A striatal circuit balances learned fear in the presence and absence of sensory cues

  1. Michael Kintscher
  2. Olexiy Kochubey
  3. Ralf Schneggenburger  Is a corresponding author
  1. Laboratory for Synaptic Mechanisms, Brain Mind Institute, School of Life Science, Ecole Polytechnique Fédérale de Lausanne, Switzerland

Abstract

During fear learning, defensive behaviors like freezing need to be finely balanced in the presence or absence of threat-predicting cues (conditioned stimulus, CS). Nevertheless, the circuits underlying such balancing are largely unknown. Here, we investigate the role of the ventral tail striatum (vTS) in auditory-cued fear learning of male mice. In vivo Ca2+ imaging showed that sizable sub-populations of direct (D1R+) and indirect pathway neurons (Adora+) in the vTS responded to footshocks, and to the initiation of movements after freezing; moreover, a sub-population of D1R+ neurons increased its responsiveness to an auditory CS during fear learning. In-vivo optogenetic silencing shows that footshock-driven activity of D1R+ neurons contributes to fear memory formation, whereas Adora+ neurons modulate freezing in the absence of a learned CS. Circuit tracing identified the posterior insular cortex (pInsCx) as an important cortical input to the vTS, and recording of optogenetically evoked EPSCs revealed long-term plasticity with opposite outcomes at the pInsCx synapses onto D1R+ - and Adora+ neurons. Thus, direct- and indirect pathways neurons of the vTS show differential signs of plasticity after fear learning, and balance defensive behaviors in the presence and absence of learned sensory cues.

Editor's evaluation

This important study examines the contribution of an understudied brain region to fear conditioning in mice. The evidence supporting the authors' conclusions is convincing. This paper will interest neuroscientists working in the fields of basal ganglia, amygdala, and fear learning.

https://doi.org/10.7554/eLife.75703.sa0

Introduction

Fear learning is an evolutionary conserved behavior, critically important for animals to detect signs of danger in an ever-changing environment. As such, fear learning is necessary for survival (Phelps and LeDoux, 2005; Janak and Tye, 2015). Nevertheless, learned defensive behaviors need to be finely regulated, so that animals can return to their normal behaviors after the cessation of threat -predicting sensory cues (Zanette et al., 2011). Furthermore, a pathological overexpression of defensive behaviors is a hallmark of several anxiety-related disorders in humans (Dunsmoor et al., 2011). Therefore, it is important to understand the neuronal circuits that balance the expression of learned defensive behaviors during and after the presence of threat-predicting sensory cues.

The study of the neuronal mechanisms of fear learning has been strongly facilitated by employing auditory-cued fear learning in model animals like rodents (Davis, 1992; LeDoux, 2000). In fear learning, subjects learn to associate an initially neutral sensory cue, the conditioned stimulus (CS; often an auditory stimulus), with an aversive or painful outcome like a footshock (the unconditioned stimulus, US). After associative learning, subjects will develop a defensive behavior when the CS is later presented alone (Fanselow, 2018). The defensive behavior that is mostly studied in the context of fear learning in rodents is behavioral arrest, also called freezing (Fanselow, 1994). Studies spanning several decades have firmly established that the amygdala has an important role in fear learning (see Davis, 1992; LeDoux, 2000; Duvarci and Pare, 2014; Tovote et al., 2015, for reviews). The lateral amygdala (LA) is viewed as an input structure to the amygdalar complex (LeDoux et al., 1990), which connects to both the basal amygdala (BA) and the central amygdala, where further integration and processing takes place (Romanski et al., 1993; Quirk et al., 1995; Amano et al., 2011; Grewe et al., 2017). Finally, the execution of learned freezing depends on a central amygdala to midbrain (periaqueductal gray) projection (LeDoux et al., 1988; Tovote et al., 2016). Nevertheless, it is likely that further neuronal circuits beyond these amygdalar circuits are involved in fear learning.

The striatum is part of the basal ganglia motor system, a neuronal system with important roles in the control of movement, action selection, and reward-based learning (for reviews, see Hikosaka et al., 2000; Graybiel, 2005; Grillner et al., 2005; Redgrave et al., 2010; Nelson and Kreitzer, 2014; Klaus et al., 2019). The principal neurons of the striatum are inhibitory projection neurons of two types. First, striato-nigral neurons project in a direct pathway to basal ganglia output structures like the Substantia nigra pars reticularis and others; these neurons selectively express D1-dopamine receptors (D1R). Second, striato-pallidal neurons project in a more indirect pathway towards the basal ganglia output structures; these neurons selectively express D2-dopamine receptors, and also adenosine-2A receptors (Adora) (Gerfen et al., 1990; Schiffmann and Vanderhaeghen, 1993). Different sub-areas of the striatum have different roles in motor control and motor learning. The dorsal striatum is involved in the learning of motor sequences and in the selection of appropriate actions (Nelson and Kreitzer, 2014; Klaus et al., 2019), as well as in habit formation (Redgrave et al., 2010; Burguière et al., 2015), whereas the ventral striatum is important for reward-based learning (Humphries and Prescott, 2010; Cox and Witten, 2019).

Recently, based on brain-wide studies of the cortical inputs to the mouse striatum, a further sub-area of the striatum was identified; a posterior area called tail striatum (Hintiryan et al., 2016; Hunnicutt et al., 2016; see Valjent and Gangarossa, 2021 for review). Interestingly, early in vivo recordings in the LA have found CS- and US-responsive neurons in the tail striatum adjacent to the LA (Romanski et al., 1993). Recent work has shown that the tail striatum is, similarly as other striatal areas, composed of D1R-expressing (D1R+) and D2R- and Adora-expressing neurons (Gangarossa et al., 2019). Furthermore, in vivo imaging studies have shown that dopaminergic axons in the tail striatum code for salient sensory stimuli, but not for rewarding stimuli (Menegas et al., 2018). However, the role of the tail striatum in fear learning has not been studied.

Here, we use in vivo miniature microscope Ca2+ imaging, as well as in vivo and ex vivo optogenetic approaches and circuit tracing, to investigate the role of D1R+, and Adora+ neurons located in the ventral part of the tail striatum (vTS) in auditory-cued fear learning.

Results

Coding for footshocks, tones and movement by D1R+ vTS neurons

We started by imaging the activity of vTS neurons during a 3-day fear learning paradigm (Figure 1A). To access the vTS, a deep brain area close to the basolateral - and central nuclei of the amygdala, we used miniature-microscope Ca2+ imaging (Figure 1B and C; Ghosh et al., 2011). In a first series of experiments, we used Drd1aCre mice, to target the expression of GCaMP6m to neurons of the direct pathway, using a Cre-dependent AAV vector (Figure 1B and C; Materials and methods). The Drd1aCremouse chosen for this purpose (line EY217 from GenSAT; see Materials and methods) shows expression of Cre in the vTS, but no expression was observed in the adjacent cortical or claustrum structures (Figure 1D; Figure 1—figure supplement 1; see also Gerfen et al., 2013). This allowed us to target striatal neurons selectively by miniature-microscope imaging with the employed mouse line (Figure 1—figure supplement 2).

Figure 1 with 5 supplements see all
Miniature microscope Ca2+-imaging of D1R+ neurons in the vTS reveals coding for tones and movement during fear learning.

(A) Outline of the fear learning protocol. (B) Experimental scheme of the injection of an AAV vector and placement of the GRIN lens in the vTS of Drd1aCre mice. (C) Posthoc fluorescence microscopy image from an injected Drd1aCre mouse expressing GCaMP6m (green channel; blue channel: DAPI). The black region indicates the position of the GRIN lens. The putative imaging area is depicted with a red rectangle. Scalebar, 500 µm. (D) tdTomato expression in Drd1aCre x Rosa26LSL-tdTomato mice indicates localization of D1R+ neurons in the vTS, but not in neighboring cortical - nor amygdalar structures. (E - G) Movement traces - and freezing-state of an example mouse (red traces and light blue areas, respectively), Z-scored Ca2+ traces for three example neurons (black traces); and color-coded Z-scored Ca2+ traces for all neurons in one example mouse (bottom). Data are from the fourth CS presentation of day 1 (E), the fourth CS-US presentation of day 2 (F), and the second CS presentation on day 3 (G). (H) Color-coded Z-scored Ca2+ responses to footshocks in all imaged D1R+ neurons from N=8 mice. Responses with Z>1 in a time interval of [0; 1 s] were considered as significant (see white dashed line). The bottom panel shows the average ± S.E.M. of Ca2+ traces for all responders (n=45 neurons; black trace ± gray shades), and the average ± S.E.M. across all neurons (n=163; dashed black trace ± gray shades). (I - K) Z-scored Ca2+ traces aligned to tone beeps (CS) during the habituation day, training day, and recall day (I, J, K, respectively). Responses with an average Z-score >0.2 in the time interval [0; 0.5 s] were considered significant. The black traces in the lower panel follow the same logics as in (H). (L) Percentage of tone - responsive neurons for each day. (M - O) Color-coded Z-scored Ca2+ traces aligned to the movement - ON events, analyzed at times in-between CS blocks, for the habituation -, training - and recall days, as indicated (top panel). The traces in the bottom panel were analyzed as in (H). (P - R) Ca2+ traces aligned to the movement - ON events, analyzed during the 30 s tone blocks (CS), for the habituation -, training - and recall days (P, Q, and R respectively). (S) Percentage of movement - ON responding neurons during the CS, and in the absence of a CS (closed, and open symbols, respectively). (T) Example traces of, from top to bottom, times of tone beeps (blue trace); movement index (red trace); and Ca2+ traces from four example neurons in one mouse; times of freezing are highlighted by light blue. The detection of Ca2+ - events and their amplitudes by a deconvolution analysis is indicated by vertical bars (see Materials and methods). (U) The amplitude - weighted frequency of Ca2+ events (average ± S.E.M.) is plotted separately for the four combinations of CS / no CS epochs, and movement / freezing states of the mice, for the habituation, training, and fear memory recall day. The presence of a CS is indicated by the blue bars. For statistical parameters, see Results text.

Figure 1—source data 1

Raw data and statistical tests for Figure 1 and its supplements.

https://cdn.elifesciences.org/articles/75703/elife-75703-fig1-data1-v2.xlsx

The fear conditioning protocol consisted of three sessions given on subsequent days (Figure 1A). On the first day, mice experienced a habituation session during which six 30 s CS stimulation blocks consisting of 7 kHz tone beeps were applied (see Materials and methods). During this session, only a small sub-population of D1R+ neurons (6/176) showed a response to the tone beeps (Figure 1E, I). One day later, each of the six CS blocks was followed by a 1 s footshock, to which 45/163 imaged D1R+ neurons responded with robust Ca2+ signals (Figure 1F and H). During this training session, an increased number of D1R+ neurons responded to tone beeps (CS) (Figure 1J and L; Chi-square test, p=0.018, Χ2df=2 = 8.004); the average Ca2+ response of the neurons that responded was similar to the one observed during the habituation session (Figure 1I and J). During the training- and recall session, we also observed that tone beeps were in 10–15% of the cases followed by movement transitions; for the calculation of the number of CS - responders, these trials were removed (see Figure 1—figure supplement 3). Finally, during a fear memory recall session on day 3, the CS was presented alone in a different context. During this session, the number of tone-responsive neurons was comparable to the one during the training session (Figure 1L); the average Z-scored Ca2+ signal was increased above the response amplitude on the training session (Figure 1J and K; 0.366±0.071 vs 0.093±0.034 for n=19 and 17 responders, respectively; 95% CI: [0.2164; 0.516]; [0.0212; 0.1649]; p=0.001, U=61, Mann-Whitney test). Thus, in vivo Ca2+ imaging showed that a significant fraction of D1R+ neurons in the vTS responds to footshocks, and furthermore, that D1R+ neurons increase their responsiveness to the CS.

During fear learning, rodents acquire a defensive behavior in response to a CS, in the form of freezing (LeDoux, 2000; Fanselow, 2018). The freezing bouts of mice typically lasted a few seconds, and were interrupted by movement re-initiation (see e.g. Figure 1E–G, top). We asked whether transitions from freezing to movement would drive activity in D1R+ vTS neurons; we first restricted this analysis to times when no tones (CS) were presented. Aligning GCaMP6m fluorescence traces to the movement onset revealed Ca2+ events in a sub- population of D1R+ neurons; we call these ‘movement-ON’ responses. The number of responding neurons decreased across the 3-day fear learning protocol, although this trend did not reach statistical significance (Figure 1M–O; Figure 1S, open symbols; p=0.117; Χ2df=2 = 4.292, Chi-square test). We next analyzed movement-ON responses during the 30 s CS presentations (Figure 1P–R). We found that a substantial number of movement-ON transitions were preceded by a tone; for the calculation of the percentage of movement-ON responders, these events were removed (Figure 1—figure supplement 4). The analysis showed that the number of D1R+ neurons that responded to a movement-ON transition during the time of the CS presentations increased during the training session, and was maintained at an elevated level during the recall session (Figure 1S, filled symbols; p=0.0155, Χ2df=2 = 8.340, Chi-square test). Furthermore, the number of neurons responding to a movement-ON transition was always higher during the CS, as compared to no - CS periods (Figure 1S). Taken together, this data suggests that subpopulations of D1R+ neurons, in addition to responding to footshocks- and to tone stimulation, also code for movement onset.

We next analyzed more comprehensively how the Ca2+-event frequency depends on the movement state of the animal and on the presence or absence of a CS. For this, we first deconvolved the fluorescence traces to obtain times of Ca2+ events and their amplitudes (Figure 1T; Materials and methods; Pnevmatikakis et al., 2016; Giovannucci et al., 2019). This allowed us to compute the amplitude-weighted frequency of Ca2+ events during the four combinations of movement states / tone presentations, for each day of the fear learning protocol (Figure 1U; Figure 1—figure supplement 5). The data was significantly different across conditions (p<0.0001, KW = 138.33, Kruskal-Wallis test). Pairwise comparisons of these conditional Ca2+ event frequencies within each session showed that on most sessions, the activity of D1R+ neurons was significantly higher when the mice moved, than when they froze, with the exception of the no-CS times on the training day (Figure 1U; Dunn’s multiple comparisons test; for p-values, see legend to Figure 1—figure supplement 5). On the training day, given that mice moved, the Ca2+ event frequency was significantly increased by the CS (Figure 1U; p=0.0163, Mean rank difference = 241.9, Kruskal-Wallis test followed by Dunn’s multiple comparisons test). Thus, in vivo Ca2+ imaging during a 3-day fear learning protocol shows that the activity of the D1R+ neurons in the vTS is higher during movement than during the freezing state of the mice. Moreover, fear learning increases the number of neurons with a phasic response to tones (CS), as well as the number of neurons with a movement-ON response during the CS presentation (Figure 1L and S).

To determine the location of the imaged neurons and to compare them across mice, we aligned the center of the GRIN lens to a mouse brain atlas in each mouse, and generated a common cell map based on the cell coordinates relative to the lens center (Materials and methods). This revealed a hotspot of footshock- and tone-responding D1R+ neurons in the posterior-ventral region of the tail striatum medial to the LA (Figure 2A and B; Figure 2—figure supplement 1). D1R+ neurons with movement-ON responses during the recall sessions were located in a similar area (Figure 2C). Venn plots of the overlay of the various response types showed that during the habituation session, neurons with responses to the CS and movement did not strongly overlap, and represented a small proportion of the imaged neurons (Figure 2D). During the training session, the neurons with tone- and movement-ON responses increased in numbers (see also Figure 1L and S), and about half of each sub-population also showed a footshock response. During the recall session, the populations of both tone - and movement responders stayed constant with respect to the training day, and these response types now overlapped substantially (Figure 2D). Taken together, in vivo Ca2+ imaging shows that a subpopulation of D1R+ vTS neurons, located in a posterior-ventral hotspot of the tail striatum, responds to footshocks. During the course of fear learning, these neurons increasingly code for an aversively motivated CS and for movement-ON transitions, suggesting that these representations in populations of D1R+ vTS neurons undergo plasticity driven by fear learning.

Figure 2 with 1 supplement see all
Spatial localization of the imaged D1R+ vTS neurons, and overlap of neurons coding for sensory events and movement state.

(A–C) Maps with the position of all imaged D1R+ vTS neurons, plotted as projection on the horizontal plane. Shown are the footshock-responses (A), the tone (CS) responses as imaged on day3 (B), and the movement-onset responses imaged on day 3 (C). Neurons are drawn in red when their response was considered significant (average Z-score >1 for footshock responses; average Z-score >0.2 for tone, and movement - ON responses; note different scales of the circles). For a coarse orientation, outlines of the amygdalar nuclei (basolateral amygdala; ‘BLA’ and central amygdala, ‘CeA’) and cortex (‘Ctx’) are shown. (D) Venn diagrams showing the overlap of neuronal populations within the D1R+ vTS neurons that respond to tones (green), to movement-ON events (blue), and to footshock stimulation (red; on day 2 only). Note the increased number of neurons responding to tones and movement - onset during fear memory recall, and the overlap of these sub-populations (right panel).

Coding for footshocks and movement by Adora+ vTS neurons

The other large population of principal neurons in the vTS are Adora+ neurons which, in analogy to other striatal areas, represent neurons of an ‘indirect’ pathway through the basal ganglia (Gerfen et al., 1990; Gangarossa et al., 2019). We next investigated the in-vivo activity of this population of vTS neurons throughout the three-day fear learning protocol, using an Adora2aCre mouse line to target the expression of GCaMP6m to Adora+ neurons in the vTS (Figure 3A; Figure 3—figure supplements 1 and 2; Materials and methods). About forty percent of the Adora+ vTS neurons responded to footshocks presented during the training session (79/201 neurons; Figure 3C and E). A small subpopulation (13/173 or ~8%) responded to tone beeps during the CS (Figure 3F). Contrasting with the D1R+ neurons, the percentage of tone (CS)-responsive neurons did not change during fear learning (Figure 3F–H; Figure 3I; p=0.415, Χ2df=2 = 1.756, Chi-square test). Thus, a large sub-population of Adora+ neurons in the vTS responds to footshocks, but the number of Adora+ neurons that responds to tones (CS) remains unchanged during fear learning.

Figure 3 with 4 supplements see all
Adora+ neurons in the vTS code for footshocks and movement onset, but CS coding is less present.

(A) Experimental scheme of the injection of an AAV vector and placement of the GRIN lens in Adora2aCre mice. (B–D) Movement traces - and freezing-state of an example mouse (red traces and light blue areas, respectively), Z-scored Ca2+ traces for three example neurons (black traces); and color-coded Z-scored Ca2+ traces for all neurons in one example mouse (bottom). Data are from the second CS presentation of day 1 (D), the fourth CS-US presentation of day 2 (E), and the second CS presentation on day 3 (F). (E) Z-scored Ca2+ responses to footshocks in all imaged Adora+ neurons from N=8 mice; responses with average Z>1 in the interval of [0; 1 s] were considered as significant (traces above dashed white line). The bottom panel shows the average ± S.E.M. of Ca2+ traces for all responders (n=79 neurons; black trace ± gray shades), and the average ± S.E.M. across all neurons (n=201; dashed black trace ± gray shades). (F – H). Color-coded Z-scored Ca2+ traces aligned to tone beeps (CS) during the habituation day, training day, and recall day (F, G, H, respectively). Responses with average Z-score >0.2 in the time interval of [0; 0.5 s] were considered significant; the black traces in the lower panel were calculated as in (E). (I) Percentage of tone-responsive neurons for each day. (J - L). Ca2+ responses to movement-ON events, analyzed outside the CS blocks, for the habituation-, training-, and recall days (J, K, and L, respectively). The top and bottom panels follow the same logics as in (E). (M - O) Ca2+ responses to movement-ON events, analyzed during the 30 s tone blocks (CS), for the habituation, training, and recall days as indicated. (P) Percentage of movement - ON responders in the presence and absence of a CS (closed, and open symbols). (Q) Illustration of the Ca2+ deconvolution approach for four example neurons in a Adora2aCre mouse. From top to bottom, times of tone beeps (blue trace); movement index (red trace); and Ca2+ traces from four example neurons; times of freezing are highlighted by light blue. Vertical gray bars indicate the timing and amplitude of the detected Ca2+ events. (R) The amplitude-weighted frequency of Ca2+ events (average ± S.E.M.), analyzed separately for the four combinations of CS / no CS times, and movement / freezing states of the mice, for the 3 fear learning days. The presence of a CS is indicated by the blue bars. The p-values for the indicated statistical comparisons are reported in the Results text. (S) Venn diagrams showing the overlap of neuronal populations within the Adora+ vTS neurons that respond to tones (green), to movement-onset transitions (blue), and to footshock stimulation.

Figure 3—source data 1

Raw data and statistical tests for Figure 3 and its supplements.

https://cdn.elifesciences.org/articles/75703/elife-75703-fig3-data1-v2.xlsx

We next analyzed whether the activity of Adora+ neurons in the vTS was modulated by the movement state of the mice. In the absence of tones, a moderate sub-population of Adora+ neurons showed movement-ON responses during the habituation session (27/173 or ~16 %); this number decreased over the course of fear learning (Figure 3J–L; Figure 3P, open symbols; p=0.0229, Χ2df=2 = 7.557; Chi-square test). In contrast, during the CS, there was a larger number of Adora+ neurons that showed movement-ON responses; their number was unchanged over the three-day fear learning protocol (Figure 3M–O; Figure 3P, closed symbols; p=0.717, Χ2df=2 = 0.67; Chi-square test). These experiments show that a substantial sub-population of Adora+ neurons in the vTS codes for movement onset, but this representation was unchanged by fear learning (Figure 3M–P), except for a decrease in the number of neurons showing a movement - ON response in the absence of a CS (Figure 3P, open symbols).

Similarly as for the D1R+ neurons, we next analyzed the activity of Adora+ neurons as a function of the four combinations of movement state (movement versus freezing) and CS presentation (presence, or absence of a CS) (Figure 3Q and R). A Kruskal-Wallis test showed that the amplitude-weighted frequency of Ca2+ events differed across categories (Figure 3R; p<0.0001, KW = 184.6). During all three behavior sessions, and irrespective of the presence or absence of a CS, the activity of Adora+ neurons was larger during movement than during freezing (Figure 3R; see legend to Figure 3—figure supplement 3 for the corresponding p-values; post-hoc Dunn’s multiple comparison test). On the other hand, the presence or absence of a CS did not significantly modulate the activity of Adora+ neurons, irrespective of whether the mice moved, or froze (Figure 3R; see legend to Figure 3—figure supplement 3 for p-values; post-hoc Dunn’s multiple comparison test). This analysis thus corroborates our finding that the activity of Adora+ neurons is only little modulated by tones and that it is more strongly modulated by movement, but that neither of the two representations are modulated in a plastic fashion by fear learning. The spatial distribution of Adora+ neurons responding to footshocks, tones and movement-ON transitions was overall similar to the one of D1R+ neurons (Figure 3—figure supplement 4). Taken together, in vivo Ca2+ imaging shows that a substantial percentage of Adora+ neurons in the vTS responds to footshocks, and to movement-ON transitions, and a smaller sub-population of these neurons responds to tones (Figure 3S). However, Adora+ neurons do not change their responses to tones and movement-ON transitions during fear learning.

D1R+ and Adora+ vTS neurons do not instruct freezing or movement

In vivo Ca2+ imaging showed that sub-populations of neurons within the two main types of principal neurons in the vTS code for footshocks, for the CS and for movement-ON transitions; furthermore, D1R+ neurons increased their representation of tones and movements with fear learning (Figures 13). We next wished to investigate how D1R+ and Adora+ vTS neurons might contribute to fear learning. A classical model of basal ganglia function postulates that D1R+ neurons in the direct pathway initiate movement, whereas Adora+ (or D2R+) neurons in the indirect pathway suppress movements (Kravitz et al., 2010; Nelson and Kreitzer, 2014; but see Klaus et al., 2019). Thus, one possible straightforward hypothesis is that activity of Adora+ neurons of the vTS instructs an arrest of movement, or freezing, and vice-versa, that D1R+ neurons instruct movement re-initiation. We wished to test this hypothesis by optogenetic activation of either D1R+ - or Adora+ vTS neurons, at times when naive mice (that had not undergone fear learning) are engaged in regular exploratory behavior. For this, the channelrhodopsin variant Chronos (Klapoetke et al., 2014) was expressed bi-laterally and Cre-dependently in D1R+, or Adora+ neurons of the vTS, using AAV1:hSyn:FLEX:Chronos-eGFP in the respective Cre-mouse line (Figure 4A; Figure 4—figure supplements 1 and 2). Three to 4 weeks later, the mice were allowed to explore the fear conditioning chamber, and trains of optogenetic stimuli were applied at pre-determined intervals, irrespective of whether the mice moved or paused from movement (pulse duration 1ms, repeated at 25 Hz for 2 s; each train given six times for each mouse). Optogenetic stimulation did not lead to changes in the movement activity of the Adora2aCre mice, nor of the Drd1aCre mice (Figure 4B and C; N=8 and 7 mice; p>0.999, W=0, and p=0.837, t6=0.215, Wilcoxon and paired t-test, respectively). In additional experiments with Adora+ neurons, we employed longer light pulses (2 and 5 ms), but we similarly did not observe effects on the movement state of the mice (Figure 4—figure supplements 3 and 4). These experiments suggest that in naive mice, the activity of neither D1R+ - nor of Adora+ neurons is sufficient to modulate the movement activity of the mice.

Figure 4 with 4 supplements see all
Optogenetic stimulation of D1R+ and Adora+ vTS neurons has no direct effect on movement.

(A) Scheme showing the bilateral placement of optic fibers over each vTS, and the injection of an AAV vector driving the Cre-dependent expression of Chronos. (B) Left, Adora2aCre mice expressing Chronos in Adora+ neurons of the vTS were allowed to explore the fear conditioning chamber, and six trains of blue light stimuli (50 pulses of 1ms length, 25 Hz) were applied. The average movement indices for N=8 mice are shown, centered around the time of light stimulation. Right, individual and average movement data for N=8 mice, for 2 s intervals before and during optogenetic stimulation. (C) Analogous experiment to the one in (B), now performed for Drd1aCre mice expressing Chronos in a Cre-dependent manner in the vTS (N=7 mice). Note that optogenetic stimulation of neither Adora+ neurons (B), nor of D1R+ neurons led to notable changes in the movement of mice (see Results for the statistical parameters).

Figure 4—source data 1

Raw data and statistical tests for Figure 4 and its supplements.

https://cdn.elifesciences.org/articles/75703/elife-75703-fig4-data1-v2.xlsx

Footshock-driven activity of D1R+ vTS neurons contributes to fear learning

We found that many D1R+ neurons respond to footshocks, and that following fear learning, these neurons show an increased response to the CS and to movement-ON transitions (Figures 1 and 2). This suggests that footshock responses might drive a plasticity of CS - representation, and of the movement-ON representation in D1R+ vTS neurons. To investigate the role of the footshock-evoked activity of D1R+ vTS neurons in fear learning, we next silenced the activity of these neurons during footshock presentation, and observed the effects of this manipulation on freezing behavior during the training day, and one day later during fear memory recall.

For optogenetic silencing, we expressed the light-sensitive proton pump Archaerhodopsin (Arch; Chow et al., 2010) in a Cre-dependent manner bilaterally in the vTS of Drd1aCre mice, and implanted optic fibers over each vTS (Figure 5A; Figure 5—figure supplement 1). Mice in a control group were injected with an AAV vector driving the expression of eGFP (Materials and methods). Four weeks later, both groups of mice underwent auditory-cued fear learning, and on the training day yellow laser light (561 nm) was applied for 3 s, starting 1 s before each footshock, with the aim to suppress the footshock-driven activity of D1R+ neurons (Figure 5B). We found in ex vivo experiments that yellow light strongly hyperpolarizes Arch-expressing D1R+ and Adora+ vTS neurons, suppresses action potential (AP) firing, and does not evoke rebound APs when the light is switched off (Figure 5—figure supplement 2).

Figure 5 with 2 supplements see all
Optogenetic inhibition of D1R+ vTS neurons during footshock reduces the formation of an auditory-cued fear memory.

(A) Scheme showing the bi-lateral injection in Drd1aCre mice of an AAV vector driving the expression of either Arch or eGFP (for controls), and the placement of optic fibers over each vTS. (B) Behavioral paradigm outlining the 3-day fear learning protocol, and the application of yellow light during the footshock presentation on the training day. (C) Freezing levels during 30 s CS presentations on the habituation day for the control mice (expressing eGFP; black circles), and for Arch-expressing mice (green circles). (D) Time-binned average percent freezing traces for control mice (black data) and Arch-expressing mice (green data; bin width is 10 s in both cases). Light blue boxes indicate the 30 s CS presentation periods. Gray bars (bottom) indicate epochs of ‘no CS’ analysis. Vertical orange lines indicate the times of footshock presentation. (E) Quantification of freezing during the training day for the control mice and Arch-expressing mice (black and green data, respectively). Freezing was averaged during the 30 s CS presentations (filled circles, left) and for the no CS epochs (open circles, right). (F) Time-resolved freezing during the cued retrieval on day 3. Note the increases in freezing driven by each CS presentation in control mice (black trace, average of N=10 eGFP expressing mice), which were smaller in amplitude in Arch-expressing mice (N=12, green trace). Light blue areas and gray bars (bottom) indicate times of CS presentation, and epochs of ‘no CS’ analysis, respectively. (G) Quantification of freezing during the 30 s CS presentation (left) and during the no CS epochs (right). Note the significant reduction of freezing during the CS in the Arch group (see Results for statistical parameters).

Figure 5—source data 1

Raw data and statistical tests for Figure 5 and its supplements.

https://cdn.elifesciences.org/articles/75703/elife-75703-fig5-data1-v2.xlsx

During the habituation day of the fear learning protocol the mice showed little freezing, as expected (Figure 5C). On the training day, during which mice received a 1 s footshock after each tone block, mice in both groups showed increasing levels of freezing throughout the training period, interrupted by low freezing activity immediately following the footshocks, caused by increased shock-evoked running and escape behavior (Figure 5D; CS periods and footshocks are indicated by blue and yellow vertical lines, respectively). To quantify freezing behavior, we averaged the percent freezing during the CS, and during six no-CS epochs (see Figure 5D, light blue vertical lines, and lower gray bars, respectively). This showed that freezing levels were not different between the Arch- and the control group, neither for the CS- nor for the no-CS epochs (Figure 5E, left, for the CS epochs: p=0.338, F1,20 = 0.965; Figure 5E right, for the no-CS epochs: p=0.454, F1,20=0.583; two-way repeated measures - ANOVA). Thus, optogenetic silencing of the footshock-evoked activity of D1R+ vTS neurons did not change the freezing behavior of mice during the training day.

On the third day, we tested for fear memory recall by applying tone stimulation (CS) alone in a different context (see Materials and methods). The time-resolved freezing analysis revealed a gradual increase of freezing levels when mice entered the conditioning chamber (to ~20%; Figure 5F, arrow). This baseline level of freezing has been observed before (see e.g. Cummings and Clem, 2020), and likely represents a residual contextual fear memory, despite the change of the context between the training day and the recall day. The 30 s tone blocks (CS) caused a vigorous increase in freezing in eGFP-expressing control mice, whereas in the Arch-expressing mice, the CS was less efficient in driving freezing (Figure 5F, black- and green average traces, N=10 eGFP- and N=12 Arch-expressing Drd1aCre mice). Averaging and statistical analysis revealed a significant difference in CS-driven freezing between control - and Arch-expressing Drd1aCre mice (Figure 5G, closed data points; p=0.040, F1,20=4.853, two-way repeated measures - ANOVA). On the other hand, freezing during the no-CS epochs was unchanged between the two groups (Figure 5G, open data points; p=0.662, F1,20=0.197 two-way repeated measures - ANOVA). Thus, optogenetic inhibition of D1R+ neurons in the vTS during the footshocks on the training day causes a diminished auditory-cued recall of fear memory 1 day later. These data suggest that footshock-driven activity of D1R+ vTS neurons contributes to auditory-cued fear learning.

Adora+ vTS neurons suppress learned fear in the absence of a CS

We next investigated the role of footshock-driven activity in the Adora+ vTS neurons for auditory-cued fear learning. For this, we silenced the activity of Adora+ vTS neurons during the footshocks presented on the training day (Figure 6A and B; Figure 6—figure supplement 1), in an approach analogous to the one used for the D1R+ neurons. During the habituation day, we observed low freezing as expected (Figure 6C). On the training day, mice in both the eGFP-expressing control group and in the Arch group showed a gradual increase in freezing with successive CS-US pairings, interrupted only by low freezing activity immediately following the footshocks, as in the Drd1aCre mice (Figure 6D). The analysis of freezing during the CS- and no-CS epochs revealed no differences in the freezing levels between the Arch- and the eGFP groups on the training day (Figure 6E; p=0.528, F1, 17 = 0.415 and p=0.312, F1,17 = 1.087, respectively; two-way repeated measures ANOVA). Thus, similar as for the D1R+ neurons, footshock-driven activity in Adora+ vTS neurons is not necessary for the freezing behavior that develops during the training day.

Figure 6 with 1 supplement see all
Optogenetic inhibition reveals a role of Adora+ vTS neurons in suppressing freezing in the absence of a learned CS.

(A) Scheme of the experimental approach, during which Arch (or eGFP, for controls) was expressed Cre-dependently in Adora2aCre mice in the vTS, and optic fibers were placed in the vTS above the injection sites. (B) Behavioral paradigm (upper panel), and scheme of the application of yellow light during the footshock stimulus (US) on the training day (lower panel). (C) Freezing levels during CS presentations on the habituation day for mice expressing eGFP (control group; black data) and for Arch-expressing mice (red data). (D) Time-resolved analysis of freezing for control mice (black trace, average of N=8 mice) and Arch-expressing mice (red trace, N=11 mice). The light blue areas and gray bars (bottom) indicate the time of CS presentation, and the time windows for ‘no CS’ analysis. (E) Quantification of freezing during the CS (left), and during the 30 s ‘no CS’ epochs (right), for control mice (black) and for Arch-expressing mice (red). (F) Time-resolved freezing during the fear memory recall day. The light blue areas and gray bars (bottom) indicate the time of CS presentation, and the analysis window for ‘no CS’ analysis. (G) Average percent of time spent freezing, analyzed during the 30 s CS presentations (left), and during the no CS epochs (right) of the fear memory retrieval day, for both groups of mice. There was a trend towards an enhanced freezing in the Arch group at times when no CS was present (right; p=0.0512; two-way repeated measures ANOVA; see Results for further statistical parameters).

On the fear memory recall day, the dynamics of the freezing behavior differed between the eGFP and the Arch group (Figure 6F and G). While the eGFP-expressing control mice displayed an increased freezing during each CS epoch followed by a relaxation to lower freezing levels, mice in the Arch group showed a delayed relaxation of freezing following the CS epochs (Figure 6F; black and red data, respectively). To quantify these effects, we analyzed the average time spent freezing during the CS and during a late no-CS epoch (Figure 6F, light blue bars, and lower gray bars). This showed a trend towards an increased freezing during the no-CS epochs in the Arch group as compared to eGFP controls, although this difference did not reach statistical significance (Figure 6G, right; p=0.0512, F1,17 = 4.40; two-way repeated measures ANOVA, N=8 and 11 eGFP and Arch mice, respectively). On the other hand, freezing during the CS epochs was unchanged between the Arch and the control group (Figure 6G, left; p=0.624, F1,17 = 0.249; two-way repeated measures ANOVA). Thus, silencing the footshock-driven activity of Adora+ neurons in the vTS induces a trend towards higher freezing during fear memory recall in the absence of the CS. These findings, together with the results obtained from silencing D1R+ vTS neurons (see above), suggest that direct and indirect pathway neurons of the vTS have separate, but functionally synergistic roles in fear learning. In fact, the action of both sub-systems together increases the difference between the strength of a defensive behavior in the presence, and absence of a learned sensory cue.

Brain-wide screening of presynaptic inputs to the vTS

Our in-vivo Ca2+ imaging and optogenetic experiments have revealed a differential role of D1R+ - and Adora+ vTS neurons in auditory-cued fear learning. To start investigating the role of synaptic afferents to vTS neurons in these plasticity processes, we next used monosynaptic retrograde rabies virus tracing to identify the presynaptic neuron pools that provide input to D1R+ and Adora+ neurons in the vTS (Figure 7A; Wickersham et al., 2007; Wall et al., 2013).

Figure 7 with 3 supplements see all
Retrograde transsynaptic tracing of brain-wide inputs to D1R+ and Adora+ vTS neurons.

(A) Scheme of the experimental protocol for rabies-virus mediated transsynaptic tracing. (B) Left, Example confocal images of the injection site in a Drd1aCre mouse, Right, confocal images at a higher magnification taken from the boxed area of the left image. Green, red and blue channels are eGFP-labeled cells expressing the helper viruses (green), rabies-virus expressing cells (red) and DAPI labeling (blue, only shown in the overlay images). Scale bars, 200 µm (left) and 50 µm (right). (C, D) Example widefield epifluorescence images of rabies virus-labeled presynaptic neurons in coronal sections, from a Drd1aCre mouse (C) and an Adora2aCre mouse (D). The abbreviations of brain areas are shown in Figure 7—source data 2. Scalebars, 250 µm. (E, F) Localizations of presynaptic neurons plotted in a 3D brain model, both for Drd1aCre (E) and Adora2aCre (F) mice. Each brain shows the results of two animals (blue and red dots) with one dataset being artificially mirrored on the other hemisphere for each genotype. For orientation, some brain areas are highlighted (S2 – green, InsCtx – yellow, PF – red, BLA – pink). (G, H) Quantification of labeled neurons for each brain area. Upwards and downward data signify cell density and absolute number of cells, quantified for N=2 mice of each genotype. Data from single animals are plotted as filled gray circles and triangles. (I, J) Distribution of presynaptic neurons along the anterior-posterior axis in the insular cortex (I), and in the secondary somatosensory cortex (J). The dashed line indicates the border between the anterior and posterior insular cortex according to Franklin and Paxinos, 2016.

Drd1aCre, or Adora2aCre mice were injected into the vTS with an AAV-helper virus driving the Cre-dependent expression of TVA, eGFP and oG, to render the infected neurons competent for later EnvA-pseudotyped rabies virus uptake (Figure 7A, see Materials and methods for the specific viruses). Three weeks later, a pseudotyped delta-G rabies vector driving the expression of dsRed was injected at the same coordinates; control experiments confirmed the specificity of the Cre-dependent expression of TVA-expressing vector, and the absence of ectopic expression of the rabies vector (Figure 7—figure supplement 1). In this way, starter cells at the injection site in the vTS could be identified by GFP- and dsRed co-labeling (red) (Figure 7B, cells appearing yellow in the overlay). Trans-synaptically labeled neurons outside of the striatum were analyzed based on their expression of dsRed. We placed all presynaptic neurons found in N=2 Drd1aCre and N=2 Adora2aCre mice into brain-wide models, with cells from the two mice arbitrarily positioned on different brain sides (Figure 7E and F; red and blue dots, respectively). In both Drd1aCre and Adora2aCre mice, we found the highest number and density of backlabeled cells in the secondary somatosensory cortex (S2), and in the dorsal (granular and dysgranular) part of the insular cortex (InsCx; see also Figure 7—source data 2 for a list of all brain structures with detected presynaptic neurons; Figure 7C, D, G and H). A sizeable number of back-labelled neurons was also found in various areas of the primary somatosensory cortex (S1), in the globus pallidus externa (GPe), and in the thalamic parafascicular nucleus (PF; Figure 7C, D, G and H; Figure 7—figure supplement 2). The GPe and the PF stood out by having high densities of presynaptic neurons to both D1R- and Adora-MSNs, even though their absolute numbers were not large (Figure 7G and H; black bars). Although the brain-wide distribution of presynaptic input neurons to D1R+ and Adora+ neurons in the vTS was overall similar, we detected differences on a finer scale. Thus, we found that structures known to process auditory- and multimodal sensory information, like the auditory cortex (AUD), the temporal association cortex (TeA), the posterior triangular thalamic nucleus (PoT), the posterior intralaminar thalamic nucleus (PiL) and the basolateral amygdala (BLA), were more strongly back-labeled in Adora2aCre mice than in Drd1aCre mice (Figure 7G and H; Figure 7—figure supplement 3; LeDoux, 2000; Weinberger, 2007; Sacco and Sacchetti, 2010; Dalmay et al., 2019; Barsy et al., 2020). Finally, we analyzed the a-p distribution of neurons in the InsCx and S2 that provide input to both D1R+ - and Adora+ neurons in the vTS (Figure 7I and J). Taken together, rabies-virus mediated circuit tracing shows that the vTS receives its main cortical input from the S2 and the InsCx, followed by primary somatosensory areas. Thalamic areas like PF and VPM, and basal ganglia like GPe, as well as limbic areas like the CeA and BLA especially for the Adora+ neurons, also contain sizeable numbers of neurons that provide input to the vTS.

The posterior insular cortex provides strong excitatory drive to the vTS

Retrograde rabies-virus labeling showed that the InsCx, and the adjacent S2 are the primary cortical input areas to the vTS (Figure 7). We next wished to functionally validate the connection from the InsCx to the vTS, using optogenetically assisted circuit-mapping (Petreanu et al., 2007; Little and Carter, 2013; Gjoni et al., 2018). For this, we focussed on the connection from the posterior InsCx (pInsCx) to the vTS, and injected a viral vector driving the expression of Chronos (Klapoetke et al., 2014), primarily targeting the pInsCx (Figure 8A and B; AAV8:hSyn:Chronos-eGFP, note that some spill-over of virus into the neighboring S2 cannot be excluded). We used Drd1aCre x Rosa26LSL-tdTomato, or Adora2aCre x Rosa26LSL-tdTomato mice, to identify each type of principal neuron by its tdTomato fluorescence in recordings 3 to 6 weeks later. Blue light pulses (1ms) at maximal light intensity evoked robust optogenetically-evoked EPSCs of 8.8±1.6 nA in D1R+ vTS neurons (95% CI: [5.28 nA; 12.2 nA]; n=12 cells), and of 2.5±0.5 nA in Adora+ vTS neurons (95% CI: [1.49 nA; 3.58 nA], n=19 cells); the EPSC amplitudes were significantly different between the two types of principal neurons (p=0.002, U=40, Mann-Whitney test; Figure 8C–E). At both connections, gradually increasing the stimulus light intensity led to a smooth increase of the EPSC amplitude, which shows that many axons from pInsCx neurons converge onto each type of principal neuron in the vTS (Figure 8C and D). At high light intensities, the optogenetically-evoked EPSC amplitudes saturated, suggesting that a maximal number of input axons was activated (Figure 8C and D). The paired-pulse ratio (PPR) did not differ between the two neuron types (Figure 8F; p=0.208, t28=1.288, ; unpaired t-test; 95% CI: [0.637; 0.827] and [0.711; 0.960] for D1R+ and Adora+ neurons). On the other hand, the ratio of direct excitation over feedforward inhibition was significantly smaller in Adora+ vTS neurons as compared to D1R+ neurons (Figure 8—figure supplement 1; p=0.0054, U=31, Mann-Whitney test; 95% CI: [2.19; 4.10] and [1.05; 2.40] for D1R+ and Adora+ neurons). Taken together, optogenetically-assisted circuit mapping shows that both D1R+ and Adora+ neurons of the vTS receive robust excitatory inputs from the pInsCx. This, together with the monosynaptic rabies tracing (Figure 7), identifies the pInsCx as providing an important cortical inputs to the vTS.

Figure 8 with 1 supplement see all
pInsCx provides strong excitatory inputs onto D1R+ - and Adora+ neurons in the vTS .

(A) Experimental scheme of injection of an AAV vector driving the expression of Chronos into the pInsCx, and subsequent slice electrophysiology in the vTS. (B) Left, example fluorescence images of the injection site in the cortex expressing Chonos-eGFP (scalebar, 500 µm); middle, overview brightfield image with the position of the patch pipette (black dotted lines) in the vTS; right, higher magnification images of a recorded example cell (brightfield, top; and tdTomato fluorescence, bottom). (C, D) Left, EPSCs recorded by stimulating with 1ms light pulses of increasing intensities (blue, photodiode-recorded trace), and right, the resulting input-output curve of EPSC amplitude versus light intensity, with data from all recorded cells overlaid. The example cell shown on the left is highlighted in black in the right panel. Data is shown for D1R+ vTS neurons (C, n=12 recordings) and Adora+ vTS neurons (D, n=19 recordings). (E) Quantification of the maximal amplitude of EPSCs (left), and of the EPSC delay (right) measured in D1R+ - and Adora+ vTS neurons (n=12 and n=19 recordings, respectively). (F) Example traces (left, and middle panel) and quantification of PPR of optogenetically evoked EPSCs in D1R+ neurons (green; n=12 recordings) and Adora+ neurons (red; n=19 recordings).

Figure 8—source data 1

Raw data and statistical tests for Figure 8 and its supplements.

https://cdn.elifesciences.org/articles/75703/elife-75703-fig8-data1-v2.xlsx

Fear learning causes opposite plasticity at cortical synapses on D1R+ and Adora+ neurons

The in-vivo imaging data showed that many D1R+ and Adora+ neurons in the vTS robustly respond to footshocks, and that during the course of fear learning, D1R+ neurons increase their responsiveness to the CS and to movement-ON transitions (Figures 13). Furthermore, silencing each vTS neuron population during the footshocks led to characteristic impairments of freezing in the presence, and absence of a learned CS (Figures 5 and 6). These findings suggest that plasticity takes place at synapses that drive D1R+ and Adora+ vTS neurons. We therefore next measured the AMPA/NMDA ratio and PPR following fear learning at the pInsCx-vTS D1R+/Adora+ synapses, to probe for postsynaptic or presynaptic forms of LTP induced by fear learning at each connection (Yin et al., 2009; Shan et al., 2014; Rothwell et al., 2015; Lucas et al., 2016; see Palchaudhuri et al., 2022 for a review).

For this purpose, we used optogenetically-assisted circuit mapping to measure optogenetically evoked EPSCs at each connection. We injected Drd1aCre x Rosa26LSL-tdTomato mice, and in a separate series of experiments Adora2aCre x Rosa26LSL-tdTomato mice, with an AAV vector driving the expression of Chronos in neurons of the pInsCx. Three to 6 weeks later, the mice were subjected to auditory-cued fear learning (Figure 9A). Following the fear-memory recall session on day 3, which was performed to validate that mice had successfully learned the CS, mice were sacrificed, and optogenetically evoked EPSCs were recorded. A control group of mice underwent the same protocols, but no footshocks were applied ("CS only" group; Figure 9A). Optogenetically evoked EPSCs at the pInsCx to D1R+ vTS connection showed a significant increase of the AMPA/NMDA ratio in the CS+US group, as compared to the CS-only group, suggesting that at this connection, a postsynaptic form of LTP had occurred during fear learning (Figure 9B and D; p=0.001, t21=3.833, unpaired t-test; 95% CI: [1.687; 2.764] and [3.128; 4.719] for CS-only and CS+US). The PPR, however, was unchanged in D1R+ neurons (Figure 9C and D; p=0.974, t21=0.033, unpaired t-test; 95% CI: [0.448; 0.765] and [0.429; 0.791] for CS-only and CS+US). Conversely, at the pInsCx to Adora+ vTS connection, the AMPA/NMDA ratio was unchanged (Figure 9E and G; p=0.162, t22=1.445, unpaired t-test; 95% CI: [1.538; 2.896] and [0.946; 2.251] for CS-only and CS+US), but instead, the PPR was increased in the CS+US group as compared to the CS-only group (Figure 9F and G; p=0.001, U=18, Mann-Whitney test; 95% CI: [0.193; 0.582] and [0.574; 0.938] for CS-only and CS+US). The latter finding suggests that fear learning induces a presynaptic form of long-term depression at the pInsCx to Adora+ vTS synapse. Thus, auditory-cued fear learning drives long-term plasticity with opposite outcomes at cortical synapses onto D1R+ and Adora+ neurons in the vTS. These differential forms of long-term plasticity might contribute to the complementary roles of the two types of vTS principal neurons in fear learning.

Fear learning induces long-term plasticity with opposite outcomes at pInsCx synapses onto D1R+ - and Adora+ vTS neurons.

(A) Timeline of the ex vivo optogenetic assessment of plasticity following fear learning. See Results for details. (B - D) Measurements of AMPA/NMDA - ratios and PPR in two experimental groups (‘CS-only’ versus ‘CS+US pairing’) in Drd1aCre x Rosa26LSL-tdTomato mice, to target the recording of direct pathway vTS neurons. (B) NMDA-EPSCs recorded at + 50 mV (top traces), and AMPA-EPSCs recorded at –70 mV (bottom traces), recorded in example D1R+ neurons from each experimental group. Vertical dashed line indicates the time of analysis of NMDA-EPSC. (C) AMPA-EPSCs (- 70 mV) recorded for the measurement of PPR (50ms inter-stimulus interval), shown for one example D1R+ neuron from each experimental group. (D) Quantification of AMPA/NMDA ratios and PPR recorded in each experimental group of Drd1aCre x Rosa26LSL-tdTomato mice (CS only, n=11 recordings; CS+US pairing, n=12). Note the significantly increased AMPA/NMDA ratio in the CS+US pairing group as compared to the CS-only group (P=0.001; unpaired t-test), whereas the PPR was unchanged (P=0.97, unpaired t-test). (E - G) Measurements of AMPA/NMDA - ratios and PPR in two experimental groups (‘CS-only’ versus ‘CS+US pairing’) in Adora2aCre x Rosa26LSL-tdTomato mice. (E) NMDA-EPSCs (top traces), and AMPA-EPSCs (bottom traces) recorded in example Adora+ neurons from each experimental group. (F) Example traces for the measurement of PPR of AMPA-EPSCs at - 70 mV, recorded in example Adora+ neurons of each experimental group. (G) Quantification of AMPA/NMDA ratios, and paired-pulse ratios recorded in each experimental group of the Adora2aCre x Rosa26LSL-tdTomato mice (CS only, n=12 recordings; CS+US pairing, n=12). Note the significantly increased PPR in the CS+US pairing group as compared to the CS-only group (p=0.001; Mann-Whitney test). For further statistical parameters, see Results text.

Discussion

Recent studies in mice have identified the posterior (tail) striatum as an anatomically, and connectionally separate part of the striatum, and evidence suggests that dopamine axons in the tail striatum code for salient sensory events (Hintiryan et al., 2016; Hunnicutt et al., 2016; Menegas et al., 2018; see Valjent and Gangarossa, 2021 for a review). Here, we have investigated the role of the tail striatum in auditory-cued fear learning, in part motivated by earlier studies which showed a convergence of US - and CS signaling in the tail striatum located close to the LA (Romanski et al., 1993). We have targeted the ventral half of the tail striatum close to the amygdala, an area which we refer to as ‘ventral tail striatum’, vTS. This region might include the amygdala-striatum transition zone (‘AStria’; located immediately adjacent to the amygdala), but our targeting was not limited to the AStria.

Using Cre-driver mouse lines to target D1R-expressing and Adora-expressing neurons of the direct and indirect pathways (Gerfen et al., 2013), we studied the role of each type of vTS principal neuron in auditory-cued fear learning. We found that both classes of vTS principal neurons were strongly activated by footshock stimuli, and coded for the movement state of the mice; smaller sub-populations of both neuron types also responded to tones in naive mice. Interestingly, the number of tone-responsive D1R+ neurons, but not of Adora+ neurons, increased during auditory-cued fear learning, not dis-similar to the acquisition of tone responsiveness of amygdalar neurons (Quirk et al., 1995; Amano et al., 2011; Grewe et al., 2017). To investigate the role of footshock-driven activity in vTS principal neuron, which most likely drives plasticity in these neurons, we optogenetically silenced neuronal activity at the time of footshock delivery. In Drd1aCre mice, this manipulation caused decreased auditory-cue driven freezing 1 day later, which leads us to conclude that footshock-driven plasticity in vTS D1R+ neurons contributes to the formation of a cue-driven fear memory. This conclusion is corroborated by our finding that fear learning induces a postsynaptic form of LTP at a major cortical input to D1R+ neurons in the vTS. On the other hand, silencing Adora+ neurons during the footshock led to a trend towards increased freezing in the absence of a learned CS, which suggests that indirect pathway neurons of the vTS have access to learned contextual freezing. Interestingly, the finding that the vTS Adora+ subsystem acts to suppress freezing in the absence of a danger-signaling CS, is reminiscent to aspects of safety learning (Christianson et al., 2012). Furthermore, a previous study has correlated the in-vivo activity of vTS neurons with safety learning, albeit without distinguishing between direct- and indirect pathway neurons (Rogan et al., 2005). Taken together, we find that direct and indirect pathway neurons of the vTS have different roles in fear learning. Both neuron types are expected to act synergistically to increase the difference in defensive behavior in the presence, and absence of a learned sensory cue.

In-vivo Ca2+ imaging also revealed that sizable sub-populations of each type of vTS principal neurons code for movement-onset, and that in D1R+ neurons, these movement-ON responses were increased after fear learning (Figures 1 and 3). Movement-correlated activity has been observed in striatal neurons (Markowitz et al., 2018), as well as in several types of sensory cortices and in a more brain-wide fashion (Niell and Stryker, 2010; Keller et al., 2012; Stringer et al., 2019). At present, the role of the movement-related activity in the vTS remains unknown. Our optogenetic activation experiments did not reveal significant acute effects on the movement - or freezing behavior of mice, in line with previous findings in the posterior striatum (Guo et al., 2018), but different from results in the dorsomedial striatum (Kravitz et al., 2010) - the latter striatal area seems to have a more immediate role in movement control than the tail striatum. Therefore, it is unlikely that the movement-ON driven activity that we found here in D1R+ or Adora+ neurons of the vTS contributes to the control of movement. Rather, one could speculate that the movement-related activity in the vTS represents a signal related to the bodily state of the animal. In this regard, the strong excitatory input from the pInsCx to the vTS which we uncovered here seems relevant (Figures 7 and 8). Recent studies have shown that the pInsCx displays state-dependent signaling relating to physiological needs, or to the aversive state of the animal (Livneh et al., 2017; Gehrlach et al., 2019); it is possible that the activity of the pInsCx is additionally modulated by movement-ON transitions of the animal. Thus, future work might investigate the origins, and functions of the movement-related signals in the vTS.

Previous studies in mice analyzed anterograde tracer experiments from cortex to striatum in a brain-wide fashion; this data has also revealed specific cortical input structures to the tail striatum (Hintiryan et al., 2016; Hunnicutt et al., 2016). We have used rabies-virus-mediated back-labeling from genetically identified neurons in the vTS, to identify presynaptic input neurons to the vTS more focally, and specifically for direct- and indirect pathway neurons (Figure 7; see also Wall et al., 2013 for a similar approach in the more anterior dorsal striatum). Our results confirm previous studies regarding the innervation of the vTS by auditory cortical areas, and by amygdalar structures (Hunnicutt et al., 2016). Beyond this, our data identifies the InsCx and the S2 as major cortical input structures to the vTS, both for D1R+ and Adora+ neurons. We confirmed the functional relevance of the putative connections by optogenetically assisted circuit mapping, which revealed unusually large EPSCs at the input from the pInsCx to the vTS (note that the S2 was not covered in our ex vivo optogenetic experiments). Based on the graded input-output curves when varying the light intensity, these EPSCs likely reflect converging inputs from many pInsCx neurons (see also Litvina and Chen, 2017; Gjoni et al., 2018). The strong inputs from the pInsCx, and presumably from the S2, suggest that in addition to auditory- and visual information (Yamamoto et al., 2012; Guo et al., 2018; Menegas et al., 2018), somatosensory signals are processed in the vTS. For example, it is possible that the pInsCx transmits somatosensory or nociceptive information about a footshock during aversively motivated learning to the vTS; similarly, movement-related signals might be transferred at this connection, as discussed above. Moreover, it was shown that a subarea of the pInsCx processes auditory information (Rodgers et al., 2008; Sawatari et al., 2011; Gogolla et al., 2014), thus tone information might additionally be transmitted at this connection. Indeed, our finding of an enhanced AMPA/NMDA ratio following fear learning at the connection from the pInsCx to D1R+ vTS neurons gives rise to the hypothesis that learned auditory signals, or movement-related signals, are transmitted at this connection, a hypothesis which could be tested in future work.

Subtle, but functionally important differences in the brain areas that project to D1R+ versus Adora+ vTS neurons might determine their different in-vivo responses, and, in conjunction with the different output projections of direct- and indirect pathway neurons, their differential roles during fear learning. Recent work has shown that a separate set of midbrain dopamine neurons innervates the tail striatum, and that these dopamine projections are activated by salient, but not by rewarding stimuli (Menegas et al., 2015; Menegas et al., 2018). Thus, the differential roles of vTS D1R+ versus Adora+ neurons in fear learning are likely additionally shaped by the differential effects of dopamine on the two types of striatal principal neurons (Gerfen et al., 1990; Tritsch and Sabatini, 2012; Hjorth et al., 2020). Classical studies showed that dopamine release in the amygdala contributes to fear learning (Lamont and Kokkinidis, 1998; Guarraci et al., 1999; Nader and LeDoux, 1999), and a recent study showed that footshock-driven activity of ventral tegmental area (VTA) dopamine neurons that project to the basal amygdala (BA), contributes to fear learning (Tang et al., 2020). Therefore, dopamine signaling in the vTS, likely from a different source than dopamine release in the BA (Menegas et al., 2015; Tang et al., 2020), might contribute to diversifying the roles of vTS direct- and indirect pathway neurons in fear learning.

Fear learning shows sex-specific differences in rats (Maren et al., 1994; Pryce et al., 1999; Gruene et al., 2015; see Lebron-Milad and Milad, 2012 for review). Thus, it was not possible to investigate female and male mice without differentiating between them; rather, we limited our study to male mice (see Materials and methods). In rodents, males in general show higher freezing than females, whereas females show a higher degree of fast bouts of escape reactions (called ‘darting’), which can also take the form of learned responses (Gruene et al., 2015). Future work might investigate whether the direct- or indirect pathway neurons in the vTS contribute to sex-specific defensive behaviors like darting in females, and whether sex-specific differences exist in the circuit wiring and plasticity mechanisms of the vTS.

Conclusions and outlook

In summary, we find that direct and indirect pathway neurons differentially modulate the degree of learned defensive behaviors in the presence, and absence of learned sensory cues. This uncovers a role of the vTS in balancing cue-specific reaction on the one hand, with a more generalized fear response on the other hand. Adaptive fear discrimination is critically important for animal survival, but is dys-regulated in anxiety disorders (Sangha et al., 2020). Indeed, it was found that during fear generalization, PTSD patients exhibited stronger responses in the striatum, amygdala and insular cortex amongst other areas (Morey et al., 2020). Because fear expression in the absence of aversive sensory cues is downregulated by Adora+ neurons in the vTS, it is an intriguing possibility to harness the differential pharmacology of direct and indirect pathway neurons together with in vivo manipulations of plasticity (Creed et al., 2015), in an attempt to mitigate the effects of fear generalization.

Materials and methods

Key resources table
Reagent type (species) or resourceDesignationSource or referenceIdentifiersAdditional information
Strain, strain background (M. musculus)C57BL/6JThe Jackson laboratoryJax:000664
Genetic reagent (M. musculus)Drd1aCrePMID:17855595; PMID:24360541MGI:4366803; MMRRC:030778-UCD; RRID:MMRRC_030778-UCDSTOCK Tg(Drd1-cre)EY217
Gsat/Mmucd
Genetic reagent (M. musculus)Adora2aCrePMID:17855595; PMID:24360541MGI:4361654; MMRRC:036158-UCD; RRID:MMRRC_036158-UCDB6.FVB(Cg)-Tg(Adora2a-
cre)KG139Gsat/Mmucd
Genetic reagent (M. musculus)Rosa26LSL-tdTomatoPMID:20023653Jax:007909; MGI:3809523; RRID:IMSR_JAX:007909B6.Cg-Gt(ROSA)
26Sortm9(CAG-tdTomato)Hze/J
Recombinant DNA reagentAAV1:hSyn:FLEX:
GCaMP6m (viral vector)
Viral vector facility, University of ZürichZurichVVF:v290-1(6.8e12/ml)
Recombinant DNA reagentAAV1:hSyn:FLEX:
Chronos-eGFP (viral vector)
University of North Carolina vector coreUNC:Boyden-AAV-Syn-Chronos-GFP(2.8e12/ml)
Recombinant DNA reagentAAV1:CAG:FLEX:
Arch-eGFP (viral vector)
University of North Carolina vector coreUNC:Boyden-AAV-CAG-FLEX-Arch-GFP(2.05e12/ml)
Recombinant DNA reagentAAV1:CBA:FLEX:
Arch-eGFP (viral vector)
University of Pennsylvania vector coreUPenn:AV-1-PV2432;Addgene:22222-AAV1(5.48e12/ml)
Recombinant DNA reagentAAV1:CAG:FLEX:
eGFP (viral vector)
University of North Carolina vector coreUNC:Boyden-AAV-CAG-FLEX-GFP(4.4e12/ml)
Recombinant DNA reagentAAV8:hSyn:Chronos-
eGFP (viral vector)
University of North Carolina vector coreUNC:Boyden-AAV-Syn-Chronos-GFP(6.5e12/ml)
Recombinant DNA reagentAAV8:CAG:FLEX:
tdTomato (viral vector)
University of North Carolina vector coreUNC:Boyden-AAV-CAG-FLEX-tdTomato(6.5e12/ml)
Recombinant DNA reagentAAV1:hSyn:FLEX:TVA-2a-eGFP-2a-oG (viral vector)Viral vector facility, University of ZürichZurichVVF:v243-1; Addgene:85225(5.3e12/ml)
Recombinant DNA reagentEnvA:deltaG–RV:
dsRed (viral vector)
PMID:21867879Addgene:32638
Commercial assay or kitFluoroshield mounting
medium with DAPI
Sigma AldrichSigma:F6057-20ML
Chemical compound, druggabazineAbcamAbcam:ab120042;
Sigma:SR95531
Software, algorithmVideoFreezeMed Associates IncMed Associates:VideoFreeze
Software, algorithmEthoVision XTNoldus Information
Technologies
Noldus:EthoVisionXT13; RRID:SCR_000441version 13
Software, algorithmezTrackPMID:31882950RRID:SCR_021496https://github.com/denisecailab/ezTrack
Software, algorithmIgor ProWavemetrics IncRRID:SCR_000325version 7.08, 64 bit
Software, algorithmInscopix Data Processing SoftwareInscopix IncInscopix:IDPS
Software, algorithmCaImAnPMID:30652683RRID:SCR_021152https://caiman.readthedocs.io/
Software, algorithmSHARP-Trackdoi:10.1101/447995https://github.com/cortex-lab/allenCCF
Software, algorithmBrainrenderPMID:33739286RRID:SCR_022328https://edspace.american.edu/openbehavior/project/brainrender/
Software, algorithmABBAdoi:10.3389/fcomp.2021.780026BIOP:ABBAhttps://github.com/BIOP/ijp-imagetoatlas
Software, algorithmFIJIPMID:22743772RRID:SCR_002285http://fiji.sc
Software, algorithmAdobe IllustratorAdobe CorporationRRID:SCR_010279http://www.adobe.com/products/illustrator.html
Software, algorithmGraphPad PrismGraphPad SoftwareRRID:SCR_002798version 9
Software, algorithmNeuroMaticPMID:29670519RRID:SCR_004186plugin for IgorPro
Othersteretotaxic frame for
small animals
David Kopf InstrumentsDavid Kopf Instruments:
Model 942
used with Model
921 mouse adapter
Otherhydraulic one-axis
manipulator
NarishigeMO-10for virus injections;
see Materials and methods
Other600 µm / 7.3 mm
ProView(TM) GRIN lens
Inscopix IncInscopix:1050–004413used with nVista3.0
system
OthernVista imaging systemInscopix IncInscopix:nVista3.0; RRID:SCR_017407for Ca2+ imaging of neurons
in freely moving mice; see
Materials and methods
Otheroptic fiber implantsThorlabs IncThorlabs:FT200EMT200 µm core / 0.39 NA /
230 µm outer diameter
Otherceramic ferruleThorlabs IncThorlabs:CFLC230230 µm bore /
1.25 mm outer diameter
Otherblue light curing dental cementIvoclar Vivadent AGIvoclar Vivadent:Tetric EvoFlowfor securing implants at
the skull surface
Otherlight curing adhesiveKulzer GmbHKulzer:iBond Total Etchfor treatment of skull before
application of dental cement
Otherfear conditioning apparatusMed Associates IncMed Associates:MED-
VFC-OPTO-M
see Materials and methods
Otherelectric footshock stimulatorMed Associates IncMed Associates:ENV-414Sused within the fear
conditioning apparatus
Other561 nm solid-state laserChangchun New Industries Optoelectronics
Technology (CNI)
CNI:MGL-FN-561-AOMfiber coupled, maximum
output 100 mW; for in-
vivo activation of Arch
Other473 nm solid-state laserChangchun New Industries Optoelectronics
Technology (CNI)
CNI:MBL-FN-473–150 mWfiber coupled, maximum
output 150 mW; for in-vivo
activation of Chronos, see
Materials and methods
Othervibrating microtome VT1200SLeica MicrosystemsRRID:SCR_020243for preparation of brain slices;
see Materials and methods
Otherpatch-clamp amplifier
EPC10/2
HEKA ElektronikRRID:SCR_018399for whole-cell patch-clamp
recordings; see Materials
and methods
Otherfluorescent microscope BX51WIOlympusRRID:SCR_018949to visualize neurons for
whole-cell patch-clamp;
see Materials and methods
Otherhigh-power LED, blueCree IncCree:XPEBRY-L1-0000-00P02460 nm; to excite Chronos
in slices; see Materials and methods
Otherhigh-power LED, greenCree IncCree:XPEBGR-L1-0000-00D02530 nm; to activate Arch
in slices; see Materials and methods
OtherLED driverMightex SystemsMightex Systems:BLS-1000–2
Othersilicone photodetectorThorlabs IncThorlabs:DET36A/Mto measure the time-course of
LED light pulse in slice
experiments; see
Figure 8C and D
and Materials and methods
Otherslide scanning fluorescent microscopeOlympusOlympus:VS120-L100; RRID:SCR_018411for imaging post-hoc histology
sections; see Materials and methods
Othersliding microtome
Microm HM450
ThermoFisher ScientificRRID:SCR_015959to prepare histological brain
sections; see Materials and methods
Otherconfocal microscopeLeica SP8RRID:SCR_018169for imaging post-hoc histology
sections; see Materials and methods

Animals

The experiments were performed with different lines of genetically modified mice (Mus musculus) of male sex. The rationale for investigating exclusively male mice was as follows. The aims of the study were to investigate with optogenetic methods whether the vTS has a role in fear learning; to image the in-vivo responses of D1R+ and Adora+ neurons of the vTS during fear learning; to identify the main cortical inputs to both types of vTS principal neurons; and to study signs of long-term plasticity at cortical input synapses to both D1R+ and Adora+ neurons after fear learning. It has been shown, mainly using rats, that sex-specific differences exist in the strength and types of learned defensive behaviors (Maren et al., 1994; Pryce et al., 1999; Gruene et al., 2015). Therefore, including mice of both sexes in the study without differentiating between them would have most likely increased the variability of the results. Thus, it would have been necessary to include male and female mice in separate groups, which would have doubled the number of experimental groups, and experimental animals used in the study. We therefore decided to perform the initial study in male mice (see also Discussion).

The experiments were performed under authorizations for animal experimentation by the veterinary office of the Canton of Vaud, Switzerland (authorizations VD3274 and VD3518). The following mouse lines were used: (1) Drd1aCre STOCK-Tg(Drd1-cre)EY217Gsat/Mmucd; see Gong et al., 2007; MMRRC: 030778-UCD; (2) Adora2aCre B6.FVB(Cg)-Tg(Adora2a-cre)KG139Gsat/Mmucd; see Gerfen et al., 2013; MMRRC: 036158-UCD; (3) Cre-dependent tdTomato reporter line, Rosa26LSL-tdTomato (B6.Cg-Gt(ROSA)26Sortm9(CAG-tdTomato)Hze/J; JAX stock #007909; also called ‘Ai9’; Madisen et al., 2010). All mice strains were back-crossed for at least five generations to a C57BL/6J background. Mice were weaned at 21 days postnatally (P21), and groups of male mice were housed together under a 12/12 hr light/dark cycle (7:00 am, light on), with food and water ad libitum. Surgery was performed at P42 - P56; mice were separated into single cages one day before surgery. For behavioral experiments (Figures 5 and 6), mice from 1 to 2 litters were randomly assigned to control - (GFP-expressing) or effect group (Arch- expressing). Behavioral testing was performed during the light cycle.

Viral vectors and injection coordinates

Request a detailed protocol

For in-vivo Ca2+-imaging experiments (Figures 13), we injected AAV1:hSyn:FLEX:GCaMP6m (200 nl; 6.8x1012 vg[vector genomes]/ml; cat. v290-1; viral vector facility University of Zürich, Switzerland), into the left vTS of Drd1aCre or Adora2aCre mice, using the following coordinates: medio-lateral (ML) 3.2 mm; anterior-posterior (AP) –0.8 to –1.0 mm; dorso-ventral (DV) –4.3 mm (from bregma). For the optogenetic activation experiments (Figure 4), we injected AAV1:hSyn:FLEX:Chronos-eGFP (200 nl; 2.80x1012 vg/ml; University of North Carolina - UNC vector core, Chapel Hill, NC, USA) bi-laterally into the vTS of Drd1aCre or Adora2aCre mice. For the optogenetic inhibition experiments (Figures 5 and 6), we injected AAV1:CAG:FLEX:Arch-eGFP (200 nl; 2.05x1012 vg/ml; UNC vector core) bi-laterally into the vTS of Drd1aCre or Adora2aCre mice. In earlier experiments AAV1:CBA:FLEX:Arch-eGFP (200 nl;5.48x1012 vg/ml; AV-1-PV2432; University of Pennsylvania vector core, Philadelphia, PA, USA, now at Addgene, 22222 - AAV1) was used; both Arch constructs correspond to the initially described ‘Arch’ (Chow et al., 2010). Mice in the control group received AAV1:CAG:FLEX:eGFP (200 nl; 4.4x1012 vg/ml; UNC vector core).

For the ex-vivo optogenetically-assisted circuit-mapping experiments (Figures 8 and 9), Drd1aCre x Rosa26LSL-tdTomato mice, or Adora2aCre x Rosa26LSL-tdTomato mice were injected unilaterally into the pInsCx with AAV8:hSyn:Chronos-eGFP (200 nl; 6.5x1012 vg/ml; UNC vector core), at the following stereotaxic coordinates: ML 4.2 mm; AP –0.55 mm, DV –3.8 mm (from bregma). In some experiments, Drd1aCre- or Adora2aCre mice were used, and an AAV8:CAG:FLEX:tdTomato (200 nl; 6.5x1012 vg/ml; UNC vector core) was additionally injected into the vTS (coordinates as above), to visualize Drd1aCre- or Adora2aCre-positive neurons for subsequent patch-clamp recordings.

Surgery for virus injection, GRIN lens - , and optical fiber implantation

Request a detailed protocol

The surgery procedures for stereotactic injection of viral vectors alone, or combined with fiber implantation were as described in Tang et al., 2020. In short, a mouse was anesthetized with isoflurane (induction with 3%, maintained at 1%) and the head was fixed in a Model 940 stereotactic injection frame (David Kopf Instruments, Tujunga, CA, USA) using non-rupture ear bars (Zygoma Ear cups, Kopf Instruments Model 921). Local anesthesia was applied subcutaneously using a mix (50 µl) of lidocaine (1 mg/ml), bupivacaine (1.25 mg/ml) and epinephrine (0.625 µg/ml). After the skull was exposed, small craniotomies were drilled above the indicated coordinates for injection and fiber insertion. For fiber implantation, an additional craniotomy was made close to Lambda to insert an anchoring micro screw. Virus suspension was injected using pulled glass pipettes and an oil hydraulic micromanipulator (MO-10, Narishige, Tokyo, Japan) with an injection speed of ~60 nl/min. For in-vivo Ca2+-imaging (Figures 13), a GRIN lens (600 µm / 7.3 mm ProView integrated lenses; cat. 1050–004413; Inscopix Inc, Palo Alto, CA, USA) was implanted 350 µm above the virus injection site. Prior to GRIN lens implantation, a 25 G medical injection needle was slowly inserted and retracted to facilitate the later insertion of the blunt-ended lens. To reduce deformation of brain tissue due to continuous vertical pressure, we lowered the GRIN lens with alternating down (150 µm) and up (50 µm) movements until the last 200 µm before the final position. For in-vivo optogenetic experiments (Figures 46), optical fiber implants were implanted bilaterally above the virus injection sites. The fiber implants were custom-made of a 200 µm core / 0.39 NA / 230 µm outer diameter optic fiber (FT200EMT; Thorlabs Inc, Newton, NJ, USA) secured inside 1.25 mm outer diameter ceramic ferrules (CFLC230; Thorlabs) as described in Sparta et al., 2012. The surface of skull, and the GRIN lens (or optical fiber implants) were treated with a light curing adhesive iBond Total Etch (Kulzer GmbH, Hanau, Germany). Blue light curing dental cement (Tetric EvoFlow, Ivoclar Vivadent, Schaan, Liechtenstein) was then applied to the skull to hold the GRIN lens (or optical fiber implants) in place. The open end of the GRIN lens at the integrated docking platform was sealed using a Kwik-Sil silicone compound (World Precision Instruments, Sarasota, FL, USA). After stitching, the skin was covered with Bepanthen Plus cream (Bayer AG, Leverkusen, Germany), the drinking water was supplemented with 1 mg/ml paracetamol and the animals were monitored for the following 6 days to ensure proper post-surgical recovery.

Behavior

Request a detailed protocol

Auditory-cued fear learning was tested three to four weeks after the surgery. Mice were handled by the experimenter and habituated to the procedure of attaching a dummy miniature-microscope (Inscopix Inc) to the GRIN lens platform (in case of Ca2+-imaging experiments), or optical patch cords (in case of optogenetic experiments), for 10–15 min on five consecutive days. An auditory-cued fear learning paradigm was performed in a conditioning chamber of a Video Fear Conditioning Optogenetics Package for Mouse (MED-VFC-OPTO-M, Med Associates Inc, Fairfax, VT, USA) under control of VideoFreeze software (Med Associates Inc). On day 1 (habituation), a mouse at a time was connected to the nVista3.0 mini-microscope (Inscopix Inc) or to the optic fiber patch cords, and the animal was placed in the conditioning chamber. The latter was a rectangular chamber with a metal grid floor, cleaned with 70% ethanol. During the ensuing habituation session, six tone blocks (CS), each consisting of 30 tone beeps (7 kHz, 80 dB, 100 ms duration, repeated at 1 Hz for 30 s), were applied 90 s apart. During a training session on day 2, the mouse was placed in the same chamber and presented with six CS blocks pseudo-randomly spaced 60–120 s apart, each followed by a 1 s foot shock US (0.6 mA, AC) delivered by a stimulator (ENV-414S, Med Associates Inc). During a fear memory recall session on day 3, the mouse was placed in a conditioning chamber within a different context, consisting of a curved wall and a smooth acrylic floor, cleaned with perfumed general-purpose soap, and four CS blocks were applied.

For optogenetic silencing experiments with Arch (Figures 5 and 6), light was delivered during 3 s starting 1 s before the footshock via 200 μm core / 0.22 NA optic fiber patch cords (Doric Lenses, Canada) from a 561 nm solid state laser (MGL-FN-561-AOM, 100 mW, Changchun New Industries Optoelectronics Technology, Changchun, China). The laser was equipped with an AOM and an additional mechanical shutter (SHB05T; Thorlabs). For optogenetic activation experiments with Chronos (Figure 4), 1 ms light pulses were delivered at 25 Hz, 2 s duration from a 473 nm solid-state laser (MBL-FN-473–150 mW, Changchun New Industries Optoelectronics Technology). The intensity of each laser was adjusted before the experiment to deliver 10 mW light power at the fiber tip.

The behavior of animals was recorded at video rate (30 Hz) by the VideoFreeze software (Med Associates Inc). Based on the behavioral videos, a movement trace was generated using ezTrack software (Pennington et al., 2019; see e.g. Figure 1E–G, red trace). The experimenter, and the person analyzing the data were blinded to the assignment of each mouse to the control - or test group. The movement index trace from ezTrack was used to compute a binary freezing trace using custom procedures in IgorPro 7 (WaveMetrics Inc, Lake Oswego, OR, USA). The animal was considered to be immobile (freezing state) if the movement index was below a threshold of 40 arbitrary units (without cable attachment) or 120 a.u. (with cable attachment), for a minimum duration of 0.5 s. The binned trace of percent time spent freezing (10 s bin size; see e.g. Figure 5D and F; Figure 6D and F) was calculated as a time-average of the freezing state from the binary trace, and then averaged across mice in each group.

Microendoscopic Ca2+-imaging data acquisition and analysis

Request a detailed protocol

We used the nVista 3.0 system (Inscopix Inc) for imaging the activity of neurons expressing GCaMP6m in the vTS over the three-day fear conditioning paradigm. Fluorescent images were acquired at 30 Hz sequentially from three focal planes, resulting in an effective sampling rate of 10 Hz per plane. The intensity of the excitation LED in the nVista3.0 miniature microscope was set to 1–1.5 mW/mm2, and the gain was adjusted to achieve pixel values within the dynamic range of the camera. The TTL pulses delivered from the behavioral setup (Med Associates) were digitized by the nVista 3.0 system to obtain synchronization between the mouse behavior and Ca2+-imaging data.

The initial processing of in-vivo Ca2+-imaging data was done using the Inscopix Data Processing Software (IDPS; Inscopix Inc). This included: (1) deinterleaving of the videos into the frames taken at individual focal planes; (2) spatial filtering; (3) motion correction; (4) export of the processed videos as TIFF image stacks; (5) export of the timestamps for each acquired frame, and for the experimental events such as CS and US timing. Next, the TIFF stacks were processed (except a movement correction step that was done by IDPS) using a Python-based package CaImAn (Giovannucci et al., 2019). The package is specifically optimized for the analysis of wide-field microendoscopic fluorescent Ca2+-imaging data using CNMF-E, an adaptation of constrained nonnegative matrix factorization algorithm (Zhou et al., 2018). Detection of ROIs for each focal plane was performed by CaImAn in an unsupervised manner, using the same set of analysis parameters across the three experimental days. This resulted in a set of background-corrected fluorescence traces for each automatically detected neuron, and also included a deconvolution step (Pnevmatikakis et al., 2016). The deconvolution returned the times, and amplitude values of Ca2+ events (see Figure 1T, gray vertical bars). The amplitudes of the deconvolved events were cumulated for each imaged neuron and each time epoch, to derive the amplitude-weighted frequency of Ca2+ events in one of four defined time epochs (during freezing, and in the absence of a CS [Frz_noCS]; during movement, and in the absence of a CS [Mov_noCS]; during freezing, and in the presence of a 30 s CS block [Frz_CS]; and during movement, and in the presence of a 30 s CS block [Mov_CS], e.g. see Figure 1—figure supplement 5).

Following the CaImAn analysis, the data were analyzed using custom routines in IgorPro 7 (WaveMetrics) as follows. Fluorescence intensity traces for each i-th cell (Fit), were standardized by calculating Z-score traces as Zit=Fit-FitσFit , where Fit and σFit are the mean and the standard deviation of the fluorescence intensity, respectively, calculated from the whole trace. Accordingly, the deconvolution traces were also normalized by the standard deviation of the respective fluorescent trace σFit .

Prior to further analysis, any duplicate cells arising from different focal planes were identified with a semi-automated routine. In brief, the candidate duplicate cells were automatically short-listed based on the lateral proximity of their centers (<20 μm lateral distance) and high temporal cross-correlation coefficient (>0.7) between their Z-score traces. Rejection of the duplicate cells featuring lower intensity signal (i.e. the cells more out of focus than the other) was validated manually.

The Z-score traces from retained cells were analyzed by their temporal alignments to the onsets of CS, US and movement-ON events as described in the Results (Figures 1 and 3 and their supplements). Neurons were classified as ‘responders’ to a given event if the time-averaged Z-score value in the relevant time range after the event onset exceeded the chosen threshold. For US events, the range was 0–1 s and the Z-score threshold was 1.0. For the CS and movement-ON events, the range was 0–0.5 s and the threshold set to 0.2.

Rabies tracing

Request a detailed protocol

For rabies tracing experiments (Figure 7) Drd1aCre or Adora2aCre mice were injected into the vTS (see above for coordinates) with a tricistronic vector; AAV1:hSyn:FLEX:TVA-2a-eGFP-2a-oG (250 nl; 5.3x1012 vg/ml; cat. v243-1; viral vector facility University of Zürich) to render cells competent for EnvA-pseudotyped rabies virus uptake (Wickersham et al., 2007; Wall et al., 2013). In earlier experiments, we used a mix of AAVs (AAV8:hSyn:FLEX:TVA-2a-oG and AAV8:EF1α:FLEX:H2B-GFP-2a-oG; 250 nl; 1:1) for the same purpose. Three weeks later, the rabies vector EnvA:deltaG–RV-dsRed was injected at the same coordinates (250 nl; viral vector core Salk Institute for Biological Studies, La Jolla, CA, USA; Osakada et al., 2011). The animals were sacrificed 7 days later, and a histological analysis was performed on every second 40-µm-thick coronal section of the entire brain, from the level of the prefrontal cortex up to the end of the cerebellum. The resulting images were analysed in a semi-automatic fashion, that is dsRed-positive neurons were marked manually, brain sections were registered to the Allen brain atlas and marked neurons were automatically mapped onto the resulting brain regions using Matlab-based software (SHARP-Track; Shamash et al., 2018). The positioning of all long-range projecting cells was analysed and plotted on a 3D brain model using the python-based Brainrender software (Claudi et al., 2020).

Electrophysiology

Request a detailed protocol

For whole-cell patch-clamp electrophysiology in slices, mice that had undergone surgery for AAV vector injection (see above) were sacrificed 3–6 weeks later. Mice were deeply anesthetized with isoflurane, and decapitated. The brain was quickly removed from the skull and placed in ice-cold preparation solution; the subsequent procedures followed the general method of Ting et al., 2014. The preparation solution contained (in mM): 110 N-methyl-D-glutamine, 2.5 KCl, 1.2 NaH2PO4, 20 HEPES, 25 Glucose, 5 Na-ascorbate, 2 Thiourea, 3 sodium pyruvate, 10 MgCl2, 0.5 CaCl2, saturated with carbogen gas (O2 95%/CO2 5%), pH of 7.4 adjusted with HCl. Coronal slices (300 μm) containing the vTS were cut using a Leica VT1200S slicer (Leica Microsystems, Wetzlar, Germany). Slices were stored for 7 min at 36 °C in the preparation solution and were then placed in a chamber containing a storage solution, composed of (in mM): 92 NaCl, 2.5 KCl, 30 NaHCO3, 1.2 NaH2PO4, 20 HEPES, 25 glucose, 5 sodium ascorbate, 2 Thiourea, 3 Na-pyruvate, 2 MgCl2 and 2 CaCl2, pH 7.4 at room temperature, saturated with carbogen (Ting et al., 2014). Whole-cell patch-clamp recordings were performed with an extracellular solution containing (in mM): 125 NaCl, 2.5 KCl, 25 NaHCO3, 1.2 NaH2PO4, 25 glucose, 0.4 Na-ascorbate, 3 Myo-Inositol, 2 Na-pyruvate, 1 MgCl2 and 2 CaCl2, pH 7.4, saturated with carbogen gas. The set-up was equipped with an EPC10/2 patch-clamp amplifier (HEKA Elektronik GmbH, Reutlingen, Germany), and an upright microscope (BX51WI; Olympus, Tokyo, Japan) with a 60 x / 0.9 NA water-immersion objective (LUMPlanFl, Olympus).

Patch-clamp experiments (Figures 8 and 9) were performed using a Cs+-based intracellular solution (in mM): 140 Cs+-gluconate, 10 HEPES, 8 TEA-Cl, 5 Na-phosphocreatine, 4 Mg-ATP, 0.3 Na-GTP, 5 EGTA, pH 7.2 adjusted with CsOH. AMPA/NMDA-ratio experiments were done in the presence of 5 µM GABAA receptor antagonist gabazine (SR-95531; Abcam, Cambridge, UK). Experiments to test the archaerhodopsin (Arch) properties (Figure 5—figure supplement 2) were performed using a K+-based solution with (in mM): 8 KCl, 145 K-gluconate, 10 HEPES, 3 Na-phosphocreatine, 4 Mg-ATP, 0.3 Na-GTP, 5 EGTA, pH 7.2 adjusted with KOH. All electrophysiological experiments were conducted at near-physiological temperature (34 °C) using an inlet heater SHM-6, a heated recording chamber RC-26GL/PM-1 and a thermostatic control unit TC-344B (all from Warner Instruments, Holliston, MA, USA). All chemicals, unless indicated, were from Sigma-Aldrich (St. Louis, MO, USA).

For activation of the excitatory and inhibitory opsins in slice experiments, and for visualization of fluorophores in brain slices, high-power LEDs (CREE XP-E2, 460 nm and 530 nm; Cree Inc, Durham, NC, USA) were custom-coupled into the epifluorescence port of the microscope. Illumination was controlled by the EPC 10/2 amplifier DAC board connected to the LED driver (BLS-1000–2, Mightex Systems, Toronto, Canada). Irradiance was measured by a photodiode (DET36A/M, Thorlabs) coupled into the illumination light path, whose readings were calibrated by the light power measured under the 60 x objective using a power-meter model 1918-R equipped with a 818-UV detector (NewPort, Irvine, CA, USA). Electrophysiological recordings were analyzed in IgorPro (WaveMetrics) using the NeuroMatic plug-in (Rothman and Silver, 2018).

Histology

Request a detailed protocol

For anatomical analysis of optic fiber- or GRIN lens positions, mice were transcardially perfused with a 4% paraformaldehyde (PFA) solution. The brains were post-fixed in PFA overnight and then transferred to 30% sucrose in phosphate-buffered solution for dehydration. Coronal brain sections of 40 µm thickness were prepared using a HM450 sliding microtome (Thermo Fisher Scientific, Waltham, MA, USA). Slices were mounted on Superfrost Plus slides (Thermo Fisher Scientific) and embedded in Fluoroshield mounting medium containing DAPI (Sigma-Aldrich) to stain cell nuclei. Slices were imaged with a slide scanning fluorescent microscope VS120-L100 (Olympus) with a 10 x /0.4 NA objective, or with a confocal microscope (Leica SP8). Brain atlas overlays are taken from Franklin and Paxinos, 2016 and were fit to the brain section image using scaling and rotations in Adobe Illustrator (Adobe, San Jose, CA, USA). Where indicated, registration of brain section images was performed onto the Allen Brain Atlas using an open-source ABBA alignment tool for FIJI (https://github.com/BIOP/ijp-imagetoatlas), developed at the Bioimaging and Optics Platform (BIOP) at EPFL (Chiaruttini et al., 2022). The majority of brain structure names and their abbreviations follows Franklin and Paxinos, 2016; their correspondence to the Allen brain atlas are given in Figure 7—source data 2.

Statistical analysis

Request a detailed protocol

Statistical analysis was performed in GraphPad Prism 9 (GraphPad, San Diego, CA, USA). Before choosing the main statistical test, the distribution of the data was tested for normality using a Shapiro-Wilk test. When normality was confirmed, we used a paired or unpaired version of the two-tailed Student’s t-tests for two-sample datasets, as indicated. For the comparison of relative datasets, we used a one-sample two-tailed t-test. When the data was not normally distributed, we used two-tailed non-parametric tests: a Wilcoxon matched-pairs signed-rank test for paired comparisons, or Mann-Whitney U test for unpaired comparisons of two-sample datasets, as indicated.

For datasets with more than two samples (in-vivo Ca2+ imaging data; Figures 1 and 3), the data showed skewed distributions which did not pass the normality test, therefore we used a non-parametric version of a one-way ANOVA, a Kruskal-Wallis test. If this test detected significant differences, it was followed by Dunn’s post-hoc test for multiple comparisons (called Dunn’s MC test). For datasets influenced by two factors such as the optogenetic silencing/control group and the time of the experiment (Figures 5 and 6), we used a repeated-measures two-way ANOVA (RM-ANOVA) separately for the training and fear retrieval days. If RM-ANOVA reported significance, it was followed by Šidák’s post-hoc tests for the respective factor.

The change in number of neurons responding in-vivo to different events across training (Figure 1L, S, Figure 3I, P) was statistically assessed using a Chi-square test. For the optogenetic silencing experiments, we determined the sample size a priori using a G*Power software (Faul et al., 2007) for an RM-ANOVA test (two groups, 4 replicate measurements for the fear retrieval day) assuming an average change in freezing of 20% (from 60 to 40%) and a standard deviation of 15% (resulting in an effect size of 0.667), significance level α=0.05, power 1-β=0.85, and the correlation between repeated measures of 0.5. The resulting total sample size was N=16 (N=8 mice per group), with a critical Fcrit = 4.6. There was no sample size estimation made for other experiments.

Each specific statistical test is mentioned in the Results text, and the input data along with statistical test summary (such as p-values, values of test statistics, degrees of freedom, etc.) and the main descriptive statistics are given in the statistics Tables for each relevant figure. The data are expressed as mean ± SEM. Statistical significance, if applicable, is indicated in the Figures using asterisks as p≤0.05 (*), p≤0.01 (**) and p≤0.001 (***).

Data availability

The underlying raw data leading to the conclusions of this paper is available at Zenodo data repository https://doi.org/10.5281/zenodo.7530512.

The following data sets were generated
    1. Kintscher M
    2. Kochubey O
    3. Schneggenburger R
    (2023) Zenodo
    A striatal circuit balances learned fear in the presence and absence of sensory cues.
    https://doi.org/10.5281/zenodo.7530512

References

  1. Book
    1. Franklin KBJ
    2. Paxinos G
    (2016)
    The Mouse Brain in Stereotaxic Coordinates (4th edition)
    Elsevier / Academic Press.

Decision letter

  1. Mario A Penzo
    Reviewing Editor; National Institute of Mental Health, United States
  2. Kate M Wassum
    Senior Editor; University of California, Los Angeles, United States

Our editorial process produces two outputs: (i) public reviews designed to be posted alongside the preprint for the benefit of readers; (ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Decision letter after peer review:

Thank you for submitting your article "A striatal circuit balances learned fear in the presence and absence of sensory cues" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Kate Wassum as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

The individual assessments and recommendations from each of the reviewers are included below. In addition, here we provide you with a list of items that we collectively consider to be essential revisions that must be addressed in order for the manuscript to be considered further for publication in eLife. Please also address the individual points raised by each of the reviewers in the public reviews and recommendations for authors, as we consider that addressing them will strengthen the overall quality of the paper. Our requested Essential Revisions are:

1) Revise the overall presentation of the manuscript to avoid the use of inaccurate wording and/or conclusions not supported by the findings presented on the manuscript. Please refer to the individual assessments from the reviewers for specific guidance on this.

2) Appropriately frame the considerations when defining the boundaries of the AStria and how this area is distinguished from the neighboring tail of the striatum. A more comprehensive discussion of anatomical features and connectivity of the AStria is also required. Please refer to comments from Reviewer 1. Maps of expression spread for each subject (e.g., overlaid) and lens or fiber placement would also help provide transparency for readers in the location of the recordings and manipulations.

3) Reanalyze the imaging data in accordance to the comments from Reviewers 2 and 3.

4) Address the reviewers' concerns related to the AP deconvolution method employed by the authors.

5) Include important missing controls as highlighted by the reviewers (e.g., positive control for optogenetic stimulation experiments, negative control for the rabies tracing data).

6) Provide a rationale for why only male subjects were included in the study. Specifically, the authors should: a) state the sex in the abstract, b) explain why only one sex was used in the methods, and c) acknowledge and discuss the limitation of the exclusion of one sex in the discussion.

7) Demonstrate that the Cre lines used are valid as tools to achieve genetic access to the neuronal populations of the AStria.

8) Please ensure your manuscript complies with the eLife policies for statistical reporting: https://reviewer.elifesciences.org/author-guide/full "Report exact p-values wherever possible alongside the summary statistics and 95% confidence intervals. These should be reported for all key questions and not only when the p-value is less than 0.05." This should be reported in the main text.

9) Please include a key resource table in your methods and clarify whether animals were tested in the light or dark cycle.

Reviewer #1 (Recommendations for the authors):

Thalamic (PIN/PIL) and cortical regions (TeA/Auv) are missing or only weakly present when the connectivity of Astr is described. Romanski and Ledoux (1993) showed strong TeA-Astr connectivity, and Ledoux et al. (1990) as well as Barsy et al., (2020) highlighted a strong thalamo-Astr projection from the area of PIL and SG. Furthermore, the latter study also investigated the CS, US and CS/US responsiveness of Astr neurons which were found to be (at least, partially) mediated by the thalamic inputs.

The LA-Astr projection seems to be also underestimated in the view of the earlier publications (eg. Jolkkonen et al., 2001).

One of the reasons causing these connectivity differences could arise from the targeting of Astr. Identification of Astr territory is not trivial, since no marker is available which could differentiate AStr from the rest of the caudal (tail) striatum. Still, as earlier studies showed that these two striatal regions may receive distinct inputs (eg. MGN to the tail, while PIN/PIL to the Astr; LA mostly targets Astr and not the other parts of the caudal striatum), it is important to investigate the same and consistent striatal region. Paxinos mouse brain atlas indicates AStr from bregma ~-0.9 (AP) between LA and LGP (and more caudal, the stria terminalis) and right above the CeA. The dorsal border is uncertain, but mostly indicated ventral to the top 'corner' of the LA (like in a recent review by Valjent and Gangarossa, 2020). However, many representative images and schematic drawing highlight more dorsal and anterior regions to be targeted which resemble the ventral part of the striatal tail, rather than Astr. While most of the foot shock and CS responsive cells were located in the posterior part of the examined regions (more posterior from bregma -1; Figure 1-3), some injection sites (AAV, RV) and fiber optic locations were more anterior. Indeed, the authors also use the tail of the striatum term. As no comprehensive functional study is present (to my best knowledge) which simultaneously investigate the tail of the striatum and Astr, it could cause many discrepancies if the two regions are mixed.

Thus, some of the discrepancies could have arisen from targeting the tail of the striatum rather than Astr.

Several recent (Barsy et al., 2020, Gilad et al., 2020; Talyor et al., 2021) and earlier studies (Bordi and Ledoux et al., 1994a,b; Apergis-Schoute et al., 2005; Han et al., 2008; Weinberger, 2011, etc) suggested that CS-US association and thus, neuronal plasticity can occur prior to the amygdala, in non-sensory (higher-order) thalamic regions, like PIN/PIL. Therefore, the following statement is only partially correct. "In the lateral (LA)- and basal amygdala (LA and BA respectively), auditory (CS) and somatosensory (US) information are integrated.."

Was there any level of dorsoventral topography found in the location of the Cs and/or US responsive cells (striatal tail vs. Astr)?

Is it possible that the examined and responded D1 and Adora+ cells are actually located in two distinct striatal regions? It is discussed that these are rather topographically non-overlapping populations.

Is it possible to follow the Ca-activity of individual cells throughout the entire learning paradigm? If yes, plasticity of a single cell could be analysed (like in the thalamus, Taylor et al., 2021).

As Figure 7 showed some level of topography in the InsCx-Astr connectivity, were different sectors of InsCx targeted in the D1 and Adore-Cre animals used for patch clamp recordings in Figure 8?

Pg 17. Between the AAV and RV, three weeks is indicated in the text, but 4 weeks in the figure.

Figure 8, Figure suppl. 4 B and C. It seems that distinct parts of striatum are targeted.

As there is very little data about Astr, it would be interesting to examine the basic anatomical organization of Astr. For example, the proportion of D1+ and Adora+ cells, the presence of other types of neurons (cholinergic, PV cells, etc). But, these may be far beyond the scope of the present study.

Reviewer #2 (Recommendations for the authors):

It is disappointing that the presentation and some of the analyses do not do justice to the work. Please see my points below. In particular, the heavy handed analyses of the fluorescence transients is too much and fraught with potential for error. More importantly, it obscures what appear to be large differences that could be extracted more easily (and convincingly). Please see the "approach section below". Lastly, at times the authors interpret their data and results too optimistically.

Specifically, the authors should:

1. Reanalyze the data using only z-score F(t).

2. Show more primary data – images of cells, etc…

3. Refrain from reaching conclusions too early in the results especially that it is the contribution of plasticity of the neurons that is being studied in the silencing experiments (as opposed to the need for activity).

4. Improve the presentation of the methods

Please read the detailed comments below.

Abstract:

1. Imprecise wording. The notion of "balancing" is unclear in "defensive behaviors need to be finely balance" which makes it hard to understand exactly what is being studied.

2. "In-vivo optogenetic silencing during the training day showed that plasticity in D1R+ AStria neurons contributes to auditory-cued fear memories"

Please see my comments on the interpretation of the silencing experiments.

Results:

1. Show some images of neurons with corresponding fluorescence transients to understand the quality of the data. I can't find them anywhere in the figures.

2. Line 111 – "Seemed to increase". What does this mean?

3. Line 114 – Claim to have derived APs by deconvolution.

4. Line 127 – "Similarly suggested an increased".

5. See comments on methods – by the time we get to summary data as I Figures IJ-L it is very hard to understand what these mean. They are in au units x 10e-4. What does this mean? If they are au then they can be remapped.

6. Line 154 "These experiments thus show that D1R+ AStria neurons increase their AP firing activity during the onset of movements after a period of immobility. "

The analysis in this section is potentially problematic as tones are being played and the animal is starting and stopping movement. It is not clear that the response is to the movement change as opposed to the tone. Please show the histogram of tone start and stop times relative to the movement onsets.

A better analysis is one that models the contributions tones, movement starts, stops, and US. This can be done with a GLM and will alone one to disambiguate these confounds.

I am especially worried because of the follow-up sentence "Line 155 were more pronounced during the CS than in the absence of a tone" which suggests that these are not movement cells.

7. Line 159 – The conclusion in this sentence cannot be justified based on the data shown so far.

8. Line 168 – One cannot make these conclusions by comparison of event-triggered averaging.

9. Overall Figure 1 is very hard to follow. The number of panels should be reduced, labels should be placed to show which analyses are relative to movement, CS, etc…

10. Figure 2 is very nice.

11. The paragraph starting on line 252 is confusing. Are the authors saying that the # of cells responding goes up (i.e. response fidelity) but that the response per active cell (estimated AP content) does not? Both the positive and negative conclusion refer to Figure 3R. I can't find a definition of response fidelity, which is used several times.

12. The analysis of Figure 3S says that significance was judged by a KS-test, which is usually used for comparisons of unbinned cdfs. The graph is an average +/- error bar. How was the KS test used?

13. If the freezing habituation data in 1S and 3S cannot be analyzed statistically, it should be removed from the panels as one cannot help but interpret the data shown quantitatively.

14. The Venn diagrams in 3V and 2D would benefit by labeling the numbers in the main intersection areas.

15. In figure 4, the baseline movement of the Adora2a and D1R groups is very different. Are the two genotypes equivalent? These are BACs so it would not be too surprising.

16. For Figure 4, what in vivo proof is there that the cells are being activated? Given the negative conclusion, such a control is important.

17. The experiments motivated by the paragraph starting at Line 325 are very nice. However, the paragraph is troubling. The manipulation is to reduce activity and it should be phrased in that way. Instead, a hypothesis based on plasticity is presented, from which the experimental manipulation of activity is indirectly motivated.

18. There are differences in baseline freezing (5F) before the light is turned on that are of similar magnitude (relative to the low level of freezing) as seen in 5H. Similarly, the comparison between no effect in 5G right to less effect in 5I left may not be fair given that the latter is in "steady state" and the former not. Is an RM-ANOVA the right way to go?

19. Similar concerns exist for the data in Figure 6.

Can the differences in Figure 5 or 6 be used to identify individual mice as ones that received silencing and ones that did not? It seems unlikely, especially for Figure H. It might have been better to run cohorts of mice in which the silencing was only applied in the retrieval portion. Otherwise, it is unclear if the relatively modest effects are due to a difference during training that is not reflected in freezing rates.

20. The main text does not reference controls for the rabies experiments which generally should include (1) TVA dependence of rabies infection, (2) G dependence of spread, (3) Cre-dependence of the DIO/floxed constructs. It is important to do these with the same batch of viruses used for the experiments (ideally in parallel!).

21. I hesitate to make too much of the differences shown in 7I/J given N=2, especially without a careful analysis of all the starter cells to show that the number and distribution of these were the same across genotypes.

22. The currents in Figures 8G-H are huge and loss of voltage clamp must have occurred, especially when measuring NMDA receptor currents.

Approach:

1. Large lesions even for optogenetics because of use of 1.25 mm cannula.

2. Line 763 -- "If mice from any experimental group showed low freezing, they were excluded from the analysis; we used a threshold of 20% time spent freezing

during the fifth and sixth CS-blocks on day 2 (7 out of 59 mice for the behavioral experiments were excluded)."

It is good of the authors to give the exclusion criteria. However, this is worrisome. Some of the loss of function experiments are based on the hypothesis that activity is necessary for learning. Therefore, this discards the poor learners, right?

What is the breakdown of exclusion across the different experimental and genetic conditions? Why not simply include all animals to capture the natural variability?

3. The deconvolution approach is worrisome for a few reasons.

a. Ca entry into these neurons is highly non-linear with single AP, burst AP, up-state APs, and GPRC-mediated influx. It is unclear that one can use a linear model of # of APs to [Ca] (to F(Gcamp6m)). It does not seem that this step is necessary (especially given the heavy filtering).

b. Data is acquired at 10Hz but then a band-pass filter at 2-4 Hz (If I read it correctly although the methods are not very clear). This is much less than the Nyquist limit. Why? Also why place a low-frequency limit (i.e. why use a band pass as opposed to a low-pass to simply get rid of shot and electronic noise?).

c. The deconvolution kernel is arbitrary and uses a rise time that cannot be captured with the 2-4 Hz band pass. The authors need to show that 0.5 s tau(decay) is justified. There is no justification for choosing an amplitude of 2.

d. Line 793 – The phrase "deconvolution yielded a… proportional to AP firing rate" must be removed as it cannot be justified.

e. The deconvolved signal (R) is further filtered with a box car – Why? The filtering is done upstream on the fluorescence. It should not be done again.

f. "local peaks" exceeding a threshold (arbitrarily set to 2) were detected using a "first-derivative" method. Why add a threshold here? It seems unjustified. Second, how does one detect a peak with a first-derivative? Shouldn't it be a second derivative? Or are they detecting fast-rate of rise, in which case why set a peak amplitude threshold?

4. All of these concerns are acknowledged later (line 801) in "we did not aim to infer exact AP spiking rates of cells". Then why do this and present it in this way?

5. They should simply get rid of all these steps and analyze the z-score raw F (calculating a DF/F and then z-scoring is no different than just z-scored the Z).

6. y comparing the

7. The section starting on line 809 indicates that different baselines were chosen for comparison for different types of cells. This introduces a circular bias. One needs to be able to statistically show that a cell is a "move-on" or CS responder without changing a baseline. Why not use the Z-score values in the time-bin without picking a base-line – i.e. compare activity in that window to all activity?

8. Line 798 – The amplitudes of the events from deconvolution (referred to as estimated AP content, EAC [a.u./s]) were proportional to the amplitude of the ca2+-transients (e.g. see Figure 1 —figure supplement 1E).

a. I can't find the referenced data in a figure. The sentence suggests a scatter plot with some kind of regression analysis?

b. If this is true, why do it all? Just use the F as suggested above and save all these potentially problematic analysis steps.

Reviewer #3 (Recommendations for the authors):

1. The study used only male mice. There is no basis for excluding females.

2. The authors did not verify specificity of D1R and Adora lines (tdTomato + RNAscope for D1R and Adora). Without this verification, it is unclear if the fluorescent patterns reported actually reflect patterns of D1R and Adora neurons.

3. Inspection of the heat plots shown in figures 1 and 3 indicate the fluorescent data contains repeating 'neurons'. For example in Figure 3, the two signals in rows 10 and 11 are identical. This is particularly notable because the repeating 'neuron' shows the largest change in fluorescence. In addition, rows ~17 and ~34 appear virtually identical as well. This is a major problem and indicates the authors have an issue in their analysis pipeline. Greater care needs to be taken to ensure repeating neuron's are not reported and analyzed.

Same issue occurs in Figure 1. Rows 3 and 4 are identical signals. Repeating 'neurons' appears to be a problem throughout.

4. GCaMP6m is a fluorescent calcium indicator. While calcium entry into neurons is required for AP-driven vesicular release, calcium entry into a neuron can reflect many other processes. In this manuscript, the authors wish to examine single unit activity by acquiring and deconvolving fluorescent signals. In fact, there are known issues relating fluorescent signals to action potentials in the striatum:

https://www.biorxiv.org/content/10.1101/2021.01.20.427525v2.full

If the authors wish to analyze spiking activity of the striatal neurons, they need to use techniques that directly record action potentials. These techniques are readily available. If the authors wish to record fluorescent changes resulting from calcium influx, they need to embrace this decision and only analyze changes in fluorescence data.

5. I found it very difficult to track the logical flow of the analyses. The authors start off by showing 'neurons' are responsive to cues. But then almost immediately pivot to showing they do not in fact respond to cues, but movement during cues. A lot of figure space and analysis is devoted to this, such that the big picture is lost. I would find it more convincing to develop analyses that directly compare cue vs. movement vs. cue x movement responding from the outset. A regression approach may be useful. Take the 30 s prior to cue and 30 s cue. Have one regressor be cue on vs cue off, another be movement on vs movement off and the last be the intersection of cue on/off and movement on/off. This would simultaneously compare each regressor to each 'neurons' activity pattern, determining which best captures change in fluorescence.

6. Do not show hypothetical behavior data (as in Figure 1A). Only show real behavior data.

In its present form, I do not feel that the authors achieved their aims or that the results support their conclusions. Addressing the points above is more likely to produce such results. Although even then, it appears that the contribution of these two cell types to tone-shock learning appears limited. If these neurons are contributing to aversive behavior, perhaps their contribution would be better captured by procedures in which movement is a more central element to behavior.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "A striatal circuit balances learned fear in the presence and absence of sensory cues" for further consideration by eLife. Your revised article has been evaluated by Kate Wassum (Senior Editor) and a Reviewing Editor.

The manuscript has been greatly improved during the revision, but there are some issues that remain to be addressed prior to acceptance, as outlined below:

Reviewer #2 (Recommendations for the authors):

The authors have greatly altered and improved the manuscript to address the reviewer concerns. In particular, they have implemented a standardized Ca analysis pipeline, altered the description of the conclusions in many places, improved the presentation of the data and added some necessary controls.

I wish that they had done a full characterization of the RV controls, instead of showing some selected images, but that is ok.

I am confused about the sentence (lines 199-201)

"These experiments show that a substantial sub-population of Adora+ neurons in the vTS codes for movement onset, but this representation was unchanged by fear learning (Figure 3M – P)."

The changes in event frequency in Figure 3R from habituation to training for freeze state are quite large yet the conclusion is that training does not influence the representation of movement in Adora2a+ neurons? Perhaps I am misinterpreting the statement.

https://doi.org/10.7554/eLife.75703.sa1

Author response

Essential revisions:

The individual assessments and recommendations from each of the reviewers are included below. In addition, here we provide you with a list of items that we collectively consider to be essential revisions that must be addressed in order for the manuscript to be considered further for publication in eLife. Please also address the individual points raised by each of the reviewers in the public reviews and recommendations for authors, as we consider that addressing them will strengthen the overall quality of the paper. Our requested Essential Revisions are:

1) Revise the overall presentation of the manuscript to avoid the use of inaccurate wording and/or conclusions not supported by the findings presented on the manuscript. Please refer to the individual assessments from the reviewers for specific guidance on this.

The manuscript has been re-written at many points, to address each point of criticism by the reviewers (see our answers to the individual points raised by each reviewer). We hope that the revision has improved the paper.

2) Appropriately frame the considerations when defining the boundaries of the AStria and how this area is distinguished from the neighboring tail of the striatum. A more comprehensive discussion of anatomical features and connectivity of the AStria is also required. Please refer to comments from Reviewer 1. Maps of expression spread for each subject (e.g., overlaid) and lens or fiber placement would also help provide transparency for readers in the location of the recordings and manipulations.

We agree with the point made by reviewer 1, point [3] about the difficulty in demarcating the AStria / ventral tail striatum (vTS). ("One of the reasons causing these connectivity differences could arise from the targeting of Astr. …"). Because of the additional concern whether the Cre mouse lines used here are well-suited to drive Cre-expression in the AStria / vTS ("Essential revision point #7, below, and reviewer 3, point 2), we performed further anatomical experiments. For this, we crossed both D1RCre and AdoraCre mice with a tdTomato reporter mouse (Ai9), and made coronal sections across the entire extent of the striatum and posteriorly including the amygdala complex (see new data in Figure 1D; Figure 1 —figure supplement 1 for D1RCre, and Figure 3 —figure supplement 1 for AdoraCre). These images show, together with a re-analysis of our fiber positions, that we have targeted neurons in the broader area of the ventral tail striatum (vTS), but not in the more narrowly defined, and more ventrally located "AStria", as suspected by reviewer 1. Therefore, in the revised version, we do not refer any more to the "AStria", but rather, name the brain area studied here the more broadly defined "ventral tail striatum" (vTS).

Furthermore, we have re-written the end of the Introduction to introduce the posterior striatum more carefully (p.4, bottom, p. 5 top), and we now briefly discuss the targeting of the vTS versus the AStria (see Discussion, p. 19, top). Moreover, we now provide the fiber placement images of all mice in two new Supplementary Figures (see Figure 5 —figure supplement 1, and Figure 6 —figure supplement 1). Please also see our answers to the corresponding individual points of the reviewers.

3) Reanalyze the imaging data in accordance to the comments from Reviewers 2 and 3.

The imaging data has been completely re-analyzed, using "Caiman" software and the built-in deconvolution software in Caiman. Please see our responses to the specific points of the reviewers 2 and 3, below.

4) Address the reviewers' concerns related to the AP deconvolution method employed by the authors.

This concern was addressed by completely re-analyzing the Ca-imaging data (see point 3 above), and by switching from our custom-written deconvolution analysis to the built-in deconvolution analysis of Caiman; the latter is based on previously published papers (Pnevmatikakis et al. 2016 Neuron; Giovannucci et al. 2019 eLife). Please see our detailed response to the corresponding individual points of reviewers 2, and 3.

5) Include important missing controls as highlighted by the reviewers (e.g., positive control for optogenetic stimulation experiments, negative control for the rabies tracing data).

Additional experiments have been performed for both points.

First, for the optogenetic stimulation experiments, we have added additional data with longer blue light pulses for the Adora+ neurons. The new experiments further confirm the results in Figure 4, showing that vTS neurons do not directly modulate movement – or freezing of the mice (see new Figure 4 —figure supplement 1).

Second, we have performed additional control experiments for the rabies virus constructs (new Figure 7 —figure supplement 1).

6) Provide a rationale for why only male subjects were included in the study. Specifically, the authors should: a) state the sex in the abstract, b) explain why only one sex was used in the methods, and c) acknowledge and discuss the limitation of the exclusion of one sex in the discussion.

We have now stated the sex of the investigated animals in the abstract, and we have explained the choice of male mice in the Methods (new text in p. 28, top). Furthermore, we have discussed possible sex-specific differences regarding the role of the vTS in fear learning (p. 22 bottom, p. 23 top).

7) Demonstrate that the Cre lines used are valid as tools to achieve genetic access to the neuronal populations of the AStria.

The use of D1RCre and AdoraCre mouse lines, to target direct- and indirect neurons/MSNs of the striatum, has been previously documented in many studies, albeit focusing on the dorsal striatum. For the vTS and including the AStria, a recent study has documented in detail the distribution of D1R+ and D2R+ neurons (Gangarossa et al., 2019; cited in our paper). We have added new anatomical data which shows the spatial distribution of Cre-expressing neurons in the D1RCre mice (Figure 1D; Figure 1 —figure supplement 1), and in the AdoraCre mice (Figure 3 —figure supplement 1) in the vTS and AStria. This data is in line with the previous findings of Gangarossa et al. 2019 regarding the spatial distribution of D1R+ and D2R+ (in our case, Adora+) MSNs in the vTS and AStria. The data furthermore shows that there are very few neurons in adjacent cortical- and amygdala areas that would give rise to Cre-dependent expression in the Cre mouse lines. On the other hand, this data has revealed that Cre – expression in the D1R+ mice does not fully reach the ventrally located AStria (see Figure 1 —figure supplement 1C). This finding supports our decision to now name the targeted brain area ventral tail striatum (vTS) and no longer "AStria" (see above, point 2). Please also note that the new data in Figure 1 —figure supplement 1 and Figure 3 —figure supplement 1 show that D1R+ – and Adora+ neurons project to the entopeduncular nucleus / GPi and to the GPe, respectively, hallmarks of the connectivity of direct- and indirect pathway neurons in the Striatum. Taken together, with these additional experiments, we are confident that the Cre mouse lines used here allow selective access to D1R+ and Adora+ neurons of the direct and indirect pathway in the vTS.

8) Please ensure your manuscript complies with the eLife policies for statistical reporting: https://reviewer.elifesciences.org/author-guide/full "Report exact p-values wherever possible alongside the summary statistics and 95% confidence intervals. These should be reported for all key questions and not only when the p-value is less than 0.05." This should be reported in the main text.

In the revised version (as well as in the first version), we include detailed statistical reporting. We have double-checked with the reporting guidelines of eLife and, to our best knowledge, we comply with them.

9) Please include a key resource table in your methods and clarify whether animals were tested in the light or dark cycle.

A Key Resource Table has now been included. Also, we now state that mice were tested in the light phase (p. 29, top).

Reviewer #1 (Recommendations for the authors):

Thalamic (PIN/PIL) and cortical regions (TeA/Auv) are missing or only weakly present when the connectivity of Astr is described. Romanski and Ledoux (1993) showed strong TeA-Astr connectivity, and Ledoux et al. (1990) as well as Barsy et al., (2020) highlighted a strong thalamo-Astr projection from the area of PIL and SG. Furthermore, the latter study also investigated the CS, US and CS/US responsiveness of Astr neurons which were found to be (at least, partially) mediated by the thalamic inputs.

The reviewer refers to previous studies which showed evidence largely from anterograde labeling studies, for auditory-related inputs to the AStria / vTS, e.g. from the temporal association areas (Romanski and LeDoux 1993) and from higher-order auditory thalamic nuclei (LeDoux et al. 1990 and Barsy et al. 2020). We, on the other hand, have investigated the presynaptic inputs to the vTS in a different approach, using rabies-mediated backlabelling techniques specific for D1R+ – and Adora+ neurons in the vTS. We find many brain areas presynaptic to either D1R+ or to Adora+ vTS neurons (Figure 7G and 7H, respectively). Of note, the temporal association area ("TeA"), and the auditory cortex ("AUD") were also revealed in our dataset (Figure 7G, H).

Furthermore, we have looked again at the data, and found that back-labelling was indeed also present both in the PiL and in the PoT, two thalamic structures located close to the auditory thalamus, and which have been shown to process auditory- and multimodal sensory information (see references in the Results). We have now removed this data from the previous "Others" categories, and have added them separately under "PoT" and "PiL" in Figure 7G, H (see also corresponding changes in the Results text; p. 15 bottom, p. 16 top).

The LA-Astr projection seems to be also underestimated in the view of the earlier publications (eg. Jolkkonen et al., 2001).

Please see our response above for the general reasoning of whether data from the rabies approach is directly comparable with previous studies largely based on anterograde labelling.

More specifically, neurons in the BLA are present in high density in our analysis (see Figure 7G, H; black bars), especially for Adora+ neurons.

One of the reasons causing these connectivity differences could arise from the targeting of Astr. Identification of Astr territory is not trivial, since no marker is available which could differentiate AStr from the rest of the caudal (tail) striatum. Still, as earlier studies showed that these two striatal regions may receive distinct inputs (eg. MGN to the tail, while PIN/PIL to the Astr; LA mostly targets Astr and not the other parts of the caudal striatum), it is important to investigate the same and consistent striatal region. Paxinos mouse brain atlas indicates AStr from bregma ~-0.9 (AP) between LA and LGP (and more caudal, the stria terminalis) and right above the CeA. The dorsal border is uncertain, but mostly indicated ventral to the top 'corner' of the LA (like in a recent review by Valjent and Gangarossa, 2020). However, many representative images and schematic drawing highlight more dorsal and anterior regions to be targeted which resemble the ventral part of the striatal tail, rather than Astr. While most of the foot shock and CS responsive cells were located in the posterior part of the examined regions (more posterior from bregma -1; Figure 1-3), some injection sites (AAV, RV) and fiber optic locations were more anterior. Indeed, the authors also use the tail of the striatum term. As no comprehensive functional study is present (to my best knowledge) which simultaneously investigate the tail of the striatum and Astr, it could cause many discrepancies if the two regions are mixed.

Thus, some of the discrepancies could have arisen from targeting the tail of the striatum rather than Astr.

We agree with the reviewer, who points out the difficulty in demarcating the AStria / ventral tail striatum (vTS). Because of the additional concern whether the Cre mouse lines used here are well-suited to drive Cre-expression in the AStria / vTS ("Essential revision point #7, above, and reviewer 3, point 2), we performed further anatomical experiments. For this, we crossed both D1RCre and AdoraCre mice with a tdTomato reporter mouse (Ai9), and made coronal sections across the entire extent of the striatum and posteriorly including the amygdala complex (see new data in Figure 1D; Figure 1 —figure supplement 1 for D1RCre, and Figure 3 —figure supplement 1 for AdoraCre). These images show, together with a reanalysis of our fiber positions, that we have targeted neurons in the broader area of the ventral tail striatum (vTS), but not in the more narrowly defined, and more ventrally located "AStria", as suspected by the reviewer. Therefore, in the revised version, we do not refer any more to the "AStria", but rather, name the brain area studied here the more broadly defined "ventral tail striatum" (vTS).

We hope that the additional anatomical data, and the re-naming of the brain area targeted in our study (vTS), have solved this issue.

Several recent (Barsy et al., 2020, Gilad et al., 2020; Talyor et al., 2021) and earlier studies (Bordi and Ledoux et al., 1994a,b; Apergis-Schoute et al., 2005; Han et al., 2008; Weinberger, 2011, etc) suggested that CS-US association and thus, neuronal plasticity can occur prior to the amygdala, in non-sensory (higher-order) thalamic regions, like PIN/PIL. Therefore, the following statement is only partially correct. "In the lateral (LA)- and basal amygdala (LA and BA respectively), auditory (CS) and somatosensory (US) information are integrated.."

We agree with the reviewer. This part of the Introduction has been re-written, to make the statements about the LA and BA more general (see p. 3, bottom).

Was there any level of dorsoventral topography found in the location of the Cs and/or US responsive cells (striatal tail vs. Astr)?

In response to this question, we now show the reconstruction of imaged neurons also in coronal planes (see Figure 2 —figure supplement 1 for in-vivo ca2+ imaging of D1R+ neurons, and Figure 3 —figure supplement 3, bottom, for Adora+). For the US- and CS-responses in D1R+ neurons (but not for the movement-ON responses), it appears that neurons located more ventrally have smaller responses. Nevertheless, such a topography would have to be investigated in more detail, and with a larger sample size. Please note that the neurons which are localized more ventrally in Figure 2 —figure supplement 1, are actually also located more anteriorly (see the "horizontal" view in Figure 2A), thus precluding a simply analysis merely in the dorso-ventral plane. Furthermore, please note that the apparent localization of these cells in the "BLA" as suggested in the coronal view (Figure 2 —figure supplement 1), is actually an artefact of the projection (compare with the "horizontal" view in Figure 2A).

Is it possible that the examined and responded D1 and Adora+ cells are actually located in two distinct striatal regions? It is discussed that these are rather topographically non-overlapping populations.

Indeed, in the previous version we had reported (p. 13, l. 292 – 293 of the original ms), that Adora+ neurons with movement-ON responses might be localized more anteriorly, than D1R+ neurons with movement-ON responses. However, because it seems difficult to back-up this observation with a statistical analysis, we have now removed the statement.

Is it possible to follow the Ca-activity of individual cells throughout the entire learning paradigm? If yes, plasticity of a single cell could be analysed (like in the thalamus, Taylor et al., 2021).

In response to the criticism by reviewer 2 and 3, we have re-analyzed the Ca-imaging data. The analysis is now based on a CNMF-E based approach to detect ROIs (in "Caiman"; see Methods and response to reviewers 2 and 3), and with this approach it should in principle be possible to register cells over various days. Unfortunately, the number of cells that can be registered successfully across days is quite low. We have therefore refrained from further analyzing the imaged neurons in a day-by-day fashion.

As Figure 7 showed some level of topography in the InsCx-Astr connectivity, were different sectors of InsCx targeted in the D1 and Adore-Cre animals used for patch clamp recordings in Figure 8?

As the reviewer remarks, there is an apparent difference in the density along the a-p axis of cortex of presynaptic neurons proving input to D1R+ versus Adora+ neuron (Figure 7I, J).

For the patch-clamp recordings, we did not target different areas but instead, we always targeted the posterior insular cortex (pInsCx) for the injections of AAV8 driving the expression of Chronos, (see Methods, p. 25, bottom).

We also clarified in the Results and Discussion that for this experiment, we targeted the pInsCx and not the S2, but that some spill – over of Chronos-expressing virus into the S2 cannot be excluded (see p. 16, middle, and p. 21 middle).

Pg 17. Between the AAV and RV, three weeks is indicated in the text, but 4 weeks in the figure.

Thank you for pointing out this inconsistency, which has been corrected (see p. 15, top)

Figure 8, Figure suppl. 4 B and C. It seems that distinct parts of striatum are targeted.

It is true that in the example images, the injection into the AdoraCre mouse was targeted somewhat more dorsally than in the D1RCre mouse.

These images were taken from the control groups for D1RCre / AdoraCre mice expressing eGFP, from the datsets in Figure 5 / 6. Because the behavioral experiments were performed with bi-lateral fiber placements (to ensure sufficiently high effect sizes) and bi-lateral expression of Arch (for the 'effect' group) and eGFP (control group), we think a certain amount of variability in the targeting is un-avoidable. This, however, should not have affected the validity of these mice as control mice.

In response to the criticism, we have now removed this figure supplement, since the anatomical analysis had remained somewhat preliminary. Furthermore, for the discussion, this data is not needed, and we feel that the targeting of the vTS in the silencing experiments of Figure 5 and 6 is now amply documented in the new figure supplements to Figure 5 and Figure 6.

As there is very little data about Astr, it would be interesting to examine the basic anatomical organization of Astr. For example, the proportion of D1+ and Adora+ cells, the presence of other types of neurons (cholinergic, PV cells, etc). But, these may be far beyond the scope of the present study.

As the reviewer noted, we think these additional ideas for studying anatomical properties and interneuron types of the vTS goes beyond the scope of the current paper.

Reviewer #2 (Recommendations for the authors):

It is disappointing that the presentation and some of the analyses do not do justice to the work. Please see my points below. In particular, the heavy handed analyses of the fluorescence transients is too much and fraught with potential for error. More importantly, it obscures what appear to be large differences that could be extracted more easily (and convincingly). Please see the "approach section below". Lastly, at times the authors interpret their data and results too optimistically.

Please see our responses to the specific points below.

Specifically, the authors should:

1. Reanalyze the data using only z-score F(t).

In response, and also to satisfy the criticism of reviewer 3 (point 4 and 5), we have newly analyzed the Ca-imaging data, from scratch. First, for the detection of ROIs (see reviewer 3, point 3), we have used the analysis routines provided by Caiman. See new display of Caimaging data in Figure 1H – R, and in Figure 3E – O, which are now based on Z-score instead of on deconvolved event frequencies.

Regarding the analysis of Ca event activity during the four different conditions (combinations of movement or freezing with presence of absence of CS; "conditional event frequencies"), we continue to think that a read-out of discrete events is advantageous over simple z-scored Ca-traces. Nevertheless, we now use the previously described Ca-deconvolution approach available in the Caiman software, based on the previous papers by Pnevmatikakis et al. 2016 Neuron, and Giovannucci et al. 2019, eLife (both papers are cited in the manuscript). The complete re-analysis of our ca2+ imaging data should address possible caveats that might have been present in our previous custom-written deconvolution analysis.

2. Show more primary data – images of cells, etc…

We now extensively show color-coded z-scored fluorescence traces aligned to different stimuli (tones, footshocks) / behavioral variables (movement – ON)(Figure 1H-R; Figure 3 EO).

3. Refrain from reaching conclusions too early in the results especially that it is the contribution of plasticity of the neurons that is being studied in the silencing experiments (as opposed to the need for activity).

We have revised the text on all occasions in which we implicated that silencing during the footshock should affect a "plasticity". Rather, we now use the term "footshock – driven activity" or similar in these instances (see Results on several occasions). Nevertheless, in the discussion we express our interpretation that footshocks likely drive plasticity in these neurons (p. 19, middle – bottom).

4. Improve the presentation of the methods

The Methods for the analysis of Ca-imaging data has been re-written, according to the new analysis we have performed (see Methods, pages 31 – 33).

Please read the detailed comments below.

Abstract:

1. Imprecise wording. The notion of "balancing" is unclear in "defensive behaviors need to be finely balance" which makes it hard to understand exactly what is being studied.

We added "in the presence or absence of a threat-predicting cue" (p. 2, l. 17), to make it clear that we mean the balancing of learned freezing behavior across the CS- and no-CS periods.

2. "In-vivo optogenetic silencing during the training day showed that plasticity in D1R+ AStria neurons contributes to auditory-cued fear memories"

Please see my comments on the interpretation of the silencing experiments.

This sentence was modified as pointed out above (major point 3), using the term "footshock-driven activity" instead of "plasticity" (p. 11, l. 254 – 255).

Results:

1. Show some images of neurons with corresponding fluorescence transients to understand the quality of the data. I can't find them anywhere in the figures.

We show images of neurons (Figure 1C), and raw traces of ca2+ (Figure 1 E-G).

2. Line 111 – "Seemed to increase". What does this mean?

This statement has now been deleted, when re-writing the Results part to accommodate the newly analyzed Ca imaging data.

3. Line 114 – Claim to have derived APs by deconvolution.

This sentence has now been removed, because – as suggested – we simply analyzed most of the Ca imaging data in the form of z-scored Ca traces.

4. Line 127 – "Similarly suggested an increased".

We feel this statement made sense in the framework of the previous analysis. An increase in the tone response throughout the population of neurons is also present in the z-score traces in the new analysis (see Figure 1 I, J, K). However, since this sentence related to the previous analysis based on Ca-event frequencies, it was now removed.

5. See comments on methods – by the time we get to summary data as I Figures IJ-L it is very hard to understand what these mean. They are in au units x 10e-4. What does this mean? If they are au then they can be remapped.

This data is now shown in the form of aligned z-scored Ca traces (Figure 1 H-J), and thus use the unit of z-score.

6. Line 154 "These experiments thus show that D1R+ AStria neurons increase their AP firing activity during the onset of movements after a period of immobility. "

The analysis in this section is potentially problematic as tones are being played and the animal is starting and stopping movement. It is not clear that the response is to the movement change as opposed to the tone. Please show the histogram of tone start and stop times relative to the movement onsets.

A better analysis is one that models the contributions tones, movement starts, stops, and US. This can be done with a GLM and will alone one to disambiguate these confounds.

I am especially worried because of the follow-up sentence "Line 155 were more pronounced during the CS than in the absence of a tone" which suggests that these are not movement cells.

In the course of the re-analysis of the data (Ca Z-scores instead of deconvolved Caevents), we have now split the analysis of movement – ON responses in two blocks. First, we analyze the movement – ON responses for the "no – CS" periods, when no tones were given (Figure 1 M-O for D1R+ neurons, and Figure 3J-L for Adora+). This analysis unambiguously shows the presence of movement – ON responses. Second, we have then aligned the z-scored Ca traces to the onset of movements during the CS. As requested by the reviewer, we have now computed the histograms of the times of CS occurrence relative to the movement – ON transition (Figure 1 —figure supplement 3 for D1R+). This indeed shows, as the reviewer suspected, that part of the movement – ON responses were preceded by tones.

In order to correct for these responses, which might have been caused by tones rather than by movement – ON transitions, we removed all Ca – traces that were preceded by 0 – 400 ms by a tone, leading to the removal of about half of all traces (see Figure 1 —figure supplement 3, numbers given at the bottom). The corrected traces are shown in red in Figure 1 —figure supplement 3. The corrected traces still show a substantial increase of Ca in D1R+ neurons during movement – ON transitions during the CS (Figure 1—figure supplement 3, red traces). For the analysis of the number of neurons responding to movement – ON transitions during the CS (Figure 1S, filled data points), we used the corresponding tone-event-corrected traces.

In summary, responses to movement – ON transitions are clearly present in D1R+ neurons, and in Adora+ neurons (see the analogous analysis for Adora+ neurons in Figure 3M-O, and Figure 3P). Our new analysis furthermore shows that the number of neurons that respond to movement – ON transitions is larger in the presence of a CS than in its absence (Figure 1S, and Figure 3P). However, because of the caveat that part of the movement – ON responses are possibly caused by preceding tones, we have removed the statements which described a positive interaction between "CS" and movement – ON responses (see Results).

7. Line 159 – The conclusion in this sentence cannot be justified based on the data shown so far.

This sentence about a non-linear interaction of a representation between tones and movement – ON responses was removed. See also our response to point [8].

8. Line 168 – One cannot make these conclusions by comparison of event-triggered averaging.

This sentence was also removed; in relation to point [8].

9. Overall Figure 1 is very hard to follow. The number of panels should be reduced, labels should be placed to show which analyses are relative to movement, CS, etc…

We have strived to make Figure 1 easier to follow by the various changes that went along with the re-analysis of the data (Z-scores). As suggested, we have also introduced labels to indicate the principal response types (US, Tone, movement); similar changes were done in Figure 3.

10. Figure 2 is very nice.

Thank you; this Figure is now based on newly analyzed data, since both the coordinates of cell maps, as well as the response strength of the cells was changed slightly after the reanalysis of the Ca imaging data.

11. The paragraph starting on line 252 is confusing. Are the authors saying that the # of cells responding goes up (i.e. response fidelity) but that the response per active cell (estimated AP content) does not? Both the positive and negative conclusion refer to Figure 3R. I can't find a definition of response fidelity, which is used several times.

Sorry for the confusion. In the revision, we now removed the analysis of "response fidelity", so this should no longer be a concern.

12. The analysis of Figure 3S says that significance was judged by a KS-test, which is usually used for comparisons of unbinned cdfs. The graph is an average +/- error bar. How was the KS test used?

This is likely a mis-understanding; we used a Kruskal-Wallis test (abbreviated as "KW"). As detailed in Materials and methods, the Kruskal-Wallis test is a non-parametric version of a one-way ANOVA. This test was used to test the "general" significance of changes according to time (three days) and class of events (four combinations of movement / freezing and CS / no-CS), before using post-hoc comparison (Dunn's test). In the revised version, we always write out the test name for "Kruskal-Wallis" to avoid a possible confusing with "KS".

13. If the freezing habituation data in 1S and 3S cannot be analyzed statistically, it should be removed from the panels as one cannot help but interpret the data shown quantitatively.

After the complete re-analysis of the ca2+ imaging data, we chose to keep the data (see Figure 1 U, 3R).

14. The Venn diagrams in 3V and 2D would benefit by labeling the numbers in the main intersection areas.

This was done accordingly, albeit only for the "simple" overlaps between tone- and movement – ON responses to keep the graph light.

15. In figure 4, the baseline movement of the Adora2a and D1R groups is very different. Are the two genotypes equivalent? These are BACs so it would not be too surprising.

We think the absolute movement indices, which are analyzed by the "ezTrack" software from the videos of the mice (see Methods, p. 31, middle), should not be compared directly. For example, the movement of the AdoraCre mice, in the newly added data (Figure 4 —figure supplement 1), shows much higher apparent levels of movement index. Indeed, the exact values depend on the calibration of the system, the camera used, and possibly on the size of mice. Thus, we refrain from deriving conclusions regarding possible differences between absolute movement indices.

16. For Figure 4, what in vivo proof is there that the cells are being activated? Given the negative conclusion, such a control is important.

We attempted optrode recordings to show that optogenetic stimulation of D1R+ neurons by blue light pulses in-vivo can drive AP-firing. Unfortunately, we were unable to record light-evoked activity with sufficiently high throughput, due to direct light -evoked artefacts in the recordings, and/or misplacements of the optrode (N = 4 mice, data not shown).

We therefore turned to a simpler approach, at least for Adora+ mice. In fact, in the original data set, we had used very brief light pulses (1 ms; 10 mW at the fiber tip), motivated by the fact that the kinetics of Chronos is fast (Klapoetke et al., 2014). We therefore validated in additional experiments, with N = 5 available Adora+ mice, whether optogenetic stimulation of Adora+ neurons in the vTS with longer light pulses (2 ms, and 5 ms; 10 mW), would produce comparable results to the ones in Figure 4. Indeed, these new experiments similarly did not show an effect on the movement of the mice (see new Figure 4 —figure supplement 1). The new experiments further corroborate our findings that optogenetic stimulation of Adora+ and D1R+ neurons in the vTS do not have a direct effect on the movement of mice.

17. The experiments motivated by the paragraph starting at Line 325 are very nice. However, the paragraph is troubling. The manipulation is to reduce activity and it should be phrased in that way. Instead, a hypothesis based on plasticity is presented, from which the experimental manipulation of activity is indirectly motivated.

The paragraph has been re-written, using the notion of "footshock-driven activity" instead of "aversively – motivated plasticity" of D1R+ neurons.

18. There are differences in baseline freezing (5F) before the light is turned on that are of similar magnitude (relative to the low level of freezing) as seen in 5H. Similarly, the comparison between no effect in 5G right to less effect in 5I left may not be fair given that the latter is in "steady state" and the former not. Is an RM-ANOVA the right way to go?

Thank you for your detailed observation of the data in Figure 5F. In Figure 5F (now Figure 5D in the revised version), and Figure 5H (5F in the revised version – note that below we only used the "new" Figure numbers), we show averaged freezing percentages across all mice in each group, at a time resolution of 10 s. The reviewer refers to "differences in baseline freezing (5F) before the light is turned on". In these experiments, 30s trains of tones (each 0.1 s, given at 1 Hz) are given (blue – shaded area), and then a final footshock is given immediately after each CS (Figure 5F, yellow vertical lines). The yellow light, to activate Arch, is given 1s before each footshock, for a duration of 3s. Thus, one would have to look at the last 1-2 freezing values before each yellow vertical line, to check for "differences in baselines freezing before the light is turned on". This would be at times when the CS is on (blue – shaded areas). Indeed, in the time-binned analysis, the freezing levels differ between the two groups especially during CS 4 and CS 5, but not in a consistent manner – once the eGFP (control) mice show higher freezing (CS5), and once the Arch mice show higher freezing (CS 4)(Figure 5E, left).

It should be noted that currently, there seems to be no consensus in the fear learning field whether the CS applied on the training day (during pairing with the US), already has acquired the properties of a CS for the animals. Thus, it is possible that freezing during the CS, on the training day, can be seen as a "prolongation" of an ongoing contextual freezing, that buildups over longer times during the training session (Figure 5D, green and black traces).

In the previous, and revised version we have analyzed the freezing in 30s time bins preceding each CS (lower grey bars in Figure 5E; 5F, right), and the freezing during each CS (Figure 5E, left). This data shows that both within the control- and the Arch mice, freezing is similar during the CS, and during the no-CS times (Figure 5F, compare data between left and right panel), which supports the view that during the training day, mice freeze in an increasing manner in response to the general context, but not specifically in response to the tones.

In summary, we think small fluctuations in freezing levels between the groups on the training day, and within CS4 and CS5 in this case, are not of biological relevance.

Please also note that, because we re-included N = 2 eGFP mice, and N = 3 Arch mice that had been previously excluded because they froze less than 20% at the end of the training day (see reviewer 2, point 26), the exact form of the traces and positions of datapoints in this Figure has slightly changed.

19. Similar concerns exist for the data in Figure 6.

Can the differences in Figure 5 or 6 be used to identify individual mice as ones that received silencing and ones that did not? It seems unlikely, especially for Figure H. It might have been better to run cohorts of mice in which the silencing was only applied in the retrieval portion. Otherwise, it is unclear if the relatively modest effects are due to a difference during training that is not reflected in freezing rates.

For a general answer regarding to the "similar concerns" in Figure 6, please see our answer above for Figure 5.

The experimental approach here was to suppress footshock-driven activity in either D1R+ or Adora+ neurons of the vTS, and then observe whether this manipulation would impair fear learning. This experiment has revealed opposite effects when applied to neurons in the direct

(D1R+) and indirect pathways (Adora+)(a contribution to auditory-cued fear memory for D1R+ neurons), as opposed to an increased contextual fear memory component for the Adora+ mice (but see also comment below on the changed statistical significance in the Adora+ dataset). The reviewer proposes that "it might have been better" to silence the activity of the neurons during the recall day. We feel, however, that the experiment as run by us makes sense, and has produced interesting results, whereas silencing during the recall day will require further considerations as to when best to silence (since the effects of D1R+ and Adora+ neurons are produced at different times: CS versus no-CS times). Thus, we think that silencing the activity of D1R+ and Adora+ neurons during the tones / context phases of the memory recall day is beyond the scope of the present paper.

Regarding the point that it is "unclear if the relatively modest effects are due to a difference during training that is not reflected in freezing rates": The rationale of these experiments was to suppress footshock – driven activity in a specific neuronal subsystem, and by this to most likely reduce plasticity in this sub-system (we express this view in the motivation paragraph to the silencing experiments; see p. 11, middle – bottom). We find it encouraging that this manipulation, which presumably impaired plasticity in the vTS, leads to selective changes in memory recall one day later, which indicates that the vTS contributes to the formation of an auditory-cued fear memory. We don't think it is problematic that the manipulations on the training day did not impair the increasing wave of "contextual" or "anticipatory" freezing observed on that same day, because other brain areas might drive the contextual, or anticipatory freezing observed on the training day.

Changed statistical significance in the Adora+ data set:

As requested by this reviewer below (point [26]), we have now re-introduced the previously excluded mice (which were excluded based on the criterium that they showed less than 20% freezing at the end of the training session, spanning the times of the fifth and sixth CS-US pairing). While the statistical significance in the D1R+ dataset was maintained (see Results), the statistical significance in the Adora+ dataset was lost, while there was still a trend in the data showing an increase in the freezing at no – CS times on the fear memory recall day (p = 0.0512; two-way repeated measures ANOVA; previously, p = 0.041). Therefore, we have now toned down our conclusions regarding the Adora+ mouse, and merely speak about a "trend" in the data towards an increased freezing during no – CS times after suppressing footshock-evoked activity in Adora+ vTS neurons (see e.g. changed text in p. 14, l. 327).

20. The main text does not reference controls for the rabies experiments which generally should include (1) TVA dependence of rabies infection, (2) G dependence of spread, (3) Cre-dependence of the DIO/floxed constructs. It is important to do these with the same batch of viruses used for the experiments (ideally in parallel!).

We have now performed additional control experiments as requested by the reviewer (see new Figure 7 —figure supplement 1, and changed Results text; p. 15 top).

21. I hesitate to make too much of the differences shown in 7I/J given N=2, especially without a careful analysis of all the starter cells to show that the number and distribution of these were the same across genotypes.

In response, we have toned-down our conclusions about this experiment, and now merely report the observed distribution (see p. 16, top).

22. The currents in Figures 8G-H are huge and loss of voltage clamp must have occurred, especially when measuring NMDA receptor currents.

Yes, the optogenetically evoked EPSCs are larger than most reports in the literature for long-range excitatory connections in the forebrain.

We have extensive experience with imposing voltage-clamp also on large and fast EPSCs. This experience dates back from our work at the calyx of Held synapses (where EPSCs can be up to 20 nA; see e.g. Meyer et al. 2001 J. Neuroscience; Kochubey et al. 2009 J. Physiology). Also, at the MNTB – LSO inhibitory connection, we more recently reported optogenetically-evoked IPSCs of up to 20 nA (Gjoni et al., 2018 J. Physiology). This requires to minimize the series resistance (Rs). Also, the finding that the optogenetically-evoked EPSC at the pInsCx – vTS connection are surprisingly large, is a sign that our voltage-clamp conditions might be appropriate, because imperfect voltage-clamp would lead to an underestimation of the true synaptic conductance (see discussion in Gjoni et al. 2018 – in this paper we reported significantly larger conductance values for IPSCs than those reported before at the same connection). For these reasons, we think we are well-positioned to report the occurrence of unusually large EPSCs.

We now discuss that the cortical inputs to the vTS are of surprisingly large amplitude (p. 21, middle).

Approach:

1. Large lesions even for optogenetics because of use of 1.25 mm cannula.

This seems to be a misunderstanding. The 1.25 mm outer diameter ceramic ferrules remain outside of the skull. The optic fibers are inserted into the brain, and have an outer diameter of 230 µm (see also improved methods description, p. 30, bottom).

2. Line 763 -- "If mice from any experimental group showed low freezing, they were excluded from the analysis; we used a threshold of 20% time spent freezing

during the fifth and sixth CS-blocks on day 2 (7 out of 59 mice for the behavioral experiments were excluded)."

It is good of the authors to give the exclusion criteria. However, this is worrisome. Some of the loss of function experiments are based on the hypothesis that activity is necessary for learning. Therefore, this discards the poor learners, right?

What is the breakdown of exclusion across the different experimental and genetic conditions? Why not simply include all animals to capture the natural variability?

We have re-added the previously excluded mice to the datasets in Figure 5 and Figure 6. Corresponding changes have been made in the Figures, and in the Results text. Please also see our answer to your points [20], [21] above.

3. The deconvolution approach is worrisome for a few reasons.

a. Ca entry into these neurons is highly non-linear with single AP, burst AP, up-state APs, and GPRC-mediated influx. It is unclear that one can use a linear model of # of APs to [Ca] (to F(Gcamp6m)). It does not seem that this step is necessary (especially given the heavy filtering).

b. Data is acquired at 10Hz but then a band-pass filter at 2-4 Hz (If I read it correctly although the methods are not very clear). This is much less than the Nyquist limit. Why? Also why place a low-frequency limit (i.e. why use a band pass as opposed to a low-pass to simply get rid of shot and electronic noise?).

c. The deconvolution kernel is arbitrary and uses a rise time that cannot be captured with the 2-4 Hz band pass. The authors need to show that 0.5 s tau(decay) is justified. There is no justification for choosing an amplitude of 2.

d. Line 793 – The phrase "deconvolution yielded a… proportional to AP firing rate" must be removed as it cannot be justified.

e. The deconvolved signal (R) is further filtered with a box car – Why? The filtering is done upstream on the fluorescence. It should not be done again.

f. "local peaks" exceeding a threshold (arbitrarily set to 2) were detected using a "first-derivative" method. Why add a threshold here? It seems unjustified. Second, how does one detect a peak with a first-derivative? Shouldn't it be a second derivative? Or are they detecting fast-rate of rise, in which case why set a peak amplitude threshold?

4. All of these concerns are acknowledged later (line 801) in "we did not aim to infer exact AP spiking rates of cells". Then why do this and present it in this way?

5. They should simply get rid of all these steps and analyze the z-score raw F (calculating a DF/F and then z-scoring is no different than just z-scored the Z).

6. y comparing the

7. The section starting on line 809 indicates that different baselines were chosen for comparison for different types of cells. This introduces a circular bias. One needs to be able to statistically show that a cell is a "move-on" or CS responder without changing a baseline. Why not use the Z-score values in the time-bin without picking a base-line – i.e. compare activity in that window to all activity?

8. Line 798 – The amplitudes of the events from deconvolution (referred to as estimated AP content, EAC [a.u./s]) were proportional to the amplitude of the ca2+-transients (e.g. see Figure 1 —figure supplement 1E).

a. I can't find the referenced data in a figure. The sentence suggests a scatter plot with some kind of regression analysis?

b. If this is true, why do it all? Just use the F as suggested above and save all these potentially problematic analysis steps.

Thank you for your detailed comments on our previous deconvolution – based analysis of Ca transients and underlying neuronal activity. As suggested by this reviewer and by reviewer 3, we have now completely re-done the analysis of the Ca imaging data. As requested, we have performed most analysis steps on the Z-scored fluorescence (Ca) traces (see new panels Figure 1H-R and Figure 3E-O for D1R+ and Adora+ neurons, respectively) (please note that the Caiman software does not allow access to deltaF/F traces so we used Zscored traces). The fluorescence data were newly extracted using an CNMF-E based approach.

We only maintained a deconvolution – based approach for the analysis of "conditional frequencies" of Ca events in panels Figure 1U, and Figure 3R. For this, we used the built-in deconvolution approach of Caiman, instead of the previous custom-written analysis. Thus, we feel that all issues raised by the reviewer in these above comments, relating to filtering and other technical issues, should have been resolved by the new analysis of the data.

Please see the new description in Material and methods, which succinctly describes how we analyzed the Ca data (p. 32 – 34). Please also see our response to the related points 3 and 4 of reviewer 3.

Reviewer #3 (Recommendations for the authors):

1. The study used only male mice. There is no basis for excluding females.

We have now stated the sex of the investigated animals in the abstract, and we have explained the choice of male mice as a model in the Methods (new text in p. 24, top). Furthermore, we have discussed possible sex-specific differences regarding the role of the vTS in fear learning (p. 22 bottom, p. 23 top).

2. The authors did not verify specificity of D1R and Adora lines (tdTomato + RNAscope for D1R and Adora). Without this verification, it is unclear if the fluorescent patterns reported actually reflect patterns of D1R and Adora neurons.

We have used a D1RCre mouse line (EY217 line from the GenSAT initiative), and an AdoraCre mouse line (KG139 line from the GenSAT initiative), which both have been used widely by previous studies in the striatum. These mouse lines have been well characterized in the GENSAT project (see expression profiles of these two lines on the GENSAT website; http://www.gensat.org/cre.jsp). We see no obvious reason that any of the two lines should express outside of its previously documented expression domain.

In response to this question, and to Reviewer 1 point [3], we have performed additional anatomical control experiments. We have crossed the D1RCre and the AdoraCre mouse line with a Cre-dependent tdTomato reporter line (Ai14), and we have then analyzed the location of tdTomato-positive cells throughout the striatum and the amygdalar complex (N = 3 mice each). Representative images on three different a-p levels, after aligning with the Paxinos mouse brain atlas, are shown in Figure 1 —figure supplement 1 and Figure 3 —figure supplement 1.

For the D1RCre x tdT mice, it can be seen that tdTomato-positive cells are present widely within the striatum ("CPu / CP in Paxinos), and that tdTomato-positive axons project to the entopeduncular nucleus / GPi (Figure 1 —figure supplement 1B, C), a hallmark for neurons of the direct pathway. Interestingly, as was also noted in the original paper describing these lines (Gerfen et al., 2013), tdTomato expression was sparse in the cortex and in the claustrum; moreover, we did not observe tdTomato expression in the intercalated cell masses that surround the BLA. This feature of the D1RCre (EY217) line is advantageous for our purpose, because it ensures that we did not inadvertently image, or silence D1R-expressing neurons in adjacent structures to the vTS (like deep layers of cortex, claustrum, intercalated cell masses surrounding the BLA). Indeed, in another D1RCre line from GENSAT, expression that matches more accurately "the full pattern of endogenous expression of the Drd1a gene" was observed (line FK150; citation from Gerfen et al. 2013). In that line, expression is also seen in cortex, claustrum, and the intercalated cell masses (see images on the GENSAT website; http://www.gensat.org/cre.jsp). One possible dis-advantage resulting from the choice of the D1RCre (EY217) line is that expression more posteriorly and ventrally becomes weak in this mouse (see Figure 1 —figure supplement 1C). In this respect, it also seems more cautious that we now refer to the neurons targeted in our study as ventral tail striatum (vTS), but no longer as "AStria" (see also Reviewer 1, point [3]).

For the AdoraCre x tdT mice, it is again seen that tdTomato-positive neurons are contained throughout the striatum as expected, and axonal projections target the GPe (Figure 3 —figure supplement 1A, B, C), a hallmark of the indirect pathway. TdTomato-positive neurons are found in the CPu/CP extending partially into the AStria. Only few Cre- positive cells are found in structures adjacent to the CP/CPu and AStria. These data show that this mouse is well-suited to target neurons of the indirect pathway (D2R+ or as here, Adora+) in the vTS.

Taken together, the present and previous characterization of the two Cre mouse lines used here is quite extensive, and we do not expect that the proposed control experiment (RNAscope for D1R, Adora, and tdTomato in each mouse line) would reveal significant weaknesses of the used lines. Moreover, our slice electrophysiology data shows that input EPSCs from the pInsCx/S2 received by D1R+ and Adora+ neurons are of different amplitudes, different E/I ratios, and show different AMPA/NMDA ratios after fear learning (Figures 8, 9). This again confirms that we are dealing with two separate neuronal populations. All this evidence suggests that the DR1Cre and AdoraCre mouse lines used here are valid tools to target neurons of the striatal direct- and indirect pathways.

In response to this criticism, we now show images of the D1RCre x tdT and AdoraCre x tdT mice in Figure 1 —figure supplement 1 and Figure 3 —figure supplement 1, to introduce the expression pattern of each Cre mouse line in the vTS.

3. Inspection of the heat plots shown in figures 1 and 3 indicate the fluorescent data contains repeating 'neurons'. For example in Figure 3, the two signals in rows 10 and 11 are identical. This is particularly notable because the repeating 'neuron' shows the largest change in fluorescence. In addition, rows ~17 and ~34 appear virtually identical as well. This is a major problem and indicates the authors have an issue in their analysis pipeline. Greater care needs to be taken to ensure repeating neuron's are not reported and analyzed.

Same issue occurs in Figure 1. Rows 3 and 4 are identical signals. Repeating 'neurons' appears to be a problem throughout.

Thank you for this detailed observation of our data, this has previously escaped our attention. With the re-analysis of all imaging data in Figure 1-4, this problem should be solved. For the previous analysis, we had hand-drawn ROIs around active cells. Since we imaged over three focal planes, apparently some cells had been included twice in the analysis. For the new analysis, we used a CNMF – E based approach in the "Caiman" software to identify active cells, and this software makes a co-variance test between cells and then excludes "repeating neurons", so the problem should not persist in the new data set.

4. GCaMP6m is a fluorescent calcium indicator. While calcium entry into neurons is required for AP-driven vesicular release, calcium entry into a neuron can reflect many other processes. In this manuscript, the authors wish to examine single unit activity by acquiring and deconvolving fluorescent signals. In fact, there are known issues relating fluorescent signals to action potentials in the striatum:

https://www.biorxiv.org/content/10.1101/2021.01.20.427525v2.full

If the authors wish to analyze spiking activity of the striatal neurons, they need to use techniques that directly record action potentials. These techniques are readily available. If the authors wish to record fluorescent changes resulting from calcium influx, they need to embrace this decision and only analyze changes in fluorescence data.

Thank you for pointing out the paper Legaria et al. 2022 (Nat. Neuroscience). In it, the authors show that measuring global Ca averaged over potentially many 100s of neurons in fiber photometry, correlates only marginally with the AP-firing activity of striatal neurons. The authors then additionally use miniature microscope Ca imaging with similar methods as used here (their Figure 3), to differentiate between the soma- and non-somatic Ca signals. Thus, it is clear that miniature microscopy has a big advantage (=cellular resolution) over fiber photometry.

This said, we have followed the advice of this reviewer and reviewer 2 and have completely re-analyzed the Ca – imaging data. We now have used CNMFE-based methods for ROI (cell body) detection as implemented in Caiman, and we have expressed most results in terms of zscored Ca traces (see new panels of Figure 1G-Q, and Figure 3E-O). We still use deconvolution of Ca-traces for the final analysis of "conditional Ca-event frequency" (see Figures 1T, 3R), because for this analysis it was necessary to count events. We now use a deconvolution analysis in the Caiman package, based on previously described methods and shared analysis programs (Pnevmatikakis et al. 2016 Neuron; Giovannucci et al. 2019 eLife).

Taken together, we hope that the new analysis of fluorescence (Ca) data in Figures 1 – 3 has improved the logical flow of the presentation of the Results. We have also dropped all statements in the text in which we had suggested that Ca-deconvolution can derive "APfiring".

Please also see our answer to "Essential revision points 3, and 4".

5. I found it very difficult to track the logical flow of the analyses. The authors start off by showing 'neurons' are responsive to cues. But then almost immediately pivot to showing they do not in fact respond to cues, but movement during cues. A lot of figure space and analysis is devoted to this, such that the big picture is lost. I would find it more convincing to develop analyses that directly compare cue vs. movement vs. cue x movement responding from the outset. A regression approach may be useful. Take the 30 s prior to cue and 30 s cue. Have one regressor be cue on vs cue off, another be movement on vs movement off and the last be the intersection of cue on/off and movement on/off. This would simultaneously compare each regressor to each 'neurons' activity pattern, determining which best captures change in fluorescence.

Thank you for this feedback. We admit that the previous version of Figure 1 was dense, and the data as presented following the deconvolution analysis was difficult to grasp. We have attempted to improve these issues of data presentation by the new analysis of the data (largely based on Z-scored fluorescence traces, see above), and by using improved organization and labels in Figures 1, and 3.

Nevertheless, the fundamental properties of the data remain the same as before; i.e. there are both responses to sensory cues (US, and CS – driven responses), as well as responses to behavioral state variables, like movement – ON transitions of the animals. We have tried to improve the logical flow, by now clearly showing movement – ON responses in the absence of tone (CS) stimulation first (Figure 1M-O; Figure 3J-L). Nevertheless, within the CS blocks (tone stimuli every s), the issue persists that movement – ON transitions can occur superimposed with tones. Indeed, a re-analysis of the occurrence of the CS (tone) events relative to movement – ON transitions shows that in a significant number of movement – ON transitions during the CS (~ 50% for D1RCre mice; see Figure 1 —figure supplement 2), the movement – ON transitions were, in fact, preceded by 0 – 400 ms by a tone (see also our answer to reviewer 2, point [8]). This phenomenon was more pronounced in the D1RCre mice, and likely indicates that tones can sometimes induce the onset of movements, with a delay of ~ 200 – 400 ms. We feel that in this situation, a generalized linear model (GLM) would be of limited usefulness, because tone- and movement – driven events can sometimes occur with essentially similar delays, and in a similar periodicity (the periodicity is imposed by the periodic tones during the CS block). Thus, we applied a timing – based analysis, in which we removed the Ca responses for those events in which the movement – ON transition happened to be preceded, by 0 – 400 ms, by a tone. The resulting "corrected" average Ca transients showed a similar overall trend as the non-corrected traces (Figure 1O-Q, compare red and black traces; similar for Figure 3 M-O for AdoraCre mice).

Therefore, we feel that have fully addressed the possible concern of the overlay of movement – ON responses, and tone responses. Because of the possible triggering of a part of the movement – ON responses by tones (see above), we now removed all statements which had claimed a non-linear interaction between movement – ON and tone responses (see Results, and Discussion).

6. Do not show hypothetical behavior data (as in Figure 1A). Only show real behavior data.

We have now changed the scheme of Figure 1A (and of Figure 5B) to retain the timing information of the behavioral protocols, without showing "hypothetical" behavior.

In its present form, I do not feel that the authors achieved their aims or that the results support their conclusions. Addressing the points above is more likely to produce such results.

We hope that the extensive revision of our paper would be able to convince the reviewer that our paper now reaches its aim. That is, we show, using an array of approaches like invivo Ca imaging, in-vivo optogenetic manipulations, and ex-vivo circuit mapping, that the ventral tail striatum (vTS) shapes learned defensive behaviors during fear learning, by different, but synergistic roles of direct, and indirect pathway neurons in this striatal area.

Although even then, it appears that the contribution of these two cell types to tone-shock learning appears limited.

We agree with the reviewer that the contribution of the ventral tail striatum to fear learning is more of a "modulatory" role, although we clearly describe that the two neuron types have roles during different phases of learned fear behavior. We have made changes in the text to indicate that the role of the vTS in fear learning is rather "modulatory" (see abstract, p. 2, l. 24; conclusions, p. 23, l. 553).

If these neurons are contributing to aversive behavior, perhaps their contribution would be better captured by procedures in which movement is a more central element to behavior.

Thank you for this interesting proposal. We think that developing alternative behavioral procedures for studying the role of the vTS is, unfortunately, beyond the scope of the present paper.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

The manuscript has been greatly improved during the revision, but there are some issues that remain to be addressed prior to acceptance, as outlined below:

Reviewer #2 (Recommendations for the authors):

The authors have greatly altered and improved the manuscript to address the reviewer concerns. In particular, they have implemented a standardized Ca analysis pipeline, altered the description of the conclusions in many places, improved the presentation of the data and added some necessary controls.

I wish that they had done a full characterization of the RV controls, instead of showing some selected images, but that is ok.

For the first revision, we have added additional control experiments for three conditions (see Figure 7 —figure supplement 1, [panels A, B]: expression of the helper constructs alone in Drd1aCre mice, [panels C, D]: expression of the same helper constructs in C57Bl/J mice; [panels E, F]: injection of the rabies virus into the vTS of a Drd1aCre mouse in the absence of previous helper virus expression). These three experiments have shown the expected results; so we are not quite sure of the statement by the reviewer "full characterization of the RV controls". Nevertheless, the reviewer finally seemed satisfied with the added data to Figure 7 —figure supplement 1 (for Revision 1).

I am confused about the sentence (lines 199-201)

"These experiments show that a substantial sub-population of Adora+ neurons in the vTS codes for movement onset, but this representation was unchanged by fear learning (Figure 3M – P)."

The changes in event frequency in Figure 3R from habituation to training for freeze state are quite large yet the conclusion is that training does not influence the representation of movement in Adora2a+ neurons? Perhaps I am misinterpreting the statement.

We appreciate that the reviewer seems to observe some differences in the ca2+ event frequency data in Adora2aCre mice induced by fear learning; however, statistical analysis shows that possible differences are not significant.

The reviewer refers to the ca2+ event frequency data shown in Figure 3R. Here, ca2+ event frequencies were plotted as average ± S.E.M. values (as indicated in the legend). In the corresponding Figure 3 —figure supplement 2, the same data are plotted as individual data points, and the median, and the interquartile ranges are superimposed on the data (as also indicated in the corresponding Figure legend). Because some of the data is highly nonnormally distributed, it is important to look at the more "raw" data display in Figure 3 —figure supplement 2 to fully understand whether specific data sets are different, or not. In addition, one needs to consult the results from statistical testing, reported in "Source Data Figure3.xlsx".

Specifically, the reviewer states "that changes in event frequency in Figure 3R from habituation to training for freeze state are quite large". For example, one might think that the average ca2+ event frequency for "Frz_noCS" on the Habituation day (leftmost blue symbol in Figure 3R) is higher than the corresponding value on the training day (third blue symbol in Figure 3R). However, when inspecting the corresponding data and their distributions in Figure 3 —figure supplement 3 ("Frz_noCS" on Habituation day versus "Frz_noCS" on the Training day), it is seen that the medians of these two datasets are almost the same (the average values are different, because the "Frz_noCS" data on the Habituation day shows a stronger skew to high values). Correspondingly, the statistical comparison between the "Frz_noCS" data of the Habituation day, versus the "Frz_noCS" data on the Training day, reports p > 0.99 (Dunn's multiple comparison test; this value can be seen in "Source Data Figure3.xlsx), sheet "Figure 3R cond_activity_Stats", line 33 ("Frz_noCS_Hab vs. Frz_noCS_Train").

Similarly, for other inter-day comparisons within the same movement state, comparisons are non-significant for the ca2+ event frequency in the Adora2aCre mice. Indeed, the display of the ca2+ event frequency data in Figure 3 —figure supplement 3 allows the conclusion that the main condition which changes the ca2+ event frequency is "movement" as compared to "freezing" (compare "red" data points with "blue" data points over all three days – all these comparisons are statistically significant as indicated by the "star" symbols above the comparisons).

Thus, we feel that our conclusion statement is justified.

Nevertheless, we have added a side sentence after the above conclusion sentence; the entire sentence now reads (l. 199 – 202):

"These experiments show that a substantial sub-population of Adora+ neurons in the vTS codes for movement onset, but this representation was unchanged by fear learning (Figure 3M – P), except for a decrease in the number of neurons showing a movement – ON response in the absence of a CS (Figure 3P, open symbols)."

The newly added sentence refers to our observation that the number of Adora+ neurons with a movement – ON responses outside the CS decreases with fear learning (data in Figure 3P, open symbols; p = 0.0229; Chi-square test, as reported on l. 196).

https://doi.org/10.7554/eLife.75703.sa2

Article and author information

Author details

  1. Michael Kintscher

    Laboratory for Synaptic Mechanisms, Brain Mind Institute, School of Life Science, Ecole Polytechnique Fédérale de Lausanne, Lausanne, Switzerland
    Contribution
    Conceptualization, Formal analysis, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing – original draft, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-2355-1369
  2. Olexiy Kochubey

    Laboratory for Synaptic Mechanisms, Brain Mind Institute, School of Life Science, Ecole Polytechnique Fédérale de Lausanne, Lausanne, Switzerland
    Contribution
    Formal analysis, Validation, Investigation, Methodology, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-8115-7733
  3. Ralf Schneggenburger

    Laboratory for Synaptic Mechanisms, Brain Mind Institute, School of Life Science, Ecole Polytechnique Fédérale de Lausanne, Lausanne, Switzerland
    Contribution
    Conceptualization, Supervision, Funding acquisition, Validation, Writing – original draft, Writing – review and editing
    For correspondence
    ralf.schneggenburger@epfl.ch
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-6223-2830

Funding

European Molecular Biology Organization (ALTF 224-2015)

  • Michael Kintscher

Swiss National Science Foundation (31003A_176332 / 1)

  • Ralf Schneggenburger

NCCR Synapsy - The Synaptic Bases of Mental Disease (Project P28)

  • Ralf Schneggenburger

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank Shriya Palchaudhuri and Denys Osypenko for helpful discussions regarding the analysis of freezing behavior, and Elena Mombelli for help with anatomical experiments. Images were acquired at the Bioimaging and Optics Platform of EPFL (BIOP). This work was supported by an EMBO fellowship to MK (# ALTF 224-2015), and by grants from the Swiss National Science foundation (SNSF; grant number 31003 A_176332/1) and from the National Competence Center for Research (NCCR) of the SNSF "Synapsy - The Synaptic Bases of Mental disease", project P28 (both to RS).

Ethics

All experimental procedures with laboratory animals (Mus musculus) were performed under authorizations for animal experimentation by the veterinary office of the Canton of Vaud, Switzerland (authorizations VD3274 and VD3518).

Senior Editor

  1. Kate M Wassum, University of California, Los Angeles, United States

Reviewing Editor

  1. Mario A Penzo, National Institute of Mental Health, United States

Publication history

  1. Received: November 19, 2021
  2. Preprint posted: December 9, 2021 (view preprint)
  3. Accepted: January 16, 2023
  4. Accepted Manuscript published: January 19, 2023 (version 1)
  5. Version of Record published: February 3, 2023 (version 2)

Copyright

© 2023, Kintscher et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 1,019
    Page views
  • 175
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Michael Kintscher
  2. Olexiy Kochubey
  3. Ralf Schneggenburger
(2023)
A striatal circuit balances learned fear in the presence and absence of sensory cues
eLife 12:e75703.
https://doi.org/10.7554/eLife.75703

Further reading

    1. Neuroscience
    Xiaosha Wang, Bijun Wang, Yanchao Bi
    Research Article Updated

    One signature of the human brain is its ability to derive knowledge from language inputs, in addition to nonlinguistic sensory channels such as vision and touch. How does human language experience modulate the mechanism by which semantic knowledge is stored in the human brain? We investigated this question using a unique human model with varying amounts and qualities of early language exposure: early deaf adults who were born to hearing parents and had reduced early exposure and delayed acquisition of any natural human language (speech or sign), with early deaf adults who acquired sign language from birth as the control group that matches on nonlinguistic sensory experiences. Neural responses in a semantic judgment task with 90 written words that were familiar to both groups were measured using fMRI. The deaf group with reduced early language exposure, compared with the deaf control group, showed reduced semantic sensitivity, in both multivariate pattern (semantic structure encoding) and univariate (abstractness effect) analyses, in the left dorsal anterior temporal lobe (dATL). These results provide positive, causal evidence that language experience drives the neural semantic representation in the dATL, highlighting the roles of language in forming human neural semantic structures beyond nonverbal sensory experiences.

    1. Neuroscience
    Ayako Yamaguchi, Manon Peltier
    Research Article Updated

    Across phyla, males often produce species-specific vocalizations to attract females. Although understanding the neural mechanisms underlying behavior has been challenging in vertebrates, we previously identified two anatomically distinct central pattern generators (CPGs) that drive the fast and slow clicks of male Xenopus laevis, using an ex vivo preparation that produces fictive vocalizations. Here, we extended this approach to four additional species, X. amieti, X. cliivi, X. petersii, and X. tropicalis, by developing ex vivo brain preparation from which fictive vocalizations are elicited in response to a chemical or electrical stimulus. We found that even though the courtship calls are species-specific, the CPGs used to generate clicks are conserved across species. The fast CPGs, which critically rely on reciprocal connections between the parabrachial nucleus and the nucleus ambiguus, are conserved among fast-click species, and slow CPGs are shared among slow-click species. In addition, our results suggest that testosterone plays a role in organizing fast CPGs in fast-click species, but not in slow-click species. Moreover, fast CPGs are not inherited by all species but monopolized by fast-click species. The results suggest that species-specific calls of the genus Xenopus have evolved by utilizing conserved slow and/or fast CPGs inherited by each species.