The Grueneberg ganglion controls odor-driven food choices in mice under threat

The ability to efficiently search for food is fundamental for animal survival. Olfactory messages are used to find food while being aware of the impending risk of predation. How these different olfactory clues are combined to optimize decision-making concerning food selection remains elusive. Here, we find that chemical danger cues drive the food selection in mice via the activation of a specific olfactory subsystem, the Grueneberg ganglion (GG). We show that a functional GG is required to decipher the threatening quality of an unfamiliar food. We also find that the increase in corticosterone, which is GG-dependent, enhances safe food preference acquired during social transmission. Moreover, we demonstrate that memory retrieval for food preference can be extinguished by activation of the GG circuitry. Our findings reveal a key function played by the GG in controlling contextual food responses and illustrate how mammalian organisms integrate environmental chemical stress to optimize decision-making. Julien Brechbühl et al. show that the Grueneberg ganglion olfactory subsystem is necessary for deciphering the threatening or safe qualities of unfamiliar food based on olfactory or social signals, respectively, in mice. These results highlight the role of this subsystem in optimizing decision-making strategies related to food preference by integrating environmental cues.

A nimals continuously assess the environmental risks and, for survival, may need to compromise between the gain of feeding and the potential cost of exposure to danger 1 . Especially, when confronted to food scarceness, behavioral tradeoffs need to be made 2,3 . Olfactory senses play an important role in this process as they help to find food but also to sense and avoid threatening situations [4][5][6][7][8][9] such as predation. For rodents, the attraction to food is primarily determined by familiarity, as scents of known food are innately favored 2,10 . These scents are detected by olfactory sensory neurons (OSNs) found in the main olfactory epithelium (MOE) 11,12 . Food familiarity can be acquired via communication between individuals, by the so-called "social transmission of food preference" (STFP), a behavioral process that occurs by sniffing the breath of conspecifics and making an association between the novel food scent and the presence of the endogenously produced carbon disulfide gas (CS 2 ) 4,13,14 . For this process to occur, the concomitant recognition of the food odorant by OSNs and the detection of the CS 2 by guanylyl cyclase-D neurons (GC-D) of the MOE is essential. On the other hand, odors that signal an impending danger are involuntarily left behind by predators 15,16 or emitted by stressed congeners as warning alarm pheromones 9,17 . We showed previously that these danger and warning cues are detected in a specialized olfactory subsystem, the Grueneberg ganglion 9,15 (GG). They generate, in the recipient mouse, stereotypical fear-related behaviors, such as freezing and risk assessment as well as stress-related systemic responses 15,18 .
In this study, we provide evidence and new biological insights on how mice exploit threatening scents to optimize and control their food selection. Using gene-targeted, surgically treated and sham-operated mice as well as a series of integrative behavioral assays, we functionally determine that the GG olfactory subsystem controls innate and acquired food preferences when mice smell an impending danger. The GG is indeed required to decipher the threatening quality of an unfamiliar food. We also find that its activation increases the systemic corticosterone level which in turn enhances safe food preferences acquired during social transmission. Moreover, we finally demonstrate that the acquisition of a food preference itself could be circumstantially reset by the activation of the GG circuitry, providing a decisionmaking advantage for mice.

Results
Selective deletion of the GG circuitry. The conflicting environmental olfactory messages carried by odorants emitted by familiar food and by olfactory danger cues need to be continuously evaluated for risk taking and final food decision-making. To address the functional relevance of the GG in this last process in mice, we first surgically disconnected it by nerve axotomy 9 (axo; Fig. 1a). Thanks to the expression of the olfactory marker protein (OMP; Fig. 1b, c), a neuronal marker for both the GG and the MOE, we then verified that the axotomy induced specific loss of the GG in axotomized (Axo) but not in control sham-operated mice (Ctrl) (Fig. 1b) without affecting the MOE (Fig. 1c). We next demonstrated that the GC-D neurons were still present in the MOE after GG axotomy (Fig. 1c) thanks to the expression of the enzyme phosphodiesterase 2 A (PDE2A; Fig. 1b and c) which is shared by GG and GC-D circuities 19 . We then used transgenic guanylyl cyclase-G (GCG, a marker of GG circuitry)-Cre-green fluorescent protein (GFP) mice 19 to selectively trace the GG connections into their olfactory bulb target (OB; Fig. 1a), the necklace glomerular complex (NG; Fig. 1a). Thereby, we distinctly differentiated, in the NG, the parallel connections between GG and GC-D circuitries and we found that the GC-D circuitry and, in particular its associated necklace glomeruli (NG), were still intact in GCG-Cre-GFP Axo mice ( Fig. 1d and Supplementary Fig. 1; Ctrl, N mouse = 3; Axo, N mouse = 5). Moreover, this GC-D circuitry was still active in the presence of CS 2 , independently from the GG axotomy procedure, as verified by immediateearly gene c-Fos expression 15 (Fig. 1e, f; Ctrl, N glomeruli = 6; Axo, N glomeruli = 7; two-tailed t-test: p = 0.544, ns). We have then systematically used the selective deletion of this olfactory subsystem to functionally dissect the relevance of the GG circuitry in odordriven food selection when mice were exposed to chemical danger cues.
GG encodes threatening scents and dominates over MOEtransmitted odor signals. Mice naturally disregard unfamiliar food based on the unknown odorants it releases, moreover familiar food soiled with danger cues is likewise avoided 2 . As a first physiological assay, we, therefore, tested odorants commonly used to odorize food such as non-synthetic spices 13 and we found that these odorants were not directly detected by GG neurons. Indeed, we revealed with calcium imaging experiments performed on acute Fura-2 acetoxymethyl ester (Fura; Fig. 2a) loaded GG slice preparations from transgenic OMP-GFP mice 9 that, in a total of 54 living GFP tagged GG neurons (N mouse = 4; n slice = 6), Cinnamon, Cocoa, Anise, Oregano, Thyme, Basil, Nutmeg, Ginger as well as the standard mice food and CS 2 did not generate any GG neuronal activity (Fig. 2b). On the other hand, the predator-derived cues 2-propylthietane (2PT) from the stoat anal glands, the mouse alarm pheromone 2-sec-butyl-4,5-dihydrothiazole (SBT) as well as mountain lion (Mt.Lion) urine 15,20 initiated reversible calcium responses in respectively 83, 57 and 100% of the GG neurons. Thus, this first approach not only suggests that the unfamiliarity of a food encoded by its emitted odorants could not be directly deciphered by the GG but also confirmed the ability of the GG to identify danger cues potentially emitted by soiled food.
We next verified, in an integrative context of food choice, the implication of the GG in decoding food unfamiliarity based on the odorants emitted. Ctrl and Axo mice were challenged to select, between familiar and unfamiliar food in a two choices assay ( Fig. 2c and Supplementary Fig. 2a). To that purpose, two familiar powdered foods were proposed to mice, odorized either with a series of never-encountered before odorants (odor #1; as unfamiliar food) or with the odorless water (odor #2; as familiar food). We placed all the tested odorants around the powdered foods to exclude any toxicity potentially displayed by synthetic cues. Subsequently, the preference for an odorized food was calculated as the ratio between the consumption of food odor #1 versus the total food consumed (food odor #1 + odor #2) in which the 0.5 value, corresponding to the non-preference threshold was subtracted. Values were thus expressed between 0.5 and −0.5 where positive scores corresponded to a preference for the food odor #1; negative scores for the food odor #2 and zero corresponded to no preference displayed. Hence we showed that rodents indeed prefer familiar food 2 as non-synthetic spices such as the unfamiliar Cinnamon or Cocoa 13 as odor #1 were not preferred by mice. Moreover, we observed that this innate choice was also performed by Axo mice (Fig. 2d), confirming that this avoidance behavior was indeed not directly dependent of the GG detection (Fig. 2a, b) but also pointed out to the conserved MOE functionality in the GG Axo mouse model. Then we found that the GC-D-related ligand CS 2 , without any social context 10 , did not influence diet selection in both Ctrl and Axo mice (Fig. 2d). We next tested butyric acid (BA) a known aversive odorant that smells rancid and has no alerting relevance 15 and found that it indeed generated food avoidance both in Ctrl and Axo mice confirming its previously reported GG-independent detection 15 ( Fig. 2d). Food odorized with predator scents were also aversive ( Fig. 2d) as observed with the pyrazine analogues (Pyrazines), found for example in Mt.Lion urine 20 , 2-PT, 2,4,5-trimethylthiazoline (TMT) from the red fox feces as well as the mouse alarm pheromone SBT. Nevertheless, in Axo mice, this dislike was reduced indicating and confirming (Fig. 2b) that the GG was indeed implicated in the perception of these chemical danger cues 15 . Finally, we used Mt.Lion urine as a natural source of predator scents 20 and found that its aversive effect on food choice was exclusively dependent on a functional GG (Fig. 2d). Thus, confirming our GG calcium imaging investigations (Fig. 2a, b) that show that mice do not recognize odor signals emitted by unfamiliar food via GG detection as demonstrated by exposing them to familiar versus unfamiliar foods (Fig. 2d).
We next highlighted that this olfactory subsystem was fundamental for mice to decipher the threatening quality of an unfamiliar food (Fig. 2e). Indeed, on sets of naive mice, when we used the pungent and unfamiliar BA as odor #2 (Fig. 2e) in the previous two choices assay (Fig. 2c), Ctrl and Axo mice now preferred Cinnamon, Cocoa and CS 2 as sources of unfamiliar odorized food (Fig. 2e) confirming the previously observed aversity of the BA (Fig. 2c). Interestingly, in the presence of danger cues such as the Pyrazines, 2PT, TMT, SBT and Mt.Lion urine, Ctrl mice now systematically preferred the aversive and unfamiliar BA (Fig. 2e). TMT and SBT were particularly efficient as they were still able to generate this innate reaction in serial dilutions ( Supplementary Fig. 2b, c). Remarkably, this observed preference for BA disappeared in the absence of a functional GG, without affecting the total food consumption ( Supplementary Fig. 2d, e). Thus, mice decode the threatening quality of unfamiliar food by GG detection.
Taken together, our results show that, the GG acts as an immediate sensor, which deciphers the threatening quality of odorants emitted by food. Mice will take their final consumption decision about the safety of the resource thanks to its smell and to its GG perception.
Threatening scents activate GG-dependent corticosterone responses. In the wild, food resources are often limited and could be located in a dangerous setting such as impending predation. Nevertheless, the motivational state is modified under hunger context 3 . To evaluate trade-offs displayed by fasting mice when confronted to actively searching a food resource in a danger context, we next challenged Ctrl and Axo mice with unreachable food resources moistened with SBT and with Mt.Lion urine, as sources of respectively intra-and inter-species conditioning stimuli (+C.S) mimicking environmental evidence of olfactory threats (Fig. 3a). Remarkably, we observed the absence of a fearlike response such as freezing (Fig. 3b) and the display of a risk assessment behavior (Fig. 3c) in both Ctrl and Axo mice 15 , indicating that, in a context of food scarceness, the opportunity to eat indeed overrules fear (Fig. 3a-c). These observed behavioral adaptations also highlight that odorants emitted by food were sufficient to initiate this motivational-related behavioral process independently from GG detection (Fig. 3c). We then followed the systemic integration of these stressful situations with the expression of the immediate-early gene c-Fos 21 . We focused on the amygdalopiriform transition area (APir; Fig. 3d) which is the brain region implicated in the increase of stress-related hormone level in the blood when mice smell volatile predator scents 15,22 . We found that this brain region was significantly activated by both SBT and the predator urine ( Fig. 3e, f). We next confirmed this result by measuring an increase of the systemic corticosterone level (Fig. 3g), implying that intra-and inter-species danger cues were both processed in this specific APir nucleus. In a dilution series of intraperitoneal corticosterone injections (Cort. i.p.; Fig. 3h), we were further able to mimic the observed stress-related hormonal elevation with an amount of Cort. i.p. of 5.0 mg kg −1 , therefore bypassing APir activation in absence of conditioning stimuli (−C.S; Fig. 3h and Supplementary Fig. 3). Moreover, we found that, in Axo mice, the activation of the APir region, as well as the elevation of the systemic corticosterone, were significantly impaired in this threatening context ( Fig. 3e-g), demonstrating that, although the GG was not involved in odor-driven foraging, it was essential for the hormonal and physiological adaptation to danger sensing.
Elevation of corticosterone optimizes learned-food selection. In the search of food resources, knowledge of familiar food obtained by STFP support the food decision-making by a recall memory process that requires the activation of a specific part of the hippocampus, the ventral subiculum 13,23 (VS; Fig. 3d). In this study, we found that the presence of danger cues enhanced this learned . c Two choices assay, illustrated here with an infrared snapshot and a procedure time-table. Odorants were placed around food resources (odor #1, yellow; odor #2, red). d, e Quantification of food Preference ratio for Ctrl (black) and Axo (gray) mice, calculated as the ratio of food consumption [(odor #1)/(odor #1 + odor #2)] -0.5. Positive scores display a preference for the food odor #1; negative scores for the food odor #2. Tested odor #1 (10%, or pure for Mt.Lion) are indicated and are opposed to odor #2 (Water, in (d) and BA (10 %), in (e)). Data are represented as mean ± SEM with aligned dot plots for individual mice values. Calculation of statistical significances of the preference ratio is performed with Z tests, #p < 0.05 (yellow or red # for a preference respectively for odor #1 or odor #2); non-significant if not mentioned. For comparisons between Ctrl vs. Axo mice, two-tailed Student's t-tests or Wilcoxon w-tests are used, *p < 0.05; **p < 0.01. 6 to 16 animals are used per condition. food-odorant preference when we performed a two choices assay under threatening conditions. For that, we first trained mice to develop a food-odorant preference. In brief, a demonstrator mouse ate an odorized demonstrating food (de.food; standard powdered food odorized for example with Cinnamon as a spice #1; Phase 1; Fig. 4a). Then it returned with its observer littermates to allow the STFP learning process to happen (Phase 2; Fig. 4a). Finally, after 24 h of fasting, the observer mice were individually confronted to a two choices assay between two odorized foods, the demonstrating food and a novel food (new.food; standard powdered food odorized for example with Cocoa as spice #2; Phase 3; Fig. 4b), both surrounded with the same conditioning Fig. 4 Chemical danger cues enhance food preferences acquired by STFP. a Acquisition of a food preference performed by STFP assay is illustrated by infrared snapshots and a procedure time-table. In phase 1, a demonstrating food (de.food, yellow) is presented to a demonstrator mouse (de). In phase 2, observer mice (ob) are in contact with the de for social interactions and acquisition of a preference for the de.food. b In phase 3, each ob mouse is individually tested in a two choices assay with two sources of food (de.food, yellow; new.food, red) surrounded with the same conditioning stimulus (+C.S, purple). c Quantification of the food preference ratio in Ctrl and Axo mice without or with STFP procedure and under the indicated environmental conditioning (STFP− or + | + C.S) or after intraperitoneal injection of 5.0 mg kg −1 corticosterone (Cort. i.p.). Statistical significances of preference ratio are performed with Z tests, #p < 0.05 (yellow, for a preference for de.food); non-significant if not mentioned. 12 to 36 animals were used per condition. The following pairs of spices were used: cinnamon 1% vs. cocoa 2%; anise 1% vs. oregano 2.4%; thyme 2% vs. basil 1.4%. d Representative c-Fos staining (dark spots) in Ctrl and Axo mice under the indicated conditioning in VS (blue dashed line). Scale bars, 50 μm. The following pair of spices were used: nutmeg 1% vs. ginger 1%. e Quantification of the density of c-Fos+ cells in VS from (d). A minimum of 4 animals were used per condition. c, e Values obtained from Ctrl (black) and Axo (gray) mice are represented as mean ± SEM with aligned dot plots. For comparisons between conditions and between Ctrl vs. Axo mice, two-tailed Student's t-tests or Wilcoxon w-tests are used, *p < 0.05; ***p < 0.001. stimulus (+C.S; Fig. 4b). To avoid any individual innate preference, spices were used in a counterbalanced mode 13 (Supplementary Fig. 4a to c). We observed in both Ctrl and Axo mice that STFP was, indeed, required to develop a food preference (STFP+ | + Water; Fig. 4c) that was, remarkably, not impacted by a chemical confusion generated by the pungent BA (STFP+ | + BA; Fig. 4c). These first effects not only confirmed the conserved functionality of the GC-D circuitry in Axo mice (Fig. 1c, e, f) but also pointed out that, under aversive distraction, mice still olfactively decipher food preferences. Surprisingly, we obtained, with the danger cue SBT (STFP+ | + SBT) and with Mt.Lion urine (STFP+ | + Mt.Lion) an enhancement of the food-odorant preference (Fig. 4c). Indeed, the observed mean food preferences were amplified by around 60 % compared to STFP under standard environmental (STFP+ | + Water) or under pungent conditions (STFP+ | + BA). This enhancement was observed in Ctrl but not in Axo mice. It was found to be both independent from food consumption (Supplementary Fig. 5a) and from an improvement of the recall memory process as no significant increase in c-Fos activity was observed in the VS brain area under these conditions (Fig. 4d, e). Interestingly, we found that injection of Cort. i.p. 5.0 mg kg −1 under neutral context (STFP+ | Cort. i.p.) was sufficient to mimic this behavioral improvement both in Ctrl and Axo mice (Fig. 4c) without affecting the memoryretrieval activity of the VS (Supplementary Fig. 5b, c). We thus demonstrated here that, when danger cues are sensed by the GG, the increase in the systemic corticosterone level that occurs (Fig. 3e) leads to an enhancement of the food preference previously acquired by STFP.
GG circuitry activation selectively erases safety food memory. Conspecific interactions are useful for mice to get food familiarity. We found here that mice can also benefit from threatening chemical information present in their environment to reset this previously acquired food-odorant familiarity when associated with a danger context. We indeed tested the impact of olfactory threats on the acquisition of a new food-odorant preference as we challenged mice 1 h after an STFP procedure (Phase 1; Fig. 5a) with a surrogate display that contained the unreachable demonstrating food moistened with an associated conditioning stimulus (Phase 2; +C.S; Fig. 5a), a procedure that allows food investigation (Fig. 3a-c). After 24 h of fasting, the observer mice were then tested in a two choices assay (Phase 3; Fig. 5b). Unexpectedly, we found that in association with the danger cues SBT (STFP+/+ SBT) or with Mt.Lion urine (STFP+/+ Mt.Lion), Ctrl mice did not display food-odorant preferences while they were still observed in Axo mice (Fig. 5c). As a control, we verified that GGrelated ligands directly associated with a demonstrating food could not act as conditioning stimuli promoting a food preference or avoidance (Supplementary Fig. 6a-c), confirming our previous observations that avoidance towards an unfamiliar odorized food is innately coded (Fig. 2d). Besides, we also observed that this apparent and selective amnesia was independent from food consumption ( Supplementary Fig. 7a) or corticosterone elevation. Indeed, mice injected with Cort. i.p. 5.0 mg kg −1 instead of the associated conditioning stimuli (STFP+/Cort. i.p.) still displayed a food-odorant preference (Fig. 5c). Moreover, we observed a striking resetting of the VS brain area (c-Fos stainings; Fig. 5d, e), that suggests an impairment in the recall memory for a food preference or for its consolidation. Remarkably, as this process was independent from the systemic corticosterone level (Supplementary Fig. 7b, c) and not observed in Axo mice, this cerebral adaptation was thus directly related with the activation of the GG neuronal circuitry itself (Fig. 5d, e).

Discussion
Acute stress can modulate both the performance of actionoutcomes and the learning processes in animals including rodents and humans 24,25 . Impending danger such as the risk of predation increases stress-related hormonal levels in the blood 5,15 and can affect memorization 26 . When faced with imminent danger, alerting senses like the olfactory senses allow the detection of chemical cues signaling for the presence of predators 27 . The olfactory senses are also essential to find food and therefore trade-offs and sorting out of the environmental chemical information has to take place followed by the appropriate behavioral responses. In this study, we investigated the biological relevance of the detection of chemical danger cues by the GG on odor-driven food selection because of its fundamental need for animal survival. We demonstrated that the GG circuitry favors innate odor-driven food decision-making in a context of danger (Fig. 2). Indeed, the preys with a functional GG are able to select the food resource, depending on its smell, while avoiding to be eaten by predators. We have also observed that the GG was not directly implicated in the detection of odorants emitted by food resources (Fig. 2) and in the linked foraging behavior (Fig. 3), we can thus consider that our GGdependent results are not exclusively specific to food selection 15,17 and that they may be extended to the global effects of threatening scents on odor-driven preferences.
Our understanding of the interplay between the pathways processing GG-induced fear-like responses and those encoding food-odorant information remains to be completed. GG neurons use multi-signaling pathways, principally linked to bitter taste signaling 28 and GCG cascades 18 to detect a large repertoire of chemical danger cues 28,29 . Interestingly, in GC-D neurons, a similar encoding strategy occurs 14 making it also sensitive to fatty acids, steroid hormones as well as to some GG-related cues such as pyrazines. This shared encoding strategy as well as the partial overlap of recognized cues between the GC-D and the GG subsystems might, at the detection level, contribute to the apparent positive collaboration between these two parallel circuitries.
In a pilot experiment performed in the NG complex of GCG-Cre-GFP mice, we noticed the presence of periglomerular cells expressing the tyrosine hydroxylase 30,31 (TH; Supplementary  Fig. 8a, b). Interestingly, their density was preferentially located around GC-D glomeruli (GC-D: 9.3 ± 0.9 TH + cells, N glomeruli = 6; GG: 5.0 ± 1.2 TH + cells, N glomeruli = 7), supporting the interconnectivity of GC-D glomeruli with other olfactory circuitries such as those of the MOE 32 . On the other hand and compared to other OB regions 33 , TH + cell density was moderate around GG glomeruli, which is consistent with their homogenous innervation 19 and their episodic afferent activity 9 . This might question the potential cross-glomerular regulatory task of these periglomerular cells. Moreover, and as a temporal hiatus, we found that the GG circuitry could impact socially learned food preferences in a delayed manner (Figs. 4 and 5), at least 1 h after the STFP procedure (Fig. 4). Therefore, periglomerular regulation emerging from GG on GC-D glomeruli requires further investigations.
The integration of the GG circuity on higher structures in the CNS have revealed specific subregions implicated in fear and anxiety responses such as the posteroventral division of the medial amygdala (MeA) and the dorsomedial subdivision of the ventromedial hypothalamus 16,34 (VMH). Our findings unravel two additional brain domains implicated in this global integration, the APir (Fig. 3) and the VS (Figs. 4 and 5). Interestingly, the APir was previously reported as a specific area of the olfactory cortex that induces stress hormone responses to volatile predator scents 21 .
Here, we demonstrate that this heterospecific response could be extended to conspecific-signaling integration, as the mouse alarm pheromone SBT likewise initiates APir activities (Fig. 3). In addition to the instinctive fear response, we, therefore, uncovered here a key consequence of the subsequent systemic corticosterone elevation and demonstrated its central outcome for the improvement of odor-driven food behavior (Fig. 4). The processing ability of the CNS structures, such as the one observed here with the APir is usually dependent of their activation. On the other hand, we found that the treatment of the information by the VS emerging from the GG circuitry consisted in its ability to reset its memory-retrieval activity for socially learned food response 13,23 (Fig. 5). This last finding is consistent with previous reports implying threatening scents in the impairment of retrieval memory 35 but also supports the VS as a key CNS area implicated in context-dependent stress Fig. 5 Acquisition of a food preference is impaired by environmental GG activation. a Acquisition of a food preference performed by STFP assay followed by an associated conditioning stimulus (+C.S, purple) is illustrated by infrared snapshots and a procedure time-table. In phase 1, acquisition of a food preference. A demonstrating food (de.food, yellow) is presented to observer mice (ob) with a surrogate display, moistened with CS 2 , mimicking social interactions. In phase 2, de.food is again presented to ob mice but associated with a +C.S. b In phase 3, each ob mouse is individually tested in a two choices assay with two sources of food (de.food, yellow; new.food, red). c Quantification of the food preference ratio in Ctrl and Axo mice without or with STFP procedure followed by the indicated associated stimulus (STFP− or +/+C.S) or after intraperitoneal injection of 5.0 mg/kg corticosterone (Cort. i.p.). Statistical significances of preference ratio are performed with Z tests, # p < 0.05 (yellow, for a preference for de.food); non-significant if not mentioned. 10-18 animals were used per condition. The following pairs of spices were used: cinnamon 1% vs. cocoa 2%; anise 1% vs. oregano 2.4%; thyme 2% vs. basil 1.4%. d Representative c-Fos staining (dark spots) in Ctrl and Axo mice under the indicated conditioning in VS (blue dashed line). Scale bars, 50 μm. The following pair of spices was used: nutmeg 1% vs. ginger 1%. e Quantification of the density of c-Fos+ cells in VS from (d). Four animals were used per condition. c, e Values obtained from Ctrl (black) and Axo (gray) mice are represented as mean ± SEM with aligned dot plots. For comparisons between conditions and between Ctrl vs. Axo mice, two-tailed Student's t-tests or Wilcoxon w-tests are used, *p < 0.05; **p < 0.01; ***p < 0.001. and integration of fear memories [36][37][38] . Moreover, we have observed that the same chemical danger cues can act both during (Fig. 4) and upstream (Fig. 5) of the decision-making event with opposing behavioral outcomes. Threatening scents could thus develop contrasting effects on odor-driven food choices in which the VS should play a context-dependent integrative role. These observations further imply a complex strategy for encoding both the predatorrelated stress as well as the retrieval of innate and learned fear 39 . At the individual level, this innate trait could be considered as a beneficial mechanism of protection. Indeed, when the food choice is presented in a context of threat (Fig. 4), the favored decision for survival should be to select a safe and familiar food. Nonetheless, to be efficient and to decrease the risk of being eaten, a compromise is made and the emerging strategy is, in the end, to erase this food familiarity and to consume the food anyway (Figs. 3 and 5). Thus, for mice, as for humans, hierarchy of prepotency as well as behavioral motivation is innately coded. Therefore, physiological needs, such as feeding, surpass safety needs 40 .
In the wild, the ability to take rapid decisions concerning food choice is critical for the survival of the individual and leads to an increase in the overall fitness of the species. It is therefore conserved throughout the animal kingdom 41,42 . For example, in the nematode Caenorhabditis elegans (C. elegans), stress linked to environmental conditions can impact food preferences when they are coupled with odorants 43 as well as with alarm pheromones and predator scents 44,45 . Interestingly, this worm modality is dependent on a specific olfactory subsystem composed of a pair of amphid wing cells of type C 46 (AWC). GG neurons were reported to share molecular and functional features with these AWC neurons, conferring to these two olfactory subsystems an interspecies orthologous status 29 .
In summary, our present study demonstrates that the GG potentiates, in mice, the decision-making process for the odordriven food choice under threat. Our results highlight how mice integrate environmental stressful stimuli to control both innate and previously acquired odor-driven food preferences. Olfactory threats are detected by the GG that operates as an immediate sensor acting both on innate perceptions, systemic hormonal levels and on the acquired food preferences to circumstantially revise mice decision-making, a mechanism that confers to the animal a beneficial survival advantage 21 .

Methods
Animals. Male and female C57BL/6 (Mus musculus; Janvier Labs), OMP-GFP 47 and GCG-Cre-GFP mice 19 (lines 43 and 52) were used. In the gene-targeted mouse strain OMP-GFP, the GFP is used as a histological reporter of mature OSNs 48 expressed under the control of the OMP promoter 47,49 . In transgenic GCG-Cre-GFP mice the expression of GFP is driven by the GCG gene which allows the selective expression of this marker in GG circuitry. Otherwise mentioned, mice were grouped-housed (3-5 animals/cage) at 21.5°C under a 12 h light/ dark cycle with ad libitum access to food (3242; Kliba Nafag) and water. Mice were euthanized by CO 2 inhalation and/or by cervical dislocation. The experimental procedures were in accordance with the Swiss legislation and approved by the EXPANIM committee of the Lemanique Animal Facility Network and the veterinary authority of the Canton de Vaud (SCAV).
Generation of control (Ctrl) and GG Axo mice. The GG completely degenerated after surgical sectioning of its axonal bundles 50 . Under deep isoflurane anesthesia, GG nerves from mice pups were Axo by using disposable 26 G needles 9 . Similar surgical procedure with only superficial sectioning was used to generate Ctrl mice in which the axotomy was not effective. Axo and Ctrl mice were kept grouped with their littermates until behavioral assays. The efficiency of the GG degeneration was assessed at the end of the behavioral assays using a histological approach. In brief, after fixation in 4% paraformaldehyde (PAF 4%; pH 7.4), mice noses were included in low melting 4-5% agar and serial coronal slices of 80 μm were generated with a vibroslicer (VT1200S, Leica). Slices of GG region were selected under a stereomicroscope (M165 FC, Leica) in accordance with their general morphology 51 . Using a floating immunohistochemical technic 29 , presence of GG neurons was detected thanks to their OMP expression. For that, slices were blocked for 3 h in a PBS solution containing 5% normal rabbit serum (NRS; Jackson ImmunoResearch) and 0.5% triton X-100. The primary antibody (Goat anti-OMP; Wako; 1:1000) was applied to the slices overnight in a PBS solution containing 2.5% NRS and 0.25% Triton X-100. Slices were then washed in 1% NRS and incubated with the secondary antibody (FITC-conjugated, Rabbit anti-Goat; Jackson ImmunoResearch; 1:200) in 1% NRS for 2 h. As control, a similar procedure was systematically performed with the MOE after a decalcification treatment (PBS -EDTA 0.5 M, pH 8.0). For double staining with the enzymatic phosphodiesterase 2A (Rabbit anti-PDE2A; FabGennix; 1:400), serum from donkey (NDS; Jackson ImmunoResearch) and appropriate secondary antibodies (FITC-conjugated, donkey anti-Goat; Jackson ImmunoResearch; 1:200; Cy3-conjugated, donkey anti-Rabbit; Jackson ImmunoResearch; 1:200) were used. For analyses, slices were mounted in antifade mounting medium with DAPI (Vectashield, H-1200; Vector Labs). Observations and acquisitions were made by LED-fluorescence microscopy (EVOS fl, AMG) and confocal microscopy (SP5; Leica) with ×4 to ×40 objectives. Maximum projection reconstructions were made with Imaris (Bitplane IMARIS 6.3). Mice with an expression of OMP-positive GG neurons were considered as the Ctrl mice and those with the absence of OMP-positive cells as Axo mice. No differences were observed between untreated and Ctrl mice 9 .
Behavioral assays. All behavioral assessments were conducted in a dedicated room under the supervision of the affiliated investigator 58 . Adult (8-16 weeks) C57BL/6 mice, with an equivalent sex ratio per conditions, were tested during the nocturnal period of the 12-h light/dark cycle. Tested odorants were only used for a single behavioral trial 59 . Delivery of powdered food was performed with homemade-ballasted cups limiting the food dispersal.
Two choices assay and quantification of the food preference ratio. To verify if the mice display a preference between two sources of odorized food (food odor #1 or food odor #2), we adapted previously designed two choices assays [60][61][62] . In brief, grouped-housed mice were food habituated for 72 consecutive hours with access to 2 g/day/mouse of powdered ordinary food followed by 24 h of food deprivation. After this period of habituation, a food preference test was performed on individual animal. For that, mice could choose, during 1 h, between two cups of odorized powdered food placed in the opposite part of the cage in an accessible semi-closed box (~500 cm 3 ) to limit odorant diffusion. Each source of food was offered in sufficient amount (between 3.0 and 3.5 g). Quantification of the food preference ratio for the food odor #1 was calculated according to the ratio between the food odor #1 consumed versus the total food consumed (food odor #1 + odor #2) minus 0.5, corresponding to the non-preference threshold. Values are then expressed between 0.5 and −0.5 where positive scores correspond to a preference for the food odor #1, negative scores for the food odor #2 and zero correspond to no preference displayed. Values obtained for a consumption ≥0.2 g were processed. To mimic food soiled by different odorants, 500 μl of the investigated substances, diluted in water, were deposited around the food without any direct contacts 60,61 . We used as tested odor #1, distilled water (Water), Cinnamon and Cocoa (10%, 1:5 from liquid stock), CS 2 (10 ppm), pyrazines (10%), 2PT (10%), TMT (10% to 0.1%), SBT (10 to 1%) and pure Mt.Lion urine. The odorless Water or the aversive BA 60,63 (10% or adapted) were used as odor #2.
STFP assay. Our STFP assay was designed according to previous reports 10,13 . For that, a demonstrator mouse (de) was first separated from its group of observer mice (ob) and fed during 1 h a spiced demonstrating food (de.food). A total consumption of ≥0.2 g by the demonstrator mouse was necessary to proceed. Then, this demonstrator mouse was returned with its littermates for 1 h of social interactions that allowed the acquisition of food preference for this de.food 13 . Depending on our investigations, STFP phases could be condensed by using a surrogate demonstrator mouse presented for 1 h to the observed mice 13 . This surrogate demonstrator mouse was composed of a cotton ball on which the unreachable de.food was supplemented with 500 μl of CS 2 (10 ppm) and enclosed in a wire mesh. To verify the acquisition of food preference for the de.food, a two choices assay was then performed after a food diet of 24 h. For that, individual food preference of the observer mice was performed between the de.food (as food odor #1) and an unfamiliar paired spiced food (new.food, as food odor #2). Quantification of the preference ratio for the de.food was then calculated. Demonstrated spices were counterbalanced within pairs in each behavioral trial and cup positions varied randomly. A couple of spices were used only once per mouse and mice could be tested with different sets of spices. Values obtained for a consumption ≥0.2 g were processed. To modulate the STFP performance, the addition of conditioning stimuli (+C.S) were considered during or before the food selection test. Respectively, 500 μl of conditioning stimuli, +Water, +BA (1:500), +SBT (1:500) or +Mt. Lion (pure Mt.Lion urine) were placed around the food during the two choices assay or were coupled for 1 h with the de.food, 1 h after food preference acquisition (STFP assay).
Systemic corticosterone measurement and injection. After powdered food habituation and 24 h of fasting, grouped of mice were challenged for 1 h with 500 μl of conditioning stimuli, +Water, +SBT (1:500) or +Mt.Lion (pure Mt.Lion urine). Each stimulus was delivered on a cotton ball containing unreachable powdered food enclosed in a wire mesh. The quantification of the freezing time and the risk assessment behavior displayed by mice 15 was performed for the first 15 min and respectively expressed as a percentage of time to freeze and as a mean score obtained per minute in a repeated measure design. Directly after euthanasia, blood samples were obtained by cardiocentesi, collected in EDTA-coated microtubes (Micro tube 1.3 ml K3E; Sarstedt), centrifuged at room temperature for 5 min at 10,000 × g and plasmas were kept at −80°C until Elisa analysis 15 (EIA Kit; Enzo Life Sciences). According to the kit manufacturer, corticosterone levels were measured in duplicate and a minimum of four mice were used to evaluate each situation. To mimic the observed increase of stress-related hormonal levels, corticosterone (Corticosterone: HBC complex; Sigma) in 0.9% saline vehicle were intraperitoneally injected (Cort. i.p., 10.0, 5.0, 2.5, 0.5, 0.0 mg kg −1 ) with 100 μl per 10 g of mice bodyweight 30 min before behavioral assays 64 .
Brain regions and c-Fos analysis. Thirty minutes after behavioral assays, mice brains were collected and fixed (PAF 4%). Brains were post-fixated overnight in 1% PAF at 4°C and rinsed in PBS before the inclusion in 4% low melting agar. Coronal slices of 60 μm were generated with a vibroslicer (VT1200S, Leica) on ice-cold PBS and according to their morphology 65 , slices comprised between Bregma −3.28 to −3.80 were selected under a stereomicroscope (M165 FC, Leica) and were used for the ventral subiculum (VS) analysis. Slices corresponding to Bregma −3.28 were used for the amygdalopiriform transition area (APir) analysis. A similar immunohistochemical process than the one used before was performed to examine the brain immediate-early genes c-Fos labeling. In brief, selected brain slices were incubated for 60 h with the primary antibody (Rabbit anti-c-Fos; Santa Cruz Biotechnology; 1:1000) in a normal goat serum (NGS; Interchim) solution and the detection was done with a fluorescent secondary antibody (Cy3-conjugated, Goat anti-Rabbit; Jackson ImmunoResearch; 1:200). LED-fluorescence microscopy (EVOS fl, AMG) with ×4 objectives and confocal microscopy (SP5; Leica) with ×20 objective were used for the acquisitions of respectively the VS and the APir regions under a black/white contrast. Using the open access imageJ software (National Institutes of Health, 1.48 v), a stereological counting approach was performed for each investigated brain (4-6 per condition). The mean density of c-Fos positive cells was measured per slice according to 2-4 surfaces per area. Average of a minimum of two slices was used for the establishment of the global density of c-Fos positive cells per area and brain.

Statistics and reproducibility.
The open-source package R version 3.1.2 was used for statistical analysis and GraphPad Prism 8.2.0 to compute bar graph with their corresponding aligned dot plots. Values are expressed as mean ± standard error of the mean (SEM). Food preferences were assessed by Z-tests according to their calculated mean z values (preference ratio/SEM). Shapiro-Wilk tests and Fisher Ftests were used for evaluation of normality and homoscedasticity. Accordingly, comparisons were performed with the two-tailed Student's t-tests or Wilcoxon (Mann-Whitney) w-tests and no corrections were applied to compare multiple data. Significance levels are indicated as follows for the preference ratio: # p < 0.05, in yellow for a significant preference for odor #1 or de.food and in red for a significant preference for odor #2 or new.food; non-significant if not mentioned. For comparisons between Ctrl and Axo mice and between conditions: *p < 0.05; **p < 0.01; ***p < 0.001.