Divergent outputs of the ventral lateral geniculate nucleus mediate visually evoked defensive behaviors

SUMMARY

Rapid alternations between exploration and defensive reactions require ongoing risk assessment. How visual cues and internal states flexibly modulate the selection of behaviors remains incompletely understood. Here, we show that the ventral lateral geniculate nucleus (vLGN)−a major retinorecipient structure−is a critical node in the network controlling defensive behaviors to visual threats. We find that vLGN^GABA neuron activity scales with the intensity of environmental illumination and is modulated by behavioral state. Chemogenetic activation of vLGN^GABA neurons reduces freezing, whereas inactivation dramatically extends the duration of freezing to visual threats. Perturbations of vLGN activity disrupt exploration in brightly illuminated environments. We describe both a vLGN→nucleus reuniens (Re) circuit and a vLGN→superior colliculus (SC) circuit, which exert opposite influences on defensive responses. These findings reveal roles for genetic- and projection-defined vLGN subpopulations in modulating the expression of behavioral threat responses according to internal state.

In brief

Salay and Huberman reveal that the ventral lateral geniculate nucleus (vLGN)−an understudied yet major visual brain area−harbors diverse populations of neurons that control the duration of visually evoked threat and anxiety responses.

Graphical abstract

INTRODUCTION

Animals often rely on vision to detect and respond to threats. While defensive behaviors can be stereotyped (i.e., freeze or run), they are dynamically modulated by internal state, context, and environmental conditions (Blanchard et al., 1986; Wei et al., 2021). Which defensive behaviors an animal will choose and how long those behaviors last require ongoing risk assessment as threat situations evolve.

There are at least two main types of visual information animals need to consider when deciding how to react to threats. The first is visual perception of the threat itself, such as light-dark contrast created by an overhead “looming” predator (Schiff et al., 1962; Yilmaz and Meister, 2013; Branco and Redgrave, 2020). The second is the local environment; for example, the presence of shelters and overall luminance conditions (Vale et al., 2017; Heap et al., 2018). These environmental factors combine with the threat stimulus to guide the selection and duration of adaptive behaviors.

Which neural circuits process information about visual threats to yield defensive behavioral responses, and how are these circuits modulated by environmental conditions? Visual information is processed first in the retina and then routed to ~40 subcortical targets (Morin and Studholme, 2014; Seabrook et al., 2017), including the superior colliculus (SC) and the dorsal lateral geniculate nucleus (dLGN). The SC responds to specific visual features characteristic of potential threats, such as expanding dark stimuli, and sends this information to other subcortical nuclei that in turn initiate defensive behaviors (Shang et al., 2018; Ito and Feldheim, 2018; Lee et al., 2020). The dLGN relays retinal information to the visual cortex (V1) for image-forming vision (Dhande et al., 2015; Seabrook et al., 2017), but the retino-dLGN-V1 pathway is dispensable for generating adaptive responses to visual threats (Shanks et al., 2016). Indeed, mice lacking V1, or their entire neocortex altogether, still exhibit robust responses to overhead looming stimuli (Zhao et al., 2014; Shanks et al., 2016; Lee et al., 2020), indicating other subcortical nuclei serve this role.

The ventral lateral geniculate nucleus (vLGN) is a large retinorecipient thalamic structure that, in rodents, is similar in volume to the dLGN (Sherman and Guillery, 2001 ; Jones, 2007). In contrast to the well-established functional roles of the dLGN, the function of the vLGN has remained unclear. The vLGN has been traditionally thought to support non-image-forming visual functions including eye movements during sleep (Harrington, 1997; Monavarfeshani et al., 2017); yet, vLGN neurons project to the SC, the periaqueductal gray area (PAG), and the nucleus reuniens (Re) of the ventral midline thalamus−all of which have been implicated in behavioral responses to visual threats (Moore et al., 2000; Evans et al., 2018; Salay et al., 2018). The fact that the vLGN is composed of several distinct cell types and connects to various visual and non-visual subcortical targets suggests that considerable functional diversity exists within the vLGN, which has yet to be explored in the behavioral context.

We hypothesized the vLGN plays a critical role in the processing and routing of threat-related visual information to brain regions involved in autonomic and defensive behavioral control. Here, we used an array of circuit investigation approaches to identify vLGN cell types, their projection targets, and their retinal presynaptic inputs. By recording calcium activity responses in the vLGN and manipulating vLGN neurons in vivo, we discovered roles for the vLGN in biasing defensive versus non-defensive (exploratory) behavioral responses in visually evoked threat situations.

RESULTS

vLGN neurons modulate the duration of visual-threat-evoked freezing

The overhead looming assay reliably triggers defensive freezing responses or escape-to-shelter responses in mice (Yilmaz and Meister, 2013). We applied this assay using 50 repetitions of the expanding looming stimulus delivered in a period of 82 s (Figures 1A and 1B). In control mice (n = 14), the total duration of freezing ranged between 13 and 82 s (Figure 1B). With repetitive looming presentations, the duration of freezing was continuously reduced over time (Figure 1C). Tail rattling−thought to reflect confrontational responses to threats (Scott, 1966; Yang et al., 2017; Salay et al., 2018)−occurred on average ~10 times throughout the 82 s experiment and predominately occurred later in the testing period (Figures 1B and 1D). Toward the end of the testing period, mice rapidly habituated to the looming stimulus by reducing defensive behaviors and resuming exploration (Figure 1E). While the repeated presentations of looming stimuli used here and in previous studies allow for testing of underlying neural circuitry, we note that in the real-world context, descent of overhead predators likely occurs as “single trials.” In either single or repeated trials, the ability of mice to assess the validity or intensity of threats and accordingly modulate their behavior is essential.

Figure 1. vLGN neurons modulate the duration of visual-threat-evoked freezing.

(A) Experimental paradigm for assessing responses to an overhead looming stimulus.

(B) Ethogram of behavioral responses during (left of blue line) and after (right of blue line) the looming threat. Black circles above the graph represent the looming stimulus (50 expansions in 82 s). Pink lines represent freezing. Black symbols represent tail rattling.

(C and D) Quantification of freezing (C, duration) and tail-rattling (D, incidence) behaviors in response to the looming stimulus (n = 14 mice; Wilcoxon test).

(E) Cumulative frequency distribution plot of the time to habituate to the looming stimulus (n = 14 mice). Blue line indicates the total duration of the looming stimulus (82 s).

(F) Example image of vLGN^GABA neurons (pink) from a Gad2-Cre;Ai9 mouse. DAPI in blue. Coronal, bregma −2.5 mm. Scale bar, 100 μm.

(G) Viral strategy for inactivation and activation of vLGN^GABA neurons.

(H) Representative image of vLGN^GABA neurons labeled with hM4Di-mCherry (pink) from an injected Gad2-Cre mouse. DAPI in blue. Coronal, bregma −2.6 mm. Scale bar, 1 mm.

(I) Time spent freezing in response to the looming stimulus in all three treatment groups (n = 9 controls, n = 7 inactivate, n = 7 activate; one-way ANOVA).

(J) Incidence of tail rattling events in response to the looming stimulus in all three treatment groups (n = 9 controls, n = 7 inactivate, n = 7 activate; Kruskal-Wallis test).

(Kand L) Ethograms of responses during (left of blue line) and after (right of blue line) the looming stimulus in mice with their vLGN^GABA neurons inactivated (K) or activated (L).

(M) Cumulative frequency distribution plots of the time to habituate to the looming stimulus for all three treatment groups (n = 9 controls, n = 7 inactivate, n = 7 activate; Kolmogrov-Smirnov test).

For all figure panels, data are mean ± SEM. Dots on bar plots in (I) and (J) represent data from individual animals. Blue line in (E), (I), and (M) indicates the total duration of the looming stimulus (82 s). *p < 0.05; **p < 0.01; ***p < 0.001; ns, not significant. If data met normality and homogeneity of variance assumptions, parametric tests were used (ex. ANOVA). If not, nonparametric tests were used (ex. Kruskal-Wallis). See Table S1 for further details of the statistical analyses. vLGN, ventral lateral geniculate nucleus; IGL, intergeniculate leaflet; dLGN, dorsal lateral geniculate nucleus.

See also Figure S1, Table S1, and Video S1.

To explore the role of the vLGN in visual-threat-evoked defensive responses, we injected the vLGN of Gad2-IRES-Cre mice with a Cre-dependent virus encoding the neuronal activity inhibitor hM4Di (AAV-DIO-hM4D-mCherry, DREADDs; n = 7 mice) or the neuronal activity activator hM3Dq (AAV-DIO-hM3Dq-mCherry; n = 7 mice; Figures 1F–1H). In a separate group, a control virus was injected into the vLGN of Gad2-Cre mice (AAV-DIO-mCherry). Two weeks later, mice were intraperitoneally injected with the ligand clozapine-N-oxide (CNO) to either decrease (hM4Di) or increase (hM3Dq) neuronal firing of vLGN^GABA neurons as they were exposed to an overhead looming stimulus (Figure 1A).

Strong bidirectional differences in defensive behaviors to looming stimuli were observed in mice with vLGN^GABA neuronal activity manipulations. Mice expressing hM4D/CNO in vLGN^GABA neurons spent significantly more time freezing in response to looming stimuli compared with control mice (Figures 1I and 1K). Conversely, mice expressing hM3Dq/CNO spent significantly less time freezing in response to looming stimuli compared with control mice (Figures 1I and 1L).

Pronounced differences in the time course and habituation thresholds for freezing behavior were observed in mice with altered vLGN neuron activity. Inactivation of vLGN^GABA cells (by hM4Di/CNO) promoted very long-lasting freezing behaviors that persisted for minutes (ranging from ~1.5 to 3 min) compared with ~50 s in control mice (Figures 1l and 1K). Whereas control mice quickly habituated to the looming stimulus over time, 100% (n = 7/7 mice) of the hM4Di/CNO treated mice froze for the entire duration of the looming stimulus presentation and continued freezing long after the looming stimulus presentation ended (Figures 1K and 1M; blue line represents the end of the looming stimulus). By contrast, increasing activation of vLGN^GABA neurons (hM3Dq/CNO) led to more rapid habituation to the looming stimulus, compared with control mice (Figures 1I, 1L, and 1M). Similar results were observed following neural activity manipulations of all cells−not just Gad2+ neurons within the vLGN (Figure S1). Taken together, these data indicate that vLGN neurons play a key role in modulating the duration of visual- threat-evoked defensive behaviors.

vLGN neural activity scales with environmental illumination

We addressed how vLGN^GABA neurons responded to visually induced threats in freely behaving mice. Using fiber photometry, we measured bulk calcium signals from vLGN^GABA neurons while mice were exposed to an overhead looming stimulus (Figures 2A and 2B). Overall, the activity of vLGN^GABA neurons did not change relative to the expansions of the overhead looming disk (Figure 2C). Whereas the bulk calcium signal from vLGN^GABA neurons was not modulated by the visual presentation of the looming disk, calcium activity was significantly reduced during behavioral epochs of visually evoked freezing (Figures 2D and 2E).

Figure 2. vLGN neural activity scales with environmental illumination.

(A) Experimental configuration for fiber photometry recordings from vLGN^GABA neurons.

(B) Representative image of vLGN^GABA neurons labeled with GCaMP (green). The location of the fiber optic tract is indicated in yellow. DAPI in blue. Coronal, bregma −2.7 mm. Scale bar, 100 μm.

(C) Mean vLGN^GABA recording trace during looming stimuli in freely behaving animals.

(D and E) Mean vLGN^GABA recording traces (D) and population activity (E) during looming-evoked freezing behaviors (n = 5 mice GCaMP, n = 5 mice GFP control; repeated-measures ANOVA).

(F) Experimental paradigm for assessing vLGN^GABA responses to luminance changes.

(G) Representative vLGN^GABA recording trace from a mouse presented with increments and decrements of full-field illumination.

(H and I) Mean vLGN^GABA recording trace (H) and vLGN^GABA population activity (I) during increments and decrements of full-field illumination (n = 5 mice GCaMP, n = 5 mice GFP control, 3 trials per mouse; repeated-measures ANOVA).

(J) Representative vLGN^GABA recording trace from a mouse presented with blue light flashes of varying intensities from 10 lux (dark blue lines) to 10,000 lux (light blue lines above trace).

(K and L) Mean vLGN^GABA recording traces (K) and population activity (L) during light flashes of varying intensities (n = 5 mice GCaMP, n = 5 mice GFP control, 3 trials per mouse; repeated-measures ANOVA).

(M) Schematic depicting rapid luminance changes during real-world threats.

For all figure panels, data are mean ± SEM. Thin lines in (E), (I), and (L) represent data from individual animals. *p < 0.05; **p < 0.01; ***p < 0.001; ns, not significant. vLGN, ventral lateral geniculate nucleus; IGL, intergeniculate leaflet; dLGN, dorsal lateral geniculate nucleus.

See also Figure S2 and Table S1.

The vLGN is known to receive information from diverse types of retinal ganglion cells (RGCs) (for review, see Monavarfeshani et al., 2017). Thus, we postulated that the vLGN^GABA neurons may encode other types of visual information relevant for modulating visually induced threat behaviors. To assess the visual responses of vLGN^GABA neurons, we presented mice with increments (brightening) or decrements (dimming) of full-field illumination (Figures 2F–2I). We consistently observed increased activity of vLGN^GABA neurons in response to increments in environmental illumination (Figures 2G–2I). vLGN^GABA activity remained elevated for the entire duration of the light exposure (Figures 2G and 2H). While activity in vLGN^GABA neurons increased during dark-to-light transitions (i.e., brightening), their activity was significantly reduced during light-to-dark transitions (i.e., dimming; Figure 2I). This supports the idea that some or all vLGN^GABA neurons encode luminance increases.

Next, we monitored the activity of vLGN^GABA neurons while we presented mice with brief flashes of blue light (3 s) that varied in intensity from dim (10 lux) to very bright (10,000 lux) (Figures 2J–2L). Overall, the activity of vLGN^GABA neurons increased during light flashes, and those activity increases scaled in an intensity-dependent manner (Figures 2K and 2L). The high-intensity light flashes (10,000 lux) caused a greater than 2-fold increase in vLGN activity compared with that of low-intensity light flashes (10 lux; Figure 2K). In control mice with GFP expressed in their vLGN rather than GCaMP expressed in their vLGN, changes in environmental illumination did not impact bulk fluorescence responses (Figures S2).

Looming stimuli are contingent upon small luminance changes as a black disk expands overhead, ranging from no change (in the shelter) to a 5 lux peak change (in the center). Whereas the expansion of the looming disk per se did not influence the vLGN^gaba neuron population activity (Figure 2C), rapid dimming and brightening responses that could signal the onset or offset of impending danger (i.e., a shadow cast from a real-world overhead predator; Figure 2M) evoked pronounced bidirectional activity changes in vLGN^GABA neurons (Figures 2I and 2K).

vLGN neurons reduce exploration in brightly illuminated environments

Next, we assessed the role of the vLGN in the context of a simulated naturalistic exploratory behavior. In the wild, mice often reside in burrows or other dark confines. They periodically exit those confines into more illuminated environments to forage for food or nesting materials or to mate (Blanchard and Blanchard, 1989; Hu and Hoekstra, 2017). The departure from a shelter to a more well-lit territory involves an assessment of risk and elevated levels of autonomic arousal (Blanchard and Blanchard, 1989; Blanchard et al., 1995; Weber et al., 2013; Barrett et al., 2019; Fink et al., 2019).

We developed a “burrow emergence assay” (BEA) consisting of a large two-chamber arena with a lower and an upper chamber, each the size of a typical open field testing arena (50 × 50 cm). The lower chamber was enclosed and kept dark (~0 lux, condition B), while the upper chamber was brightly illuminated by an overhead light with a spectral composition similar to daylight (~1,000 lux, condition A; Figure 3A). In a separate “control” behavioral test, both the lower and upper chambers were kept dark (condition B, ~0 lux; Figures 3A and 3B). To emerge from the burrow (lower chamber), mice passed through a small tunnel directly in the center of the arena. In the control condition (where both the upper and lower chambers were kept dark), mice spent approximately equal time in each chamber (Figures 3B–3D). When the upper chamber was brightly illuminated, however, mice spent significantly less time in the upper chamber (Figures 3B–3D).

Figure 3. Inactivation of vLGN neurons reduces exploration in brightly illuminated environments.

(A) Experimental paradigm of the burrow emergence assay (BEA) performed under brightly illuminated (top; condition A, ~1,000 lux) or dark conditions (bottom; condition B, ~0 lux).

(B) Representative trace of a mouse in the top chamber during the BEA under the two lighting conditions.

(C) Cumulative percentage of time spent in the top chamber that is either brightly illuminated (in yellow, n = 20 mice) or dark (in black, n = 7 mice; one-way ANOVA).

(D) Mean latency to enter the top chamber that is either brightly illuminated (yellow bar, n = 20 mice) or dark (gray bar, n = 7 mice; Mann-Whitney U test).

(E) Viral strategy for inactivation and activation of vLGN^GABA neurons.

(F) Percentage of time spent in the top chamber under bright (left) and dark (right) conditions in all three treatment groups (n = 12 control, n = 5 inactivate, n = 5 activate; one-way ANOVA).

(G) Mean latency to enter the top chamber under bright (left) and dark (right) conditions in all three treatment groups (n = 12 control, n = 5 inactivate, n = 5 activate one-way ANOVA).

(H) Experimental paradigm for the light-dark test.

(I) Percentage of time spent in the light chamber during the light-dark test in all three treatment groups (n = 12 control, n = 5 inactivate, n = 5 activate; one-way ANOVA).

(J) Mean latency to enter the light chamber during the light-dark test in all three treatment groups (n = 12 control, n = 5 inactivate, n = 5 activate; Kruskal-Wallis test).

(K) Experimental paradigm for the open field test under brightly illuminated conditions.

(L) Percentage of time spent in the center of the arena in the open field test in all three treatment groups (n = 12 control, n = 5 inactivate, n = 5 activate; one-way ANOVA).

For all figure panels, data are mean ± SEM. Dots on bar plots represent data from individual animals. *p < 0.05; **p < 0.01; ***p < 0.001; ns, not significant.

See also Table S1 and Video S2.

Next, we explored the functional role of vLGN cells in the exploration of brightly illuminated environments. Gad2-Cre mice with either hM4Di or hM3Dq expressed in their vLGN (Figure 3E) were given CNO and tested on the BEA. Mice with their vLGN^GABA neuron activity reduced (hM4Di/CNO) spent significantly less time in the brightly illuminated top chamber relative to controls and instead resided in the dark lower chamber (burrow) for more than 90% of the 20-min test (Figure 3F, left panel). Moreover, hM4Di/CNO-treated mice did not initiate exploration of the bright upper chamber until more than 5 min had elapsed, and some mice never entered the upper chamber at all (Figure 3G, left panel). Mice with their vLGN^GABA neuron activity increased (hM3D/CNO) also took several minutes to initiate exploration of the bright upper chamber (Figure 3G); however, the total amount of time these mice spent in the upper (bright) versus lower (dark) chamber was not significantly different relative to controls (Figure 3F).

Reductions in the activity of vLGN^GABA neurons also resulted in changes in exploratory behavior under conditions where both chambers were dark (Figures 3F and 3G, right panels). These changes were less pronounced (~75% of time residing in the lower chamber), yet consistent with the observed reductions in exploration under conditions where the upper chamber was brightly illuminated. These data suggest that the vLGN adjusts exploratory behaviors in a manner that scales with environmental luminance and/or perceived risk.

To compare with other traditional risk avoidance assays, we tested mice in the light-dark test (Figures 3H–3J) and open field test (Figures 3K and 3L) while manipulating the activity of vLGN^GABA neurons (Shimada et al., 1995; Prut and Belzung, 2003). Mice with reduced vLGN^GABA neuron activity spent significantly less time in the brightly illuminated chamber in the light-dark test relative to controls (Figure 3I). In the open field test, which involves mice exploring a large brightly illuminated arena, hM4Di/CNO-induced reductions in vLGN^GABA activity caused mice to spend significantly less time (~6% total) in the center portion of the arena relative to controls (~20% total), which reflects elevated levels of anxiety (Figures 3K and 3L). Taken together, these data indicate that the function of vLGN^GABA neurons extends beyond modulating behavioral responses to looming visual threats and points to a key role for the vLGN in adjusting adaptive behaviors in response to spatial variation in lighting conditions.

Distinct behavioral influences of immediate versus sustained vLGN activation

The duration of a behavioral response to a threat can be enhanced or reduced depending on levels of arousal or stress (Li et al., 2018; Zacarias et al., 2018; Zelikowsky et al., 2018; Salay et al., 2018). To test whether the vLGN exerts a role in modulating autonomic arousal, we recorded the heart rates of freely behaving mice while manipulating the activity of their vLGN^gaba neurons (Figure 4A). Control mice and mice with hM4Di expressed in their vLGN^GABA neurons displayed no significant differences in heart rate in conditions with and without CNO (Figures 4B, 4C, and 4E). By contrast, mice with hM3Dq expressed in vLGN^GABA neurons (i.e., increased vLGN^GABA activity) exhibited sustained decreases in their heart rates following CNO injections, indicating a net reduction in autonomic arousal (Figures 4D and 4E).

Figure 4. vLGN neurons modulate duration of threat responses independently of their influence on levels of autonomic arousal.

(A) Viral strategy for inactivation and activation of vLGN^GABA neurons.

(B-D) Heart rate responses of mice following CNO injections in control mice (B), mice with their vLGN^GABA neurons inactivated (C), and mice with their vLGN^GABA neurons activated relative to saline injections (D; n = 11 control, n = 7 inactivate, n = 7 activate).

(E) Mean heart rate of mice with and without CNO across all three treatment groups (n = 11 control, n = 7 inactivate, n = 7 activate; paired t test).

(F and G) Viral strategy (F) and experimental paradigm (G) for optogenetic activation of vLGN^GABA neurons via stabilized-step function opsin (SSFO).

(H) Heart rate responses of mice prior, during, and after induction of vLGN^GABA neuron activation (n = 8 mice).

(I and J) Experimental protocol for prior activation (I) and concurrent activation (J) of vLGN^GABA neuron relative to the presentation of the looming stimulus.

(K and L) Ethograms of responses in mice with their vLGN^GABA neurons activated before (K, prior activation) or during (L, concurrent activation) the looming stimulus presentation. Gray line represents the onset (left line) and blue line represents the offset (right line) of the looming stimulus. Pink lines represent freezing. Black symbols represent tail rattling.

(M) Time spent freezing under conditions in which vLGN^GABA activation occurred before or during the looming stimulus. Thin lines represent paired data from individual animals (n = 8 mice; Wilcoxon test).

For all figure panels, data are mean ± SEM. *p < 0.05; **p < 0.01; ns, not significant.

See also Table S1.

We next asked whether the observed reduction in heart rate occurs rapidly upon activation of vLGN^GABA neurons or follows a delayed onset. To address this, we used a stabilized step-function opsin (SSFO; ChR2(C128S/D156A)) to temporally control vLGN^GABA neuron activation in freely behaving mice while we monitored their heart rates (Figures 4F and 4G). SSFO allows for consistent neuronal activation on the order of ~30 min following a brief 473 nm blue light pulse that can be deactivated with a brief 595 nm orange light pulse (Yizhar et al., 2011). Heart rate was not significantly changed during or in the time immediately following the initiation of vLGN^GABA neuron activation. Thirty minutes after initiation of vLGN^GABA neuron activation, we observed sustained decreases in heart rate (Figure 4H). The heart rate reduction continued even after the vLGN activation had ceased. Together, these data suggest that consistent activation of vLGN^GABA neurons promotes a delayed onset, yet persistent reduction in autonomic arousal.

We wondered whether the reductions in looming-induced freezing observed during vLGN^GABA activation reflected shifts in autonomic arousal. We therefore tested two questions: (1) What happens to the visual threat responses 30 min following activation of vLGN^GABA neurons (the time when heart rate is reduced)? (2) What happens to the visual threat responses immediately after activation of vLGN^GABA neurons (the time when heart rate is not yet reduced)?

To test the first question, mice received a brief pulse of blue light to initiate activation of vLGN^GABA neurons. We then waited 30 min−at which time, their heart rate is reduced (Figures 4H and 4I). Next, we deactivated vLGN^GABA neurons with orange (595 nm) light and tested their responses to looming stimuli immediately afterward (Figures 4I and 4K). We observed no change in freezing responses to looming stimuli (Figures 4K and 4M).

To test the second question, mice received a brief pulse of blue light to initiate activation of vLGN^GABA neurons and then we assessed their freezing responses to looming stimuli immediately afterward (a time when heart rate is not yet reduced) (Figures 4H and 4J). This resulted in a significant reduction in freezing to looming threats (Figures 4L and 4M). Thus, the reduction in freezing response caused by vLGN^GABA neuron activation is not the consequence of reduced autonomic arousal. Taken together, vLGN^GABA neuron activation imparts rapid influences on visual threat responses and, by contrast, imparts delayed changes in autonomic arousal.

Divergent vLGN outputs exert opposite influences on defensive behaviors

Next, we explored the role of specific output pathways of the vLGN with a particular emphasis on the outputs from vLGN^GABA neurons. We injected AAV-DIO-GFP into the vLGN of Gad2-Cre mice, thereby selectively labeling vLGN^GABA neurons and their axons with GFP (Figure 5A). We observed GFP-expressing vLGN^GABA axons in the SC (Figure 5B) and in the Re (Figures 5C and S3), both of which have been previously implicated in defensive behavioral responses to visually induced threats (Wei et al., 2015; Evans et al., 2018; Salay et al., 2018). Two-color retrograde labeling of vLGN cells that project to the SC (in green) or to the Re (in red) confirmed that the vLGN→SC and vLGN→Re circuits represent distinct vLGN output pathways (Figures 5D and 5E).

Figure 5. Divergent vLGN outputs impart opposing influences on freezing behaviors.

(A) Viral strategy for circuit tracing experiments of vLGN^GABA neuron output projections.

(B and C) Example images of vLGN^GABA neuron axons in the SC (B) and in the Re portion of the vMT (C). DAPI in blue. Coronal, bregma −4.2 mm (B) and −0.9 mm (C). Scale bars, 100 μm.

(D and E) Experimental schematic (D) and quantification (E) of retrograde-labeled vLGN neurons that project to the SC (green, vLGN→SC), to the Re (pink, vLGN→Re), or both (yellow).

(F) Viral strategy for inactivation and activation of vLGN→Re neurons (left) and vLGN^GABA→Re neurons (right).

(G and H) Quantification of freezing (G, duration) and tail-rattling (H, incidence) behaviors in response to the looming stimulus for all vLGN→Re treatment groups (left vLGN→Re: n = 9 control, n = 6 inactivate, n = 9 activate; freeze and rattle: Kruskal-Wallis test; right vLGN^GABA→Re: n = 7 control, n = 5 inactivate, n = 5 activate; freeze and rattle: one-way ANOVA).

(I) Cumulative frequency distribution plots of the time to habituate to the looming stimulus for vLGN^GABA→Re neuron treatment groups (n = 7 control, n = 5 inactivate, n = 5 activate; Kolmogrov-Smirnov test).

(J) Viral strategy for inactivation and activation of vLGN→SC neurons (left) and vLGN^GABA→SC neurons (right).

(K and L) Quantification of freezing (K, duration) and tail rattling (L, incidence) behaviors in response to the looming stimulus for all vLGN→SC treatment groups (vLGN→SC; n = 5 control, n = 6 inactivate, n = 6 activate; freeze: one-way ANOVA; rattle: Kruskal-Wallis test; vLGN^GABA→SC; n = 8 control, n = 6 inactivate, n = 6 activate; freeze and rattle: Kruskal-Wallis test).

(M) Cumulative frequency distribution plots of the time to habituate to the looming stimulus for vLGN^GABA→SC neuron treatment groups (n = 8 control, n = 6 inactivate, n = 6 activate; Kolmogrov-Smirnov test).

(N) Schematic depicting the influence of activating vLGN→Re and vLGN→SC projection neurons during the looming stimulus.

(O) Viral strategy for simultaneous activation of vLGN→Re and vLGN→SC neurons.

(P and Q) Quantification of freezing (P, duration) and tail rattling (Q, incidence) behaviors in mice with simultaneous activation of vLGN→Re and vLGN→SC neurons during the looming stimulus (n = 6 control, n = 5 activate; freeze: unpaired Student’s t test; rattle: Man-Whitney test).

(R) Cumulative frequency distribution plots ofthetimeto habituation in mice with simultaneous activation of vLGN→Re and vLGN→SC neurons (n = 6 control, n = 5 activate; Kolmogrov-Smirnov test).

For all figure panels, data are mean ± SEM. Dots on bar plots represent paired data from individual animals. Blue line in (G), (I), (K), (M), (P), and (Q) indicates the total duration of the looming stimulus (82 s). *p < 0.05; **p < 0.01; ***p < 0.00; ns, not significant. vLGN, ventral lateral geniculate nucleus; IGL, intergeniculate leaflet; dLGN, dorsal lateral geniculate nucleus; Xi, xiphoid; Re, nucleus reuniens; ventral midline thalamus, vMT; SC, superior colliculus; sSC, superficial SC; dSC, deep SC.

See also Figures S3, S4, S5, S6, and S8; Table S1 ; and Video S3.

We then either increased or decreased the activity of all vLGN neurons projecting to the Re (Figure 5F, left panel) and compared the effects to increasing or decreasing the activity specifically of Gad2+ vLGN^GABA neurons projecting to the Re (Figure 5F, right panel) during looming visual threats (Figures 5G–5I). Reducing the activity levels of all vLGN→Re or vLGN^GABA→Re did not alter the duration of time mice spent freezing, whereas increasing the activity of all vLGN→Re or vLGN^GABA→Re neurons resulted in significantly increased freezing (Figure 5G). Activation of vLGN^GABA→Re neurons resulted in dramatically prolonged defensive responses and thereby shifted the threshold for habituation to looming stimuli (Figure 5I). Notably, retrograde tracing revealed that all labeled vLGN→Re neurons were in fact GABAergic (100% Vgat+ neurons), with approximately half of all neurons specifically labeled by Gad2 (Figure S3). Monitoring bulk calcium activity of Re neurons revealed freezing-related reductions in Re population activity (Figure S4). Thus, these data suggest that the selection and expression of visual threat behaviors in part rely on the activity within the nucleus reuniens.

Since the effect of activating the vLGN→Re neurons was opposite to that of increasing the activity all vLGN neurons (which includes vLGN outputs to the Re and SC), it raised the question of whether the outputs from the vLGN to the SC may play a pivotal and perhaps even opposite role to those projecting to the Re. We therefore altered the activity of all vLGN→SC neurons or just the Gad2+ vLGN^GABA→SC neurons (Figure 5J) during looming visual threats (Figures 5K–5M). We found that activation of all vLGN→SC neurons or just vLGN^GABA→SC neurons decreased freezing and resulted in the rapid habituation to looming stimuli-consistent with the main vLGN manipulation results we observed (Figures 5K and 5M; referto Figure 1). Inactivation of all vLGN→SC neurons, but not the vLGN^GABA→SC Gad2 neurons, also decreased freezing, suggesting that another non-Gad2 cell type is responsible for this observed effect (Figure 5K).

Our data revealed two divergent vLGN^GABA output pathways to the Re and SC, each of which exert opposite influences on freezing behaviors (Figure 5N). We thus asked whether the outputs to the SC dominate the response in conditions where both outputs (vLGN→SC and vLGN→Re) are simultaneously activated. We activated both the vLGN output pathways to the SC and Re in the context of looming threats (Figures 5O–5R). This resulted in a reduction in freezing behaviors (Figure 5P), consistent with the “whole vLGN” and vLGN→SC manipulations, thereby thus revealing a dominant role for the outputs to the SC in modulating defensive behavioral responses to looming.

In contrast to the “whole vLGN” inactivation result, neither vLGN→SC nor vLGN→Re circuit manipulations influenced exploratory behaviors in the BEA (Figure S5). Moreover, none of the output-specific inactivation experiments to the SC (Figure 5G) to the Re (Figure 5K), or to the PAG (Figure S6), caused the pronounced freezing behaviors observed following “whole vLGN” inactivation. Taken together, these data implicate the role of other vLGN cell types or outputs in contributing to changes in visual-threat-induced defensive behaviors.

vLGN neurons evoke defensive behaviors in the absence of threat stimuli

Within the vLGN, there exists considerable diversity of cell types, including excitatory glutamatergic neurons as well as several subtypes of inhibitory GABAergic neurons (Harrington, 1997; Monavarfeshani et al., 2017; Sabbagh et al., 2020). Given that reducing the activity of all vLGN→SC neurons, but not the Gad2+ vLGN^GABA→SC neurons, decreased freezing (Figures 5G; schematized in Figure 6A), we decided to explore what other vLGN cell types may contribute to the modulation of visual- threat-evoked behaviors.

Figure 6. Activation of vLGN^Glut outputs trigger defensive behaviors in the absence of threat stimuli.

(A) Schematic depicting the influence of inactivating vLGN→SC and vLGN→SC GABA-specific projection neurons during the looming stimulus.

(B) Viral strategy for mapping the cell types of vLGN→SC neurons in Gad2-Cre mice.

(C) Example image of vLGN Gad2+ (green) and Gad2− (pink) neurons that project to the SC. DAPI in blue. Coronal, bregma −2.5 mm. Scale bars, 100 μm.

(D) Fraction of vLGN Cre+ (green) and Cre− (pink) neurons that project to the SC from Gad2-Cre, Vgat-Cre, and Vglut2-Cre mice.

(E) Viral strategy for inactivation and activation of vLGN→SC glutamatergic neurons.

(F and G) Quantification of freezing (F, duration) and tail-rattling (G, incidence) behaviors in response to the looming stimulus for all glutamatergic vLGN→SC treatment groups (n = 7 control, n = 5 inactivate, n = 6 activate; vLGN→SC; freeze: one-way ANOVA; rattle: Kruskal-Wallis test).

(H) Cumulative frequency distribution plots of the time to habituate to the looming stimulus for all glutamatergic vLGN→SC treatment groups (n = 7 control, n = 5 inactivate, n = 6 activate; Kolmogrov-Smirnov test).

(I) Time spent freezing before, during, and after the looming stimulus for all treatment groups (n = 7 control, n = 5 inactivate, n = 6 activate; Friedman test with Dunn’s correction).

(J) Ethograms of behavior responses in the home cage without a threat present in mice with their vLGN→SC glutamatergic neurons activated. Pink lines represent freezing. Black symbols represent tail rattling. Green circles represent running.

(K and M) Quantification of freezing (K, percentage of time), tail-rattling (L, incidence), and running (M, incidence) behaviors in the home cage without a threat present for all glutamatergic vLGN→SC treatment groups (n = 7 control, n = 5 inactivate, n = 6 activate; Kruskal-Wallis test).

(N) Experimental paradigm of the BEA performed under brightly illuminated conditions (~1,000 lux; condition A).

(O) Percentage of time spent in the top chamber under brightly illuminated conditions for all glutamatergic vLGN→SC treatment groups (n = 7 control, n = 5 inactivate, n = 6 activate; one-way ANOVA).

To explore the cell types of vLGN neurons that project to the SC, we injected a retrograde color-flipping switch virus AAVretro-Nucflox(mCherry)-EGFP into the SC of Gad2-Cre mice, which expresses mCherry in Cre-negative cells (Gad2−; in pink), but inverts to express GFP in Cre-positive cells (Gad2+; in green; Figures 6B–6D). We found that approximately two-thirds (~65%) of vLGN cells that project to the SC were Gad2+ cells. We also injected this retrograde color-flipping switch virus into the SC of Vgat-Cre and Vglut2-Cre mice. The majority of vLGN cells that project to the SC were Vgat+ (~90%; i.e., GABAergic), which consists of both Gad2 and Gad1 cell types (Figure 6D; Sab- bagh et al., 2020). The remaining vLGN→SC neurons (~10%) were Vglut2+ (i.e., glutamatergic; Figure 6D).

We next asked whether the glutamatergic neurons in the vLGN that project to the SC participate in the modulation of visual threat behaviors. We addressed this question by using an intersectional approach to selectively express hM4Di or hM3Dq in glutamatergic vLGN→SC neurons of Vglut2-Cre mice (Figure 6E). Next, mice were injected with CNO and subsequently exposed to an overhead looming stimulus (Figures 6F–6I). Mice with their vLGN^Glut→SC neurons inactivated spent significantly less time freezing in response to the looming stimulus relative to controls (Figure 6F). These data are consistent with the hypothesis that a non-Gad2 vLGN→SC neuronal subtype contributes to the reduction in visual-threat-induced freezing behaviors (Figure 6A).

In contrast to the observed reductions in looming-induced freezing following vLGN^Glut→SC inactivation (Figure 6F), we observed increased freezing behaviors following vLGN^Glut→SC activation (Figure 6I). Control mice and the mice that had their vLGN^Glut→SC neurons inactivated performed almost all freezing behaviors during the period in which the looming stimulus was present. By contrast, mice with their vLGN^Glut→SC neurons activated froze during the entire testing period before, during, and after the looming stimulus presentation (Figure 6I). Remarkably on the very first trial, these mice spent approximately half of the time freezing (~50%) well before the onset of the looming stimulus (Figure 6I). Furthermore, mice with their vLGN^Glut→SC neurons activated never habituated and, indeed, continued to perform defensive freezing and tail rattle behaviors for the entire duration of the testing period (over 15 min; Figure 6H).

We then asked whether manipulations of vLGN^Glut→SC neurons results in the generation of defensive behaviors in different environments−such as in the safety of the mouse’s own home cage. Vglut2-Cre mice with either hM4Di or hM3Dq expressed in their vLGN→SC neurons were given CNO and monitored for 5 min in their home cage environment in the dark (Figure 6J). Activation of vLGN^Glut→SC neurons resulted in the generation of diverse, long-lasting defensive behaviors even in the absence of looming stimuli, with the mice simply confined to their home cage (Figures 6J–6M). All mice (6/6) with their vLGN^Glut→SC neurons activated froze in their home cage in contrast with control mice that did not freeze at all in their home cage (0/6 mice; Figures 6J and 6K). Control mice and mice with their vLGN^Glut→SC neurons inactivated did not perform any tail rattling or running behaviors in their home cage. By contrast, we observed robust tail rattling (Figures 6J and 6L) as well running behavior (Figures 6J and 6M) in mice with their vLGN^Glut→SC neurons activated. Similarly, activating vLGN^Glut→SC neurons caused mice to completely avoid exploration of the upper, well-illuminated chamber in the BEA, whereas there was no effect of inactivating vLGN^Glut→SC neurons.

Robust bidirectional influences on freezing behaviors to looming stimuli were observed following inactivation and activation of all vLGN^Glut neurons, not just the ones that project to the SC (Figure S7). The influences on freezing and exploratory behaviors following vLGN glutamatergic neuron manipulations were opposite to those observed following vLGN GABAergic manipulations. Together, these data reveal that distinct circuits arising from the vLGN could play opposing roles to balance the demand to either respond (defend) or not respond (explore) in threat situations.

DISCUSSION

Rapid and dynamic control of defensive behaviors requires ongoing assessment of threats as they evolve, such as shifts in the trajectory and proximity of predators (Blanchard et al., 1986; Ydenberg and Dill, 1986; Fanselow and Lester, 1988). To be adaptive, defensive responses must also adjust to context and environmental factors including the presence of shelters or ambient illumination (Blanchard and Blanchard, 1989; Liang et al., 2015; De Franceschi et al., 2016). The neural circuits that confer behavioral flexibility to visual threats remain unclear.

Here, we find that neurons in the vLGN scale their activity with environmental luminance in an intensity-dependent manner and strongly modulate the presence and persistence of defensive behaviors to visual threats. Typically, “freeze,” “escape,” or “confrontational” behaviors are viewed as all or none, but in the ethological context, each can be of varying duration, which plays a crucial role in whether they are adaptive or not (Blanchard and Blanchard, 1989; Fanselow and Lester, 1988; Eilam, 2005). The fact that vLGN neurons powerfully adjust the absolute duration of looming-evoked freezing and that vLGN neurons bidirectionally shift the threshold for habituation indicates that the vLGN, together with its downstream targets: the Re and SC, integrate sensory stimuli with internal states of arousal.

Our findings and those of recent electrophysiological recordings from mouse vLGN neurons (Ciftcioglu et al., 2020) show that vLGN neuronal activity scales with increments in illumination (i.e., brightening; Figure 2). Furthermore, we find that vLGN^GABA neuron activity is rapidly reduced when mice view decrements in illumination. Real-world predators generally impart a reduction in luminance as they descend and cast shadows from overhead. As a predator casts its shadow, vLGN^GABA neurons could release their downstream targets from inhibition and thereby promote robust defensive responses (Figure 7). Indeed, work across many species−from flies to primates−has demonstrated the importance of rapid reductions in luminance for initiating defensive reactions (Schiffet al., 1962; Gibson et al., 2015; Heap et al., 2018).

Figure 7. Summary: The vLGN bidirectionally regulates responses to visual threats.

For all figure panels, data are mean ± SEM. Dots on bar plots represent paired data from individual animals. Blue line in (F) and (H) indicates the total duration of the looming stimulus (82 s). *p < 0.05; **p < 0.01; ***p < 0.00; ns, not significant. vLGN, ventral lateral geniculate nucleus; vLGNe, vLGN external subdivision, vLGNi, vLGN internal subdivision; IGL, intergeniculate leaflet; dLGN, dorsal lateral geniculate nucleus; SC, superior colliculus.

See also Figures S7 and S8 and Table S1.

Reducing the activity of vLGN^GABA neurons promoted long-lasting freezing behaviors when looming stimuli were present (Figure 1). However, in the absence of threat stimuli, vLGN^GABA activity manipulations did not induce freezing. Thus, our findings support a modulatory role of the vLGN^GABA neurons in this context. By contrast, increasing the activity of vLGN^Glut neurons that project to the SC elicited long-lasting defensive freezing, running, and tail-rattling behaviors even in the absence of a threat (Figure 6). It is notable that the vLGN^Glut neurons can promote distinct categories of threat behaviors (i.e., freeze, flight, and confrontational responses) on their own, indicating the vLGN has the potential to both modulate and to directly generate a diverse repertoire of defensive behaviors.

In the wild, mice must balance their need to forage for essential resources with the risk of exposing themselves to detection and harm from predators (Stephens and Krebs, 1986; Lima and Dill, 1990). Mice are largely nocturnal (mostly active at night) and often reside in burrows or other dark confines, especially during daylight (Johnson, 1926; Behney, 1936; Metz et al., 2017) We designed the BEA to simulate naturally occurring contexts of exploratory behavior. We find that distinct cell types in the vLGN contributed to unique aspects of exploration in the BEA (Figure 3). Reducing the activity of vLGN^GABA neurons or increasing the activity of vLGN^Glut neurons promoted risk aversion in the BEA; the mice spent less time in the brightly illuminated area. The activity manipulations that reduced exploration in the BEA congruently promoted long-lasting defensive responses to looming visual threats. These findings extend the role of the vLGN beyond simply modulating responses to looming threats and suggest a broader role in risk-associated behaviors.

In addition to luminance-induced changes in vLGN neuron activity, we find that vLGN^GABA neurons reduce their activity during looming-induced freezing (Figure 2). Therefore, the vLGN may be a key node for convergence of visual and non-visual information about behavioral state (Davidowa and Albrecht, 1992; Kolmac and Mitrofanis, 2000; Horowitz et al., 2004; Pienaar et al., 2018). Indeed, transsynaptic tracing (Schwarz et al., 2015) revealed that vLGN neurons receive inputs from several RGC types and from non-visual sources (Figure S8). The vLGN is composed of an external and an internal subdivision demarcated by RGC axonal arbors (Kolmac and Mitrofanis, 2000; Cosenza and Moore, 1984). vLGN neurons that target the SC or the Re reside in both divisions and thereby may pool information from visual as well as from non-visual sources.

Separate populations of vLGN neurons project to the Re versus the SC and exert opposing influences on defensive behaviors to visual threat (Figure 5). Mice with their vLGN^GABA→SC neurons activated spent less time freezing and rapidly habituated to looming stimuli by resuming exploratory behaviors. By contrast, mice with their vLGN^GABA→Re neurons activated spent significantly more time freezing and took longer to habituate to looming stimuli. These findings align with previous work showing that Re/vMT activity reduces with freezing (Figure S4; Salay et al., 2018) and that inactivation of Re/vMT increases freezing behaviors to threats (Salay et al., 2018; Ramanathan et al., 2018). The presence of excitatory and inhibitory neurons in the vLGN is notable; glutamatergic and GABAergic populations within the vLGN promote opposite influences on defensive freezing and the threshold to habituation (Figure 7). Thus, it appears the vLGN and its downstream targets are poised to modulate the duration of defensive behaviors in push-pull fashion−a common theme for thalamic circuits (Jones, 2007) and hypothalamic control of motivated behaviors (Hong et al., 2014; Wei et al., 2021; Karigo et al., 2021; Zhang et al., 2021).

The rodent vLGN is homologous to the primate pregeniculate nucleus and the cat geniculate wing (Niimi et al., 1963; Nakamura and Itoh, 2004; Jones, 2007). Traditionally, the vLGN was thought to be involved in circadian-related phenomena by virtue of the fact that it receives input from intrinsically photosensitive “melanopsin” RGCs (M1 ipRGCs; Harrington, 1997; Berson et al., 2002; Hattar et al., 2006; Chen et al., 2011). However, we and others have observed that the vLGN receives diverse types of retinal inputs in addition to input from M1 ipRGCs (Figure S8; Osterhout et al., 2011; Dhande et al., 2015; Sonoda et al., 2020; Beier et al., 2021). Recent work points to a role for the vLGN in light-mediated effects on depression and spatial memory−processes that occur on the timescale of days (Huang et al., 2019, 2021). We find that increasing the activity in vLGN^GABA neurons induced immediate changes in defensive responses within seconds, yet delayed changes in autonomic arousal that took many minutes (Figure 4). Together, these data point to the considerable functional diversity of this major thalamic area both based on its connectivity and the broad range of timescales with which it imparts its effects.

Given the presence of the vLGN and its homologs across species, and its role in defensive and anxiety-related behaviors shown here, the vLGN is an interesting structure to consider in the context of depression, generalized anxiety, and phobias. In each of these disorders, sensory stimuli such as light, sound, and time of day have been shown to modulate the presence and scale of adverse behavioral and emotional responses (Stephenson et al., 2012; Cho et al., 2015; Liberman et al., 2017).

STAR★METHODS

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Andrew Huberman (adh1@stanford.edu).

Materials availability

This study did no generate new unique reagents.

Data and code availability

• All data reported in this paper will be shared by the lead contact upon request.

• This paper does not report original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Mice

All protocols conducted according to National Institutes of Health (NIH) guidelines for animal research and were approved by the Stanford University (IACUC). Mice were housed under a 12-h light/dark cycle with food and water provided ad libitum. Gad2-IRES-Cre (stock 028867; Taniguchi et al., 2011), Vgat-IRES-Cre (stock #028862; Vong et al., 2011), Vglut2-IRES-Cre (stock 028863; Vong et al., 2011), Ai9 (stock 007909; Madisen et al., 2010) and C57BL/6J (stock 000664) imported from Jackson Laboratory. Adult male and female mice were 8–12 weeks old at the time of viral injection. Animals were randomly assigned to different experimental conditions. Group sample sizes were chosen on the basis of previous studies.

METHOD DETAILS

Surgery

Mice were anesthetized with 1.5%−3.0% isoflurane and given the analgesic buprenorphine (subcutaneously, 0.3 mg kg⁻¹). A small craniotomy hole was drilled with a dental drill. Using a Nanoject II (Drummond) injector, 0.3–0.5 μL of viral vectors (titer, 10¹² particles per ml) were injected into the vLGN (bregma: −2.5 mm, midline: 2.3 mm, dorsal surface: −3.3 mm), the Re/vMT (bregma: −1.0 mm, midline: 0.2 mm, dorsal surface: −4.2 mm), the SC (bregma: −3.8 mm, midline: 0.65 mm, dorsal surface: −1.5 mm), the PAG (bregma: −4.6 mm, midline: 0.35 mm, dorsal surface: −2.2 mm), the mPFC (bregma: +2.5 mm; midline: 0.25 mm; dorsal surface 1.5 mm) or the BLA (bregma: −1.3 mm, midline: 3.0 mm, dorsal surface: 4.75 mm). All injections were bilateral except those at the midline (Re/vMT). The virus was slowly infused, and 10 min after infusion the glass pipette (with a resistance of approximately 0.5 MΩ) was retracted. Half of the mice in each cage were randomly assigned to either treatment or control groups. All control mice were treated with the same experimental procedures, but a control virus was injected instead.

For activating or inactivating vLGN^GABA neurons, AAV-DIO-hM3Dq-mCherry or AAV-DIO-hM4Di-mCherry was injected into the vLGN of Gad2-IRES-Cre mice. For activating or inactivating all vLGN neurons, AAV-hM3Dq-mCherry or AAV-hM4Di-mCherry was injected into the vLGN of WT mice. For activating or inactivating vLGN^Glut neurons, AAV-DIO-hM3Dq-mCherry or AAV-DIO-hM4Di-mCherry was injected into the vLGN of Vglut2-IRES-Cre mice.

For activity manipulations of Vlgn→Re neurons, AAVretro-Cre was injected into the Re. In the same mice, AAV-DIO-hM3Dq- mCherry or AAV-DIO-hM4Di-mCherry was injected into the vLGN. For activity manipulations of vLGN→SC neurons, AAVretro- Cre was injected into the SC and either AAV-DIO-hM3Dq-mCherry or AAV-DIO-hM4Di-mCherry was injected into the vLGN. For simultaneous activation experiments, AAVretro-Cre was injected into the Re, AAVretro-Cre-eGFP was injected into the SC, and AAV-DIO-hM3Dq-mCherry was injected into the vLGN. For inactivating vLGN→PAG neurons, AAVretro-Cre was injected into the PAG and AAV-DIO-hM4Di-mCherry was injected into the vLGN. Mice were tested after at least three weeks of recovery.

For specific activity manipulations of vLGN GABAergic projection neurons, AAVretro-DIO-FLPo was injected into the downstream target (Re or SC) and either AAV-fDIO-hM3Dq-mCherry or AAV-fDIO-hM4Di-mCherry was injected into the vLGN of Gad2-IRES-Cre mice. For specific activity manipulations vLGN glutamatergic projection neurons, AAVretro-DIO-FLPo was injected into the SC and either AAV-fDIO-hM3Dq-mCherry or AAV-fDIO-hM4Di-mCherry was injected into the vLGN of Vglut2-IRES-Cre mice.

For fiber photometry recordings, AAV-EF1a-DIO-GCaMP6m was injected into the vLGN of Gad2-IRES-Cre mice and a fiber optic cannula was implanted over the vLGN. Control mice were injected with AAV-DIO-GFP instead of GCaMP6 and a fiber optic cannula was implanted over the vLGN. For vMT photometry recordings, AAV-hSyn-GCaMP6m was injected into the vMT and a fiber optic cannula was implanted over the vMT. Mice were given at least two weeks to recover before behavioral testing.

For optogenetic experiments, AAV-EF1a-DIO-hChR2(C128S/D156A)-mCherry was injected into the vLGN of Gad2-IRES-Cre mice and a fiber optic cannula was implanted over the vLGN. Control mice were injected with AAV-DIO-mCherry and a fiber optic cannula was implanted over the vLGN.

To determine the outputs of the vLGN inhibitory cells, AAV-DIO-ChR2-eYFP was injected into the vLGN of Gad2-IRES-Cre mice. For dual-color retrograde tracing experiments, the Re was injected with AAVretro-mCherry and, in the same mouse, AAVretro-GFP was injected into the SC. To determine the cell types of vLGN projection neurons, AAVretro-EF1a-Nuc-flox(mcherry)-eGFP was injected into the Re or SC of Gad2-IRES-Cre mice, Vgat-IRES-Cre mice and Vglut2-IRES-Cre mice.

To examine the location within the vMT that vLGN cells innervate, AAV-DIO-GFP was injected into the vLGN of Gad2-IRES-Cre mice. The same mouse was injected with CTβ–647 (far red) into the mPFC to label the Nucleus Reuniens and with CTβ–594 (red) into the BLA to label the Xiphoid Nucleus. The vMT was examined for the location of vLGN axons relative to the locations of CTβ–647 and CTβ–594 labeled cell bodies.

For vLGN input-output tracing, the Re or SC was injected with AAVretro-Cre and the vLGN was injected with AAV-hSyn-flex-TVA-P2A-EGFP-2A-oG. Three weeks later, ENVA-ΔG-Rabies-mCherry was injected into the vLGN. Mice were housed in a biosafety room for 4–6 days to allow the rabies virus to infect and expresses mCherry in presynaptic cells. Retinas were harvested and immuno- stained prior to imaging. The entire brain was sectioned and imaged using a Zeiss slide scanner.

Fiber photometry

To measure bulk florescence, mice with fiber optic implants were connected to an optical fiber (400 μm diameter, 0.57 NA; Doric Lenses) to both deliver excitation of light and collect emitted florescence. The light power at the fiber tip was about 0.01–0.02 mW to minimize bleaching. We used both 405 nm and 465 nm LEDs as an excitation source for performing Ca²+ independent (isosbestic control) and Ca²+ dependent (from GCaMP6) measurements, respectively. Each excitation wavelength (405 and 465 nm) was sinusoidally modulated at a distinct carrier frequency (208 Hz and 572 HZ, respectively) that is demodulated to recover the original calcium sensor response (i.e., lock-in amplification; Lerneret al., 2015). This modulation step minimizes contamination of the calcium signal by changes in overall ambient light and low-frequency noise (refer to Figure S2). The fluorescence emitted by GCaMP6 was filtered through a GFP emissions filter and focused on a photoreceiver using a lens (Doric Lenses). The photoreceiver was sampled at 12 kHz and each modulated signal was independently recovered. The outputs were then low pass filtered at 12 HZ and decimated at a factor of 200. The isosbestic control signal was aligned to and subtracted from the 465 nm signal. The fluorescence change (ΔF/F) was calculated as (F-F₀)/F₀, where F₀ is the least-squares linear fit of the whole data series. For z-scored data, traces were z-score normalized before averaging.

Mice were habituated to the optic fiber cable and head-fixation prior to behavioral testing. For behavior testing, mice were dark adapted and presented with various visual stimuli while recording from vLGN and vMT neurons using fiber photometry. This includes blue light flashes ranging from 10 lux to 10,000 lux, full-field luminance steps from 100 lux to 1000 lux and looming stimuli under both freely moving and head-fixed conditions.

Histology

Mice were transcardial perfused with saline and followed by 4% paraformaldehyde (PFA). Brains were collected and post-fixed overnight, cryoprotected in 30% sucrose and sectioned at 45 μm (coronally). To enhance the mCherry signal, brains were kept at 4 °C overnight with rabbit-anti-DsRed (1:1000,Takara Bio Cat# 632496). For secondary detection, Alexa Fluor 594 donkey anti-rabbit was used. Brain tissue was imaged with a Zeiss LSM 880 Airyscan confocal microscope or a Zeiss AxioScan microscope (Zeiss, Germany). Optic fiber placement above the vLGN was verified for fiber photometry studies. Viral injection locations were confirmed. For vLGN injections, viral injections were located in both the external and internal divisions of the vLGN in all mice. Mice with injections that missed the target or spread into nearby areas (ex. dLGN) were excluded.

For transsynaptic tracing experiments, the eyes were removed following transcardial perfusion and post-fixed for 1 hr. For whole mount, the retinas were extracted and then relieving cuts were performed. Retinas were processed and immunostained as previously described (Huberman et al., 2008). Primary antibodies used were: mouse anti-SMI-32 (1:1000, Covance, Cat# SMI-32P), rabbit anti-melanopsin (1:1000, Advanced Targeting Systems Cat# AB-N39), guinea pig anti-VAChT (1:1000, Millipore, Cat #AB1588) and goat anti-ChAT (1:100, Millipore, Cat# AB144P). Species specific secondary antibodies conjugated to Alexa Flour 488 or 647 (1:500) were used. All retinas were imaged with a Zeiss LSM 880 Airyscan confocal microscope.

Clozapine N -oxide delivery

For chemicogenetic experiments, mice were intraperitoneal injected with clozapine N-oxide (CNO, Tocris; 5.0 mg kg⁻¹ for mice with hM4Di, 1.5 mg kg⁻¹ for mice with hM3Dq) 40 minutes prior to behavioral testing.

Optogenetic stimulation

Mice with fiber optic implants were connected to an optic fiber (200 μm diameter, 0.22 NA; Doric Lenses) and allowed to habituate before behavioral testing. The optic fiber was connected to either a 473 nm laser to deliver blue light (Shanghai Laser and Optics Century) or a 595 nm LED (Thor Labs) to deliver orange light. The light power at the fiber tip was set to about 5–10 mW. For stabilized step-function opsin activation, one 5 s blue light pulse was delivered. For deactivation, one 5 s orange light pulse was delivered.

Behavioral paradigms

Mice were handled for three days prior to behavior testing. All behavior was performed during consistent hours in the afternoon. Male and female mice were used in about equal proportions in each treatment group. No significant differences in behavioral responses were observed between male and female mice testing. Mice that were tested in several behavior assays were given at least 48 h rest between tests. All behaviors were performed and scored blind to treatment group.

Looming behavioral assay

Mice were placed in a glass chamber (50 cm × 25 cm × 40 cm) with an overhead 24-inch LCD monitor facing downward to display the visual stimulus. The floor and three walls of the chamber were covered with a matte coating (Krylon) to prevent reelections of the looming stimulus. The chamber was dimly illuminated (~15 lux measured in the center of the arena, ~0 lux measured in the shelter). A black shelving board was placed on one side of the chamber to provide shelter. A top-view camera (Yi Action Camera) and a side- view camera (Panasonic-HC-W850) were used to record the mouse’s behavior during the session.

Mice were given 10 minutes to habituate to the arena. Then, the looming stimulus (black expanding disc on a white background) was manually triggered when the mouse entered the center of the arena and continually presented over 82 s. This involved 10 continuous repetitions of 5 dark expanding circles with 3 s between each set. The disk expanded from 4 to 20 cm (~5 to 25° of visual angle) over 600 ms at which it maintained the same size for 500 ms before repeating. The overall luminance in the chamber decreases (i.e., dims) as the black disk expands. Peak looming-induced luminance changes occurred directly in the center (~5 lux change), while the luminance changes in the shelter was negligible (~0 lux change).

For optogenetic experiments, mice with either stabilized step-function opsin or a control virus were tested twice in the looming assay. In one condition, mice received a short pulse of blue light to activate the vLGN neurons and then immediately after were exposed to the looming stimulus. In the other condition, mice received a short pulse of blue light and then 30 minutes later received a short pulse of orange light to deactivate the vLGN neurons. Immediately after the orange light, mice were exposed to the looming stimulus. These two conditions were counterbalanced to control for habituation effects. For all other experiments, mice were tested only once owing to the fact that mice habituate to the looming stimulus even during one trial of looming presentation (referto Figure 1).

Locomotor behaviors of the mice were recorded and analyzed automatically with a video tracking system (Biobserve software). Behaviors were scored blind to treatment group based on previous criterion (Salay et al., 2018). Briefly, freezing was defined as episodes lasting three seconds or more of complete immobility. Total freezing is the summation of all freezing bouts. Tail rattling was defined as an event in which the mouse’s tail rapidly moved back and forth (1 bout = 1 s of tail rattling). Running events were defined as an event in which the mouse’s speed was more than two times the average speed prior to the looming onset. These events occurred infrequently and thus were characterized but not included in our analyses. Ambulation was defined as all other locomotor behaviors. Habituation was defined as the complete cessation of defensive behaviors.

Heart rate measurements

Heart rate was continuously measured using a pulse oximeter and then the number of beats were reported at 1 Hz intervals (MouseOx Plus Software; Starr Life Sciences) in freely moving animals in their home cages. Heart rate values ranged from ~500 to 780 bmp. The pulse oximeter can track heart rates from 90 to 900 bpm. To allow the red light to pass through to the detector, the necks of mice were shaved at the location that the detectors were placed. For chemogenetic experiments, recordings were obtained for 10 minutes in conditions in which mice received either CNO or saline injections. For optogenetic experiments, recordings were obtained for ~1hr including before, during and after stabilized step-function opsin activation.

Burrow emergence assay (BEA)

Mice were placed into a two-chamber acrylic box (50 cm x 50 cm x 5.5 cm below, 50 cm x 50 cm x 50 cm above). The lower chamber is dark (0 lux) and the upper chamber is brightly illuminated by an overhead light with spectral composition similar to daylight (~1000 lux; Sylvania; referto Keenan et al., 2016). A small tunnel (4.5 cm hole) in the center of the arena separates the two chambers. Under the hole, is a short platform (1.7 cm) to allow mice to enter the upper and lower chamber with ease. Mice were placed in the dark chamber and monitored for 20 minutes. The latency to enter the brightly illuminated upper chamber was recorded. The mice were tracked with automated tracking software (Biobserve) and the time spent in each chamber was recorded.

Light-dark test

Mice were placed into a two-chamber acrylic box (40 cm × 20 cm × 25 cm, each chamber 20 cm × 20 cm). One chamber is dark (0 lux) and the other is brightly illuminated by an overhead light (~500 lux). Mice were placed in the dark chamber and monitored for 5 minutes. The latency to enter the brightly illuminated chamber was recorded. The mice were tracked with automated tracking software (Biobserve) and the time spent in each chamber was recorded.

Open-field test

Mice were placed in the outer portion of the open-field chamber (50 cm × 50 cm) and monitored for 5 minutes. The chamber was brightly illuminated by an overhead with a light with spectral composition similar to daylight (~1000 lux, Sylvania; refer to Keenan et al., 2016). Total time and distance in the center (36 cm × 36 cm) were analyzed using automated tracking software (Biobserve).

Home cage behavioral monitoring

Mice were placed alone in their home cages in the dark and were monitored for 5 minutes following CNO injections. The mice were tracked with automated tracking software (Biobserve). Freezing, tail rattling and running behaviors were scored as defined in the looming behavioral assay (see above).

Quantification of axon density and cell counts

For mice with injections to the vLGN, we quantified the relative density of labeled axons in the Re versus Xi portions of the vMT. For mice with CTβ injections to the mPFC or BLA, we quantified the number of retrogradely labeled cells within the Re and Xi. For Gad2-IRES-Cre mice with injections of AAV-DIO-hM3D-mCherry, we quantified the number of labeled hM3D/mCherry+ cells within the external and internal division of the vLGN. For mice with injections of AAVretro-EF1a-Nuc-flox(mcherry)-eGFP, we quantified the number of retrogradely labeled GFP+ and mCherry+ cells within the vLGN. The brains were imaged on an automated slide scanner (Zeiss). A brain map was overlain on the digital image to identify the appropriate regions using landmark structures, ventricles and optic tract as reference guides. An experimenter blind to condition analyzed the images in ImageJ. For axon density, the mean fluorescence intensity was measured in the Re and the Xi portion of the vMT. For cell counts, the number of labeled cells were analyzing using the automated ImageJ cell counter.

QUANTIFICATION AND STATISTICAL ANALYSIS

All statistical analyses were performed using R v.3.6.3. (https://www.r-project.org/) and Prism 9. The sample sizes were chosen based on common practice in animal behavior experiments. No statistical methods were used to predetermine sample size. Data were first tested for normality using a Shapiro-Wilk test and tested for homogeneity of variance using a Levene’s test. If data met normality and homogeneity of variance assumptions, parametric tests were used (for example, Student’s t test, one-way or repeated-measures ANOVA with Tukey’s multiple comparison). If not, non-parametric tests were used (for example, Mann-Whitney U test or Kruskal-Wallis with Dunn’s multiple comparison test). Paired tests were used to compare within-group repeated-measures data (for example, Wilcoxon signed rank test and Friedman test with Dunn’s multiple comparison). For data with repeated-measures, the mean of repeated trials within an individual animal was computed and then statistical tests were performed across animals. All statistical tests were two-tailed. Significance levels are indicated as follows: *p < 0.05; **p < 0.01; ***p < 0.001. For representative images, similar results were obtained in at least three independent trials. Refer to Table S1 for full statistical analyses.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit polyclonal anti-DsRed	Takara Bio	Cat# 632496; RRID: AB_10013483
Mouse monoclonal anti-SMI-32	Covance	Cat# SMI-32P; RRID: AB_2314912
Rabbit polyclonal anti-melanopsin	Advanced Targeting Systems	Cat# AB-N39; RRID: AB_1608076
Guinea pig polyclonal anti-VAChT	Millipore	Cat# AB1588; RRID: AB_2187981
Goat polyclonal anti-ChAT	Millipore	Cat# AB144P; RRID: AB_2079751
Bacterial and virus strains
AAV2-hSyn-DIO-hM4Di-mCherry	Addgene; Krashes et al., 2011	Addgene viral prep # 44362-AAV2
AAV2-hSyn-DIO-hM3Dq-mCherry	Addgene; Krashes et al., 2011	Addgene viral prep # 44361-AAV2
AAV2-hSyn-hM4Di-mCherry	Addgene	Addgene viral prep # 50475-AAV2
AAV2-hSyn-hM3Dq-mCherry	Addgene	Addgene viral prep # 50474-AAV2
AAVDJ-hSyn-fDIO-hM4Di-mCherry	Stanford Vector Core	N/A
AAVDJ-hSyn-fDIO-hM3Dq-mCherry	Stanford Vector Core	N/A
AAV2-hSyn-DIO-mCherry	Addgene	Addgene viral prep # 50459-AAV2
AAVretro-EF1a-Nuc-flox(mCherry)-eGFP	Addgene; Back et al., 2019	Addgene viral prep # 112677-AAVrg
AAVretro-hSyn-GFP	Addgene	Addgene viral prep # 50465-AAVrg
AAVretro-hSyn-mCherry	Stanford Vector Core	N/A
AAVretro-hSyn-Cre	Addgene	Addgene viral prep # 105553-AAVrg
AAVretro-hSyn-HI-GFP-Cre	Addgene	Addgene viral prep # 105540-AAVrg
AAVretro-pEF1a-DIO-FLPo	Addgene; Zingg et al., 2017	Addgene viral prep # 87306-AAVrg
AAV-DJ-EF1a-DIO-hChR2(C128S/D156A)-mCherry	Stanford Vector Core	N/A
AAV-DJ-EF1a-DIO-ChR2(H134R)-eYFP	Stanford Vector Core	N/A
AAV-DJ-EF1a-DIO-GCaMP6m	Stanford Vector Core	N/A
AAV8-hSyn-flex-TVA-P2A-EGFP-2A-oG	Salk Vector Core	Cat# 85225
ENVA-ΔG-Rabies-mCherry	Salk Vector Core	Cat# 32636
Chemicals, peptides, and recombinant proteins
Alexa Fluor 647-conjugated Cholera Toxin Subunit B	ThermoFisher	Cat# C34778
Alexa Fluor 594-conjugated Cholera Toxin Subunit B	ThermoFisher	Cat# C34777
Clozapine N-oxide (CNO)	Tocris	Cat# 4936
Experimental models: Organisms/strains
Mouse: C57BL/6J	The Jackson Laboratory	JAX:000664
Mouse: Ai9	The Jackson Laboratory	JAX:007909
Mouse: Gad2-IRES-cre	The Jackson Laboratory	JAX:028867
Mouse: Vgat-ires-cre	The Jackson Laboratory	JAX:028862
Mouse: Vglut2-ires-cre	The Jackson Laboratory	JAX:028863
Software and algorithms
MATLAB	MathWorks	https://www.mathworks.com/
R v.3.6.3.	R Foundation	https://www.r-project.org/
ImageJ	NIH	https://imagej.nih.gov/ij
Prism 9	GraphPad Software	https://www.graphpad.com/

Highlights.

• vLGN neurons modulate the duration of visually evoked threat responses

• GABAergic vLGN neurons encode rapid changes in environmental illumination

• Divergent vLGN^GABA outputs to the Re and SC exert opposing influences on freezing

• vLGN^Glut→SC neurons promote defensive reactions even in the absence of a threat