, which was not updated in the included figure. It has now been correctly updated. They do not have the same proportions as shown in Fig. 9, although they are similar.

Also, these mice could be recorded during the open field, so the original neural classification (class 1, 2,3) could be applied to these same cells, and then the authors can see whether each cell type defined in the open field has a different response to the different avoid types. As it stands, the paper simply finds that during movement and during avoidance behaviors, different cells in the STN do different things.

We included a new analysis in Fig. 7 that classifies neurons based on the cross-correlation with movement. The inclusion of the models now clearly assigns variance to movement versus the other factors, and this analysis leads to the classification based on avoid modes.

(5) The use of the same colors to mean two different things in Figure 9 is confusing. AA1 vs AA2 shouldn't be the same colors as light-naïve vs light signaling CS.

Optogenetic excitation is no longer part of the study.

(6) The exact timeline of the optogenetics experiments should be presented as a schematic for understanding. It is not clear which conditions each mouse experienced in which order. This is critical to the interpretation of Figure 9 and the reduction of passive avoids during STN stimulation. Did these mice have the CS1+STN stimulation pairing or the STN+US pairing prior to this experiment? If they did, the stimulation of the STN could be strongly associated with either punishment or with the CS1that predicts punishment. If that is the case, stimulating the STN during CS2 could be like presentingCS1+CS2 at the same time and could be confusing. The authors should make it clear whether the mice were naïve during this passive avoid experiment or whether they had experienced STN stimulation paired with anything prior to this experiment.

Optogenetic excitation is no longer part of the study.

(20) Similarly, the duration of the STN stimulation should be made clear on the plots that show behavior over time (e.g., Figure 9E).

Optogenetic excitation is no longer part of the study.

(21) There is just so much data and so many conditions for each experiment here. The paper is dense and difficult to read. It would really benefit readability if the authors put only the key experiments and key figure panels in the main text and moved much of the repetitive figure panels to supplemental figures. The addition of schematic drawings for behavioral experiment timing and for the different AA1, AA2, and AA3 conditions would also really improve clarity.

By focusing the study, we believe it has substantially improved clarity and readability.

Reviewer #3 (Recommendations for the authors):

(1) Minor error in results 'Cre-AAV in the STN of Vglut2-Cre' Fixed.

(2) In some Figure 2 panels, the peaks appear to be cut off, and blue traces are obscured by red.

In Fig. 2, the peaks of movement (speed) traces are intentionally truncated to emphasize the rising phase of the turn, which would otherwise be obscured if the full y-axis range were displayed (peaks and other measures are statistically compared). This adjustment enhances clarity without omitting essential detail and is now noted in the legend.

Figures and data

Calcium imaging fiber photometry reveals that STN glutamatergic neurons activate during spontaneous exploratory movement.

STN glutamatergic neurons code the direction of spontaneous contraversive exploratory turning movements.

A subgroup of STN glutamatergic neurons code contraversive movements.

STN glutamatergic neurons discharge to auditory tones in association with movement.

STN glutamatergic neuron activation in the context of signaled active avoidance.

STN glutamatergic neurons track the avoidance and escape movement.

Different classes of STN glutamatergic neurons during signaled active avoidance, classified by the cross-correlation between head speed and ΔF/F.

Activation of STN glutamatergic neurons across distinct modes of signaled active avoidance.

Subtypes of STN glutamatergic neurons across distinct modes of signaled active avoidance.

Optogenetic inhibition of STN glutamatergic neurons impairs signaled avoidance.

Lesions of STN glutamatergic neurons impair signaled avoidance learning and performance.

Optogenetic inhibition of STN glutamatergic fibers in Midbrain impairs signaled avoidance.