Abstract
A dynamic environment, such as the one we inhabit, requires organisms to continuously update their knowledge of the setting. While the prefrontal cortex is recognized for its pivotal role in regulating such adaptive behavior, the specific contribution of each prefrontal area remains elusive. In the current work, we investigated the direct involvement of two major prefrontal subregions, the medial prefrontal cortex (mPFC, A32D + A32V) and the orbitofrontal cortex (OFC, VO + LO), in updating pavlovian stimulus–outcome (S–O) associations following contingency degradation in male rats. Specifically, animals had to learn that a particular cue, previously fully predicting the delivery of a specific reward, was no longer a reliable predictor. First, we found that chemogenetic inhibition of mPFC, but not of OFC, neurons altered the rats’ ability to adaptively respond to degraded and non-degraded cues. Next, given the growing evidence pointing at noradrenaline (NA) as a main neuromodulator of adaptive behavior, we decided to investigate the possible involvement of NA projections to the two subregions in this higher-order cognitive process. Employing a pair of novel retrograde vectors, we traced NA projections from the locus ceruleus (LC) to both structures and observed an equivalent yet relatively segregated amount of inputs. Then, we showed that chemogenetic inhibition of NA projections to the mPFC, but not to the OFC, also impaired the rats’ ability to adaptively respond to the degradation procedure. Altogether, our findings provide important evidence of functional parcellation within the prefrontal cortex and point at mPFC NA as key for updating pavlovian S–O associations.
Significance Statement
The ability to update stimulus–outcome (S–O) associations is a key adaptive behavior, essential for surviving and thriving in an ever-changing environment. The prefrontal cortex is well-known for playing a key role in this process. The discrete contribution of each prefrontal subregion and of different neurotransmitters, however, remains unclear. In the current study, we show that inhibiting medial prefrontal (mPFC), but not orbitofrontal cortex (OFC), neurons impairs rats’ ability to update S–O associations following contingency degradation. Moreover, we demonstrate that discrete noradrenergic projections to the two subregions exist and that inhibiting the ones projecting to the mPFC once again impairs the animals’ behavior, thereby implying a substantial contribution of noradrenaline in orchestrating this higher-order cognitive process.
Introduction
Adaptation to an ever-changing environment requires the extraction of the predictive relationships between events. For instance, in urban settings abundant with visual stimuli, it is common for a specific visual cue (e.g., a fast-food sign) to reliably forecast the presence of a particular meal. Consequently, the probability of heading toward the meal location upon spotting the visual cue tends to rise. Nevertheless, the predictive influence of the same cue could shift if the meal becomes accessible without the cue.
Previous research has convincingly demonstrated that the prefrontal cortex plays a pivotal role in behavioral adaptations, specifically in the encoding and updating of instrumental action–outcome (A–O) and of pavlovian stimulus–outcome (S–O) associations, with a high parcellation of function existing among prefrontal subregions. For instance, the medial prefrontal cortex (mPFC) has been historically linked to the acquisition (Killcross and Coutureau, 2003; Tran-Tu-Yen et al., 2009) and updating (Boitard et al., 2016) of instrumental associations and more recently has been suggested to participate in the updating of pavlovian associations following outcome devaluation (Niedringhaus and West, 2022), while the orbitofrontal cortex (OFC) has been implicated, directly or indirectly, in the updating of previously established instrumental (Parkes et al., 2018; Fresno et al., 2019; Cerpa et al., 2023) and pavlovian associations (Ostlund and Balleine, 2007; Alcaraz et al., 2015; Morceau et al., 2022).
The neuromodulation of adaptive behaviors has long been associated with dopamine (DA), from the notion that associative learning is driven by DA transients that correlate with prediction errors (Schultz, 2016; Sharpe et al., 2017; Langdon et al., 2018), to the fact that optogenetic stimulation of DA neurons paired with a cue evokes pavlovian conditioned responses (Saunders et al., 2018). However, recent theories (Cerpa et al., 2021) highlight the importance of including noradrenaline (NA) alongside DA for better understanding adaptive behavior. Indeed, NA, a key regulator of attention and arousal in the central nervous system, has been proposed as the key neurotransmitter tracking uncertainty and driving behavioral flexibility (Bouret and Sara, 2004; Tait et al., 2007; McGaughy et al., 2008; Tervo et al., 2014; Uematsu et al., 2017; Jahn et al., 2018; Cope et al., 2019). Notably, the mPFC and the OFC are believed to receive discrete and direct far-reaching NA projections from the brainstem locus ceruleus (LC), the primary source of NA to the mammalian cortex (Agster et al., 2013; Chandler et al., 2013, 2014). This notion has been recently backed up by the fact that chemogenetic inhibition of LC NA projections to the OFC (VO + LO), but not to the mPFC (A32D + A32V), impaired the ability to update A–O associations following reversal in rats (Cerpa et al., 2023). To our knowledge, this specific anatomical dissociation of effects (mPFC vs OFC) and the possible implication of NA in the updating of S–O contingencies have never been investigated.
Therefore, in the current work, we looked into the discrete involvement of these two major prefrontal subregions in pavlovian contingency degradation. Specifically, male rats were challenged to learn that a particular cue, previously the sole predictor of a specific reward, was no longer a better predictor than the absence of that same cue. First, we found that chemogenetic inhibition of mPFC, but not OFC, neurons during the degradation phase altered the rats’ ability to adaptively learn and respond to degraded and non-degraded cues. Next, employing a pair of novel retrograde vectors, we traced LC NA projections to both structures and observed an equivalent yet relatively segregated amount of inputs. At last, we showed that chemogenetic inhibition of NA projections to the mPFC, but not to the OFC, impaired the animals’ ability to retrieve the degraded S–O contingencies.
The present work contributes to integrate and revisit our knowledge of how the prefrontal cortex controls flexible, goal-directed behaviors. Specifically, we question the implication of the OFC in updating S–O contingencies and point instead at the mPFC and its NA innervation as crucial for executing such behavioral adaptation.
Materials and Methods
Subjects and housing conditions
A total of 95 male Long–Evans rats, aged 2–3 months, were acquired from the Centre d’Elevage Janvier (France). Animals were housed in pairs with ad libitum access to water and standard lab chow, then put on food restriction 2 d before the start of behavioral experiments, and maintained at ∼90% of their ad libitum feeding weight. Rats were handled daily for 3 d before the beginning of the experiments. The facility was maintained at 21 ± 1°C on a 12 h light/dark cycle (lights on at 8:00 A.M.). Environmental enrichment was provided by tinted polycarbonate tubing elements, in accordance with current French (Council directive 2013-118, February 1, 2013) and European (directive 2010-63, September 22, 2010, European Community) laws and policies regarding animal experimentation. The experiments received approval from the ethics and animal experimentation committee of the French Ministry of Higher Education and Innovation (reference number of the project: APAFIS#27928-2020110918011853 v2).
Viral vectors
In Experiment 1 and Experiment 2, an adeno-associated viral vector carrying the inhibitory hM4Di designer receptor exclusively activated by designer drugs (DREADDs) was acquired from the Viral Vector Production Unit at Universitat Autonoma de Barcelona, Spain. This AAV8-CaMKII-hM4D(Gi)-mCherry vector was derived from a plasmid obtained from Addgene (pAAV-CaMKIIa-hM4D(Gi)-mCherry; Addgene plasmid #50477) and used at a titer of 3.5 × 1012 GC/ml, as validated in a previous study (Parkes et al., 2018). In Experiment 3, we employed two retrograde canine adenovirus type 2 (CAV-2) vectors harboring a PRS (NA-specific cis-regulatory element identified in the human dopamine β-hydroxylase) promoter that enables the expression of fluorescent proteins in noradrenergic neurons projecting to the target regions of interest (Hwang et al., 2001). Both CAV-2-PRS-FsRed and CAV-2-PRS-eGFP were supplied by Marina Lavigne and Eric J. Kremer at the Institut Genetique Moleculaire de Montpellier, CNRS UMR 5535, France. They were used at a concentration of 3.6 × 1012 VP/ml and 3.45 × 1012 VP/ml, respectively. In Experiment 4 and Experiment 5, we employed another CAV-2 vector equipped with a PRS promoter, this time controlling the expression of an hM4Di receptor and therefore enabling selective chemogenetic targeting of noradrenergic projections. This CAV-2-PRS-HA-hM4Di-hSyn-mCherry vector was also generously provided by Eric J. Kremer’s team and used at a titer of 3.5 × 1012 VP/ml, as validated in a previous study (Cerpa et al., 2023).
Stereotaxic surgery
For all experiments, rats were anesthetized with 5% inhalant isoflurane gas with oxygen and placed in a stereotaxic frame with atraumatic ear bars (Kopf Instruments) in a flat skull position. Anesthesia was maintained with 1.5% isoflurane and complemented with a subcutaneous injection of ropivacaïne (a bolus of 0.15 ml at 2 mg/ml) at the incision site and then disinfected using Betadine. All viral vectors were injected using a microinjector (UMP3 UltraMicroPump II with Micro4 Controller, World Precision Instruments) connected to a 10 µl Hamilton syringe. For Experiments 1, 2, 4, and 5, each animal received two injections per hemisphere of either AAV8-CaMKII-hM4D(Gi)-mCherry or CAV-2-PRS-HA-hM4Di-hSyn-mCherry viruses in the mPFC (A32D + A32V) or in the OFC (VO + LO). For Experiment 3, animals received counterbalanced unilateral injections of CAV-2-PRS-eGFP and CAV-2-PRS-FsRed in the mPFC and the OFC. Each injection consisted of 1 µl infused at a rate of 200 nl/min. After each injection, the syringe remained in place for an additional 5 min before being removed. The coordinates used for the mPFC (A32D + A32V) were the following: AP +3.2, ML ± 0.6, DV −3.4 and AP +3.0, ML ± 0.7, DV −5.4 for AAV8 injections; AP +3.8, ML ± 0.6, DV −3.6 and AP +3.2, ML ± 0.6, DV −3.6 for CAV-2 injections. The coordinates used for the OFC (VO + LO) were the following: AP +3.7, ML ± 2.0, DV −5.0 and AP +3.2, ML ± 2.8, DV −5.2 for both AAV8 and CAV-2 injections. During surgery, a heating pad was placed under each rat to maintain its body temperature. Following surgery, rats were subcutaneously administered with a non-steroidal anti-inflammatory drug (meloxicam, 2 mg/ml/kg) and individually housed in a warmed cage with facilitated access to food and water for 2 h postsurgery. Rats underwent a recovery period of 3 weeks to allow proper transgene expression. Injection sites were confirmed histologically after the completion of behavioral experiments.
Chemogenetics
The DREADDs agonist deschloroclozapine (DCZ, MedChemExpress) was dissolved in dimethyl sulfoxide (DMSO) at a 50 mg/ml concentration and stored at −80°C (stock solution). The stock solution was then diluted in sterile saline to a final concentration of 0.1 mg/kg and administered intraperitoneally (10 ml/kg, i.p.) 40–45 min before behavioral experiments, as validated in a previous study (Cerpa et al., 2023). The vehicle solution (Veh) was made using 0.2% DMSO in sterile saline. Injectable solutions were freshly prepared on the day of use and manipulated under low-light conditions.
Behavioral apparatus
Training and testing procedures took place in eight identical operant chambers (40 cm width × 30 cm depth × 35 cm height, Imetronic) individually enclosed in sound- and light-resistant wooden chambers (74 × 46 × 50 cm). Each chamber was equipped with two pellet dispensers that delivered grain (Rodent Grain-Based Diet, 45 mg, Bio-Serv) or sugar (LabTab Sucrose Tablet, 45 mg, TestDiet) pellets into a food port (magazine) when activated. Two retracted levers were located on each side of the magazine. Each chamber had a ventilation fan producing a background noise of 55 dB and was illuminated by four LEDs in the ceiling. Experimental events were controlled and recorded by a computer located in the room and equipped with the POLY software (Imetronic). Rats were habituated to the magazine through three daily sessions of training before the beginning of the experiments. During these sessions, they received 40 sugar and 40 grain pellets in the magazine, pseudorandomly interspersed, on average every 60 s.
Pavlovian conditioning
Rats first underwent 8 × 40 min daily sessions of pavlovian conditioning, during which two auditory CS (tone and clicker) were paired with two distinct food rewards (sucrose and grain). Animals were divided into two groups, one with tone–sucrose and clicker–pellet pairings, and the other with the opposite combinations. Each CS was presented for 20 s, during which the corresponding food reward was delivered twice in a pseudorandom manner. Auditory cues were presented each day in two blocks of 15 CS with alternate starting order. Rats therefore received a total of 60 rewards per daily session. The time between each CS was pseudorandomized and ranged from 30 to 90 s, with an average of 60 s.
Pavlovian contingency degradation
Following pavlovian conditioning, rats experienced six daily sessions of contingency degradation during which the predictability of one of the cues was degraded with non-contingent delivery of the reward during the interval period of the CS (60 s on average). Specifically, the 30 reward deliveries occurred at the same frequency both during and outside of the CS presentation. The degraded CS was therefore no longer a reliable predictive cue of the reward. On the other hand, the non-degraded CS was kept at the same schedule of the pavlovian conditioning, with two rewards being delivered during each 20 s CS, with pseudorandomized 60 s average intervals (±30 s). During the degradation sessions, rats were injected systemically (10 ml/kg, i.p.) with either DCZ (0.1 mg/kg) or Veh solutions 40–45 min prior being placed in the operant chambers. All CS–O associations and schedules (Degraded vs Non-degraded) were counterbalanced between experimental groups.
Test in extinction
Following the last session of contingency degradation, rats underwent two tests in extinction, according to a within-subject experimental design. Specifically, each experimental group was divided into two subgroups, one receiving DCZ during the first test, and the other receiving Veh. Conditions were reversed for the second test. During the tests, both CSs were presented four times each for 20 s with 60 s intervals on an alternated schedule with no reward being delivered. One resting day was given in between tests.
Behavioral measures
For both the pavlovian conditioning and extinction tests, the measurement of the conditioned response elicited by a conditioned stimulus (CS) was determined by quantifying magazine visits during the 20 s of the CS, while subtracting the visits during the PreCS period, which refers to the 20 s preceding said CS presentation. This measure is referred to as CS-PreCS magazine visits. CS-PreCS during the first Non-degraded and Degraded CS was used as the critical measure for the pavlovian extinction tests quantifying the cue-evoked reward-seeking response for each CS presentation. During the degradation phase, we assessed the conditioned response by dividing the magazine visits during the CS by the mean visits for that CS in the last session of conditioning (% baseline) to avoid potential contamination of the PreCS period by non-contingent reward delivery.
Histology
At the end of behavioral experiments, rats were killed with an overdose of sodium pentobarbital (Exagon Euthasol) and perfused transcardially with 60 ml of saline followed by 260 ml of 4% paraformaldehyde (PFA) in 0.1 M phosphate buffer (PB). Brains were removed and postfixed in the same PFA 4% solution overnight and then transferred to a 0.1 M PB solution. Subsequently, 40 µm coronal sections were cut using a VT1200S Vibratome (Leica Microsystems). Every fourth section was collected to form a series. Immunofluorescence staining was performed for mCherry (AAV8-CaMKII-hM4D(Gi)-mCherry) and FsRed (CAV-2-PRS-FsRed) proteins: free-floating sections were first rinsed in 0.1 M phosphate-buffered saline (PBS; 4 × 5 min) and then transferred to PBS containing 0.3% Triton X-100 (PBST) for further rinsing (3 × 5 min), before being incubated in blocking solution (PBST containing 3% goat serum) for 1 h at room temperature; sections were then incubated for 24 h at room temperature with a primary antibody (rabbit polyclonal anti-RFP, 1/1,000) diluted in blocking solution; after washes (4 × 5 min) in PBST, sections were incubated for 2 h at room temperature with the secondary antibody (goat polyclonal anti-rabbit TRITC-conjugated, 1/200) diluted in PBST; following rinses in PBS (4 × 5 min), sections were collected on gelatin-coated slides using 0.05 M PB, before being mounted and coverslipped using Fluoroshield with DAPI mounting medium. DAB staining was performed for mCherry to confirm CAV-2-PRS-HA-hM4Di-hSyn-mCherry injection sites: free-floating sections were rinsed in PBST (4 × 5 min) before being incubated in an H2O2 (0.5%) in PBST solution for 30 min at room temperature; sections were rinsed again in PBST (4 × 5 min) before being incubated in the blocking solution; following blocking, they were incubated overnight at room temperature with the primary antibody (rabbit polyclonal anti-RFP, 1/2,500) in blocking solution; after washing in PBST (4 × 5 min), sections were placed with the secondary antibody (biotinylated goat anti-rabbit, 1/1,000) diluted in PBST for 2 h at room temperature; following washes (4 × 5 min) in PBST, they were then incubated with the avidin–biotin–peroxydase complex (1/500 in PBST) for 90 min in a dark environment at room temperature. For the final DAB staining, H202 was added to the solution (10 mg tablet dissolved in 50 ml of 0.1 M Tris buffer) just before incubation of 3–4 min; sections were then rinsed with 0.05 M Tris buffer (2 × 5 min) and 0.05 M PB (2 × 5 min), before mounting in 0.05 M PB and coverslipping using a Fluoroshield medium.
Tracing analysis
To quantify neuronal density within the LC, each animal received a unique identification number to allow blind counting. eGFP+ and FsRed+ neurons were captured through a Leica VM5500B Fluorescence Motorized Microscope at 10× magnification. Images were processed using MicroManager. ImageJ was used to quantify cell bodies in each section. The Paxinos rat atlas served as a reference to identify anteroposterior levels (“The Rat Brain in Stereotaxic Coordinates: Hard Cover Edition - George Paxinos, Charles Watson - Google Books,” n.d.).
Statistical analysis
Each rat was assigned a unique identification number that was used to conduct blind testing and data analysis. A three-way analysis of variance (ANOVA) with Period (CS vs PreCS), Stimulus (Non-degraded vs Degraded), and Session as within-subject factors was used to analyze pavlovian conditioning (Figs. 1B, 2B, 4B, 5B). A three-way ANOVA with Treatment at degradation (Veh vs DCZ, to be administered) as a between-subjects factor and Stimulus (Non-degraded vs Degraded) and Session as within-subject factors was also used to analyze pavlovian conditioning in Figure 4B. A three-way ANOVA with Treatment at degradation (Veh vs DCZ) as a between-subjects factor and Stimulus (Non-degraded vs Degraded) and Session as within-subject factors was used to analyze pavlovian contingency degradation (Figs. 1C, 2C, 4C, 5C). A two-way ANOVA with Treatment (Veh vs DCZ) as a between-subjects factor and Session as a within-subject factor was used to compare the responding to the degraded cue in Figure 1C. A three-way ANOVA with Treatment at degradation (Veh vs DCZ) as a between-subjects factor and Treatment at test (Veh vs DCZ) and Stimulus (Non-degraded vs Degraded) as within-subject factors was used to analyze extinction tests (Figs. 1D, 2D, 4C, 5D). A two-way ANOVA with Treatment (Veh vs DCZ) as a between-subjects factor and Stimulus (Non-degraded vs Degraded) as a within-subject factor was also used to analyze extinction tests in Figures 1D and 4D. The accepted value for significance was p < 0.05. Following significant interaction effects, Bonferroni’s post hoc tests corrected for multiple comparisons were performed to clarify statistical interactions. Statistical analyses were performed using GraphPad Prism. Data graphs were created using GraphPad Prism and Adobe Illustrator.
Results
Experiment 1 and Experiment 2
In a first set of experiments, we sought to explore the potential role of the mPFC (A32D + A32V) and the OFC (VO + LO) in updating stimulus–outcome (S–O) associations. To this end, we implemented a pavlovian contingency degradation task where animals had to learn that a cue, which had previously solely predicted a specific outcome, was no longer a reliable predictor. The task design is shown in Table 1. To investigate the involvement of our regions of interest, we used a chemogenetic strategy to selectively inhibit CaMKII+ neurons during the degradation phase, ensuring that the conditioning process remained unaltered. Rats were bilaterally injected with a AAV8-CaMKII-hM4D(Gi)-mCherry virus targeting either the mPFC (Experiment 1) or the OFC (Experiment 2). Then, following a 3-week recovery period, animals underwent eight daily sessions of pavlovian conditioning during which they received two specific and counterbalanced S–O associations (15 presentations of each CS/session). Following conditioning, rats were split into two groups with similar performances. Then, they underwent six pavlovian degradation sessions. Forty-five minutes before each degradation session, half of the animals received an injection of Veh, half an injection of DCZ. During each session, the predictability of one of the two cues was degraded by the non-contingent delivery of the corresponding reward during the intertrial interval period with the frequency of delivery matching the one of the stimulus presentation, accompanied by a reduction in the likelihood of delivery at the cue itself. After completing the degradation phase, rats underwent two tests in extinction during which they received either Veh or DCZ, administered in a counterbalanced order. This within-subject experimental designed allowed us to assess the specific impact of chemogenetic inhibition during degradation (DCZ/Veh), test (Veh/DCZ), or both (DCZ/DCZ), as compared with a control group (Veh/Veh). Experimental groups’ sizes were the following: n = 8 Veh during degradation, n = 10 DCZ during degradation for Experiment 1; n = 10 Veh during degradation, n = 11 DCZ during degradation for Experiment 2.
Pavlovian contingency degradation task
Histology
Figures 1A and 2A show the location of AAV8-CaMKII-hM4D(Gi)-mCherry injections (with a representative image and a viral spread map) within the mPFC (A32D + A32V) and the OFC (VO + LO), respectively. Two rats from the mPFC group were excluded due to one-sided transgene expression.
Silencing mPFC CaMKII+ neurons impairs adaptive learning and responding to pavlovian S–O associations following contingency degradation. A, Representative image depicting the distribution of AAV8-CaMKII-hM4D(Gi)-mCherry expression within the mPFC (A32D + A32V) and schematic representation delineating injection site and viral expression extent within the subregion for all subjects, where each rat is represented as a unique stacked layer (n = 18). Scale bar, 1 mm. B, Rate of CS-PreCS magazine visits across pavlovian conditioning sessions. Data are shown based on the treatment to be received in the subsequent phase (Veh, n = 8 vs DCZ, n = 10) and the stimulus (Non-degraded vs Degraded). C, Magazine visit rate during the CS period expressed relative to the CS magazine visits in the last session of conditioning (% baseline). Data are shown based on the treatment (Veh vs DCZ) and the stimulus (Non-degraded vs Degraded). D, Rate of CS-PreCS magazine visits for the first presentation of each CS during the tests in extinction. Data are shown based on the treatment administered during degradation (Degrad) and test in extinction (Test). Data are expressed as mean + SEM. *p < 0.05. ***p < 0.001. †p < 0.05 (vs Veh/Veh Degraded).
Silencing OFC CaMKII+ neurons does not affect adaptive learning or responding to pavlovian S–O associations following contingency degradation. A, Representative image depicting the distribution of AAV8-CaMKII-hM4D(Gi)-mCherry expression within the OFC (VO + LO) and schematic representation delineating injection site and viral expression extent within the subregion for all subjects, where each rat is represented as a unique stacked layer (n = 21). Scale bar, 1 mm. B, Rate of CS-PreCS magazine visits across pavlovian conditioning sessions. Data are shown based on the treatment to be received in the subsequent phase (Veh, n = 10 vs DCZ, n = 11) and the stimulus (Non-degraded vs Degraded). C, Magazine visit rate during the CS period expressed relative to the CS magazine visits in the last session of conditioning (% baseline). Data are shown based on the treatment (Veh vs DCZ) and the stimulus (Non-degraded vs Degraded). D, Rate of CS-PreCS magazine visits for the first presentation of each CS during the tests in extinction. Data are shown based on the treatment administered during degradation (Degrad) and test in extinction (Test). Data are expressed as mean + SEM. **p < 0.01.
Pavlovian conditioning
Pavlovian conditioning (Figs. 1B, 2B) was measured by comparing magazine visit rates during the CS (20 s) and PreCS periods (20 s prior). Stimulus-evoked responding was revealed by an increased magazine visit rate during the CS period (Period) for both mPFC (F(1,32) = 62.67; p < 0.0001) and OFC animals (F(1,40) = 89.80; p < 0.0001). This gradual increase in responding to the CS was revealed by a significant interaction between Period and Session for both conditions (F(7,224) = 2.294, p = 0.0282 for mPFC and F(7,280) = 2.084, p = 0.0454 for OFC). The lack of effect of Stimulus (F(1,32) = 0.2775, p = 0.6020 for mPFC and F(1,40) = 0.0256, p = 0.8737 for OFC) or any interaction with this factor (F’s < 1, p > 0.05 for both groups) indicates that CSs were properly counterbalanced prior the degradation phase and conditioning was similar for both groups. One subject failed to acquire conditioned responding to both CSs and was therefore excluded from the data analysis.
Pavlovian contingency degradation
Figures 1C and 2C show the rate of magazine visits during each CS, relative to its corresponding baseline during the last session of conditioning, for mPFC and OFC rats, respectively. This measure was favored here over CS-PreCS magazine visits, given that non-contingent reward deliveries can occur during the PreCS period for the degraded CS, introducing a potential bias to this measurement. For the mPFC group (Fig. 1C), both Veh- and DCZ-treated animals displayed adaptive responding, characterized by a reduced cue-evoked response to the degraded CS, while maintaining a similar response rate to the non-degraded one. This was supported by statistical analysis, which indicated a significant effect of Stimulus (F(1,16) = 23.91; p = 0.0002) and a significant interaction between Stimulus and Session (F(5,80) = 5.298; p = 0.0003). Interestingly, statistical analysis also revealed a significant interaction between Session and Treatment at degradation (F(5,80) = 2.835; p = 0.0208). Upon looking at Figure 1C, this effect appeared to be driven by a heightened responding to the degraded cue on the last few sessions of degradation. Indeed, statistical comparison between Veh-degraded and DCZ-degraded data also revealed a significant interaction effect between Session and Treatment at degradation (F(5,80) = 6.168; p < 0.0001), with the DCZ group responding to the degraded cue significantly more on Day 5, as compared with controls (p = 0.0167). On the other hand, both Veh- and DCZ-treated animals of the OFC group displayed adaptive responding to the degraded CS (Fig. 2C). This was corroborated by statistical analyses, showing a significant effect of Stimulus (F(1,19) = 27.38; p < 0.0001), as well as a significant interaction between Stimulus and Session (F(5,95) = 5.024; p = 0.0004). Overall, we found that chemogenetic inhibition of mPFC (A32D + A32V), but not of OFC (VO + LO), CaMKII+ neurons during pavlovian contingency degradation impairs adaptive learning, as indicated by a higher responding to the degraded cue in comparison with controls.
Test in extinction
Figures 1D and 2D depict the rate of magazine visits during the first presentation of each CS, subtracting the magazine visits rate during the corresponding PreCS period serving as a baseline. For the mPFC group (Fig. 1D), rats treated with Veh during the degradation phase consistently exhibited a strong preference for the non-degraded CS, regardless of the treatment administered during the subsequent test in extinction (Veh/Veh and Veh/DCZ). In contrast, rats treated with DCZ during the degradation phase demonstrated a comparable rate of responding for both cues, irrespective of the treatment received during test (DCZ/Veh and DCZ/DCZ). This was confirmed by statistical analyses, revealing a significant interaction between Stimulus and Treatment at degradation (F(1,16) = 11.80; p = 0.0034), with post hoc tests confirming significance for Veh-treated animals (p < 0.0001) and non-significance for DCZ-treated animals (p = 0.3414). The response rate appeared to be lower for rats treated with DCZ at test, a trend that was nearly discernible in the statistical analysis, indicating a marginally non-significant effect of Treatment at test (F(1,16) = 4.316; p = 0.0542). Therefore, we decided to analyze data split by treatment at test. In the groups that received Veh at test (Fig. 1D, left half), we found a significant interaction between Stimulus and Treatment at degradation (F(1,16) = 8.938; p = 0.0087), with post hoc tests confirming significance for Veh/Veh animals (p = 0.0001) and non-significance for DCZ/Veh animals (p = 0.2824). Interestingly, DCZ-treated animals responded significantly more to the degraded cue, as compared with controls (p = 0.0478), an effect that mirrors what observed on Day 5 of the degradation phase. In the groups that received DCZ at test (Fig. 1D, right half), the interaction between Stimulus and Treatment at degradation was close to significance (F(1,16) = 3.834; p = 0.0679), with post hoc tests confirming significance for Veh/DCZ animals (p = 0.0128) and non-significance for DCZ/DCZ animals (p > 0.9999). No significant effects were observed for any other factors or interactions (F’s < 3; p > 0.05), except for an overall Stimulus effect (F(1,16) = 28.60; p < 0.0001). For the OFC rats (Fig. 2D), all groups seemed to favor the non-degraded CS, regardless of the treatment administered during the degradation phase or at test. This observation was corroborated by statistical analyses revealing a significant effect of Stimulus (F(1,19) = 13.43, p = 0.0016), with all other factors yielding non-significant results (F’s < 2; p > 0.05). Overall, we found that chemogenetic inhibition of mPFC (A32D + A32V), but not of OFC (VO + LO), CaMKII+ neurons during pavlovian contingency degradation impairs adaptive responding in unrewarded testing conditions, independently from treatment at testing, as indicated by an inability to discern between non-degraded and degraded cues, as compared with controls.
Experiment 3
In light of the differential effects observed in Experiment 1 and Experiment 2, and the hypothesized influence of the LC NA system in prefrontal-dependent learning processes, as a next step we aimed at characterizing NA projections to the mPFC (LC:mPFC) and the OFC (LC:OFC). To do so, we employed a unilateral double-retrograde viral strategy to simultaneously target the mPFC (A32D + A32V) and the OFC (VO + LO) within the same subject (n = 8). One region was injected with a green CAV-2 vector (CAV-2-PRS-eGFP), while the other was injected with a red CAV-2 vector (CAV-2-PRS-FsRed), counterbalancing hemispheres and regions between animals. These vectors enable the expression of either fluorescent protein under the control of a synthetic DβH promoter (PRS), therefore enabling selective targeting of NA neurons projecting to the subregions of interest. This allowed us to count the number of eGFP- and FsRed-positive cell bodies within the LC, i.e., quantify the amount of NA neurons projecting to either one subregion, the other, or both (Fig. 3A). Quantification of neuronal density within the LC region revealed 32% of labeled cell bodies projecting exclusively to the mPFC (320 labeled somas), 49% projecting exclusively to the OFC (492 labeled somas), and 19% projecting to both subregions (197 labeled somas), suggesting relatively segregated neuronal populations (Fig. 3B). Importantly, we found no difference of expression between neither vectors nor hemispheres. Next, we examined the distribution of labeled cell bodies within the LC along its anteroposterior axis (Fig. 3C). Our analysis revealed no distinct topographical organization along either the anteroposterior or dorsoventral axis. Instead, the organization of both mPFC- and OFC-projecting neurons appeared to align with the surface of the LC at each anteroposterior level, with no observed differences between those projecting to either region or both (Fig. 3D). Overall, the significant degree of segregation observed among NA projections to the mPFC and the OFC prompted us to investigate whether inhibiting these projections specifically would also result in a subregion-specific deficit and therefore offer a neuromodulatory interpretation for what observed in Experiment 1 and Experiment 2.
LC NA cells exhibit target-specific projections. A, Schematic illustrating unilateral double-viral injections of retrograde CAV-2-PRS-eGFP and CAV-2-PRS-FsRed vectors within the mPFC (A32D + A32V) and the OFC (VO + LO) and representative immunofluorescent staining images of corresponding FsRed+, eGFP+, and overlay retrogradely traced NA cell bodies within the LC. Scale bar, 100 µm. B, Overall quantification of NA cell bodies projecting to the mPFC (red), the OFC (green), or both (yellow) across all rats (n = 8). C, Spatial organization of NA cell bodies projecting to the mPFC (red), the OFC (green), or both (yellow) along the anteroposterior axis (from −9.5 mm to −10.3 mm relative to Bregma) in two representative rats. D, Spatial distribution of NA cell bodies projecting to the mPFC (red), the OFC (green), or both (yellow) along the LC anteroposterior axis across all rats. Data are expressed as sum of cell bodies.
Experiment 4 and Experiment 5
Considering the differential impact of manipulating mPFC and OFC neuronal activity on S–O updating observed in Experiment 1 and Experiment 2, along with the substantial segregation of NA projections to these subregions observed in Experiment 3, in a last set of experiments we aimed to explore whether chemogenetic inhibition of these sets of projections would yield any effect on S–O updating. To this end, rats were bilaterally injected with a retrograde CAV-2-PRS-HA-hM4Di(Gi)-hSyn-mCherry vector within the mPFC (A32D + A32V) or the OFC (VO + LO), in order to enable selective chemogenetic targeting of NA projections. Subsequently, the animals underwent the same recovery and behavioral procedures outlined in Experiment 1 and Experiment 2. Experimental groups sizes were the following: n = 9 Veh during degradation, n = 10 DCZ during degradation for Experiment 4; n = 13 Veh during degradation, n = 13 DCZ during degradation for Experiment 5.
Histology
Figures 4A and 5A illustrate the spread of CAV-2-PRS-HA-hM4Di(Gi)-hSyn-mCherry injections within the mPFC (A32D + A32V) and the OFC (VO + LO), respectively, together with a representative picture of mCherry-expressing retrogradely targeted neurons in the LC. We previously showed (Cerpa et al., 2023) that, while mCherry staining is also present at injection sites, reflecting local corticocortical connections that are not NA-dependent, HA-immunoreactive cell bodies were found exclusively in the LC, observations consistent with NA-specific expression of the HA-tagged hM4Di due to the PRS promoter, and nonselective expression of mCherry due to the hSyn promoter. However, considering that mCherry and HA immunoreactivity fully colocalize in the LC, in the current study we opted to use mCherry as a proxy of DREADDs expression.
Silencing mPFC NA inputs (LC:mPFC) impairs adaptive responding to pavlovian S–O associations following contingency degradation. A, Schematic representation delineating CAV-2-PRS-HA-hM4Di-hSyn-mCherry viral expression extent within the mPFC (A32D + A32V) for all subjects, where each rat is represented as a unique stacked layer (n = 19), and representative image showing retrogradely targeted LC cell bodies. Scale bar, 50 µm. B, Rate of CS-PreCS magazine visits across pavlovian conditioning sessions. Data are shown based on the treatment to be received in the subsequent phase (Veh, n = 9 vs DCZ, n = 10) and the stimulus (Non-degraded vs Degraded). C, Magazine visit rate during the CS period expressed relative to the CS magazine visits in the last session of conditioning (% baseline). Data are shown based on the treatment (Veh vs DCZ) and the stimulus (Non-degraded vs Degraded). D, Rate of CS-PreCS magazine visits for the first presentation of each CS during the tests in extinction. Data are shown based on the treatment administered during degradation (Degrad) and test in extinction (Test). Data are expressed as mean + SEM. **p < 0.01. ***p < 0.001.
Silencing OFC NA inputs (LC:OFC) does not affect adaptive learning or responding to pavlovian S–O associations following contingency degradation. A, Schematic representation delineating CAV-2-PRS-HA-hM4Di-hSyn-mCherry viral expression extent within the OFC (VO + LO) for all subjects, where each rat is represented as a unique stacked layer (n = 26), and representative image showing retrogradely targeted LC cell bodies. Scale bar, 50 µm. B, Rate of CS-PreCS magazine visits across pavlovian conditioning sessions. Data are shown based on the treatment to be received in the subsequent phase (Veh, n = 13 vs DCZ, n = 13) and the stimulus (Non-degraded vs Degraded). C, Magazine visit rate during the CS period expressed relative to the CS magazine visits in the last session of conditioning (% baseline). Data are shown based on the treatment (Veh vs DCZ) and the stimulus (Non-degraded vs Degraded). D, Rate of CS-PreCS magazine visits for the first presentation of each CS during the tests in extinction. Data are shown based on the treatment administered during degradation (Degrad) and test in extinction (Test). Data are expressed as mean + SEM. ***p < 0.001.
Pavlovian conditioning
Pavlovian conditioning (Figs. 4B and 5B) was measured by comparing magazine visit rates during the CS (20 s) and PreCS periods (20 s prior). Stimulus-evoked responding was revealed by an increased magazine visit rate during the CS period (Period) for both mPFC (F(1,36) = 222.8; p < 0.0001) and OFC animals (F(1,50) = 131.8; p < 0.0001). This gradual increase in responding to the CS was revealed by a significant interaction between Period and Session for both conditions (F(7,252) = 6.955, p < 0.0001 for mPFC and F(7,350) = 22.13, p < 0.0001 for OFC). No effect of Stimulus or any interaction with this factor were observed for the OFC group (F’s < 2; p > 0.05). An effect of Stimulus was found in the mPFC group (F(1,36) = 5.508; p = 0.0245). However, this effect disappeared when considering a CS-PreCS measure (F(1,17) = 2.312; p = 0.1468). Moreover, no interaction of this factor with Treatment at degradation (to be administered) was present (F(1,17) = 0.069; p = 0.7960), indicating similar performances between experimental groups.
Pavlovian contingency degradation
Similarly to our first set of experiments, we opted for the last session of conditioning as a baseline for responding to each CS, rather than relying on the corresponding PreCS period. We observed a similar impact of degradation in mPFC and OFC groups, with evidence of adaptive responding, characterized by a reduction in the CS-evoked response to the degraded cue, while the response to the non-degraded CS remained unchanged compared with their respective baselines (Figs. 4C and 5C). This was substantiated by statistical analysis, with a significant effect of Stimulus (F(1,17) = 20.74, p = 0.0003 for mPFC and F(1,24) = 46.71, p < 0.0001 for OFC) and a significant interaction between Stimulus and Session (F(5,85) = 2.835, p = 0.0204 for mPFC and F(5,120) = 2.937, p = 0.0155 for OFC). Overall, this suggests that both the mPFC and the OFC groups appeared to have successfully adapted to the degraded cue, irrespective of the treatment received during the degradation phase.
Test in extinction
Figures 4D and 5D depict the rate of magazine visits during the first presentation of each CS, subtracting the magazine visits rate during the corresponding PreCS period serving as a baseline. For the mPFC group (Fig. 1D), statistical analyses revealed a significant interaction between Stimulus and Treatment at degradation (F(1,17) = 7.973; p = 0.0117). On the one hand, rats treated with Veh during the degradation phase consistently exhibited a strong preference for the non-degraded CS (p < 0.0001), regardless of the treatment administered during the subsequent test in extinction (Veh/Veh and Veh/DCZ). On the other hand, rats treated with DCZ during the degradation phase demonstrated a comparable rate of responding for both cues (p = 0.1275), irrespective of the treatment received during tests (DCZ/Veh and DCZ/DCZ). Despite this analysis, the response rate appeared to be impaired exclusively in rats treated with DCZ at both degradation and test, although no Treatment at degradation × Treatment at test (F(1,17) = 0.2873; p = 0.5989) or Stimulus × Treatment at degradation × Treatment at test (F(1,17) = 0.2873; p = 0.5989) interaction effects were present. For this reason, we decided to analyze data split by treatment at test, just like in Experiment 1. On the one hand, in the groups that received Veh at test (Fig. 4D, left half), we found a significant effect of Stimulus (F(1,17) = 13.35; p = 0.0020), but no interaction between Stimulus and Treatment at degradation (F(1,17) = 0.8629; p = 0.3659). On the other hand, in the groups that received DCZ at test (Fig. 4D, right half), we found a significant interaction between Stimulus and Treatment at degradation (F(1,17) = 10.19; p = 0.0053), with post hoc tests confirming significance for Veh/DCZ animals (p = 0.0004) and non-significance for DCZ/DCZ animals (p > 0.9999). No significant effects were observed for any other factors or interactions (F’s < 2; p > 0.05), except for an overall Stimulus effect (F(1,17) = 30.84; p < 0.0001). For the OFC rats (Fig. 5D), all groups seemed to favor the non- degraded CS, regardless of the treatment administered during the degradation phase or at test. This observation was corroborated by statistical analyses revealing a significant effect of Stimulus (F(1,24) = 24.18; p < 0.0001), with all other factors yielding non-significant results (F’s < 3; p > 0.05). Overall, we found that chemogenetic inhibition of NA projections to the mPFC (A32D + A32V), but not to the OFC (VO + LO), during pavlovian contingency degradation impairs adaptive responding in unrewarded testing conditions, as indicated by an inability to discern between non-degraded and degraded cues, as compared with controls. Importantly, this deficit appears evident only if those same NA projections are inhibited at testing as well, indicating that mPFC NA plays a role in both the updating and the retrieval of degraded S–O associations.
Discussion
The significance of incoming signals frequently changes in natural settings. This solicits organisms to regularly track the ongoing predictive value of environmental cues and adjust their behavior accordingly. Degrading the contingency between a stimulus (S) and its associated outcome (O) is one effective way to study such adaptive behavior. The current study presents evidence of mPFC (A32D + A32V) neurons, and more specifically of noradrenergic (NA) transmission within the subregion, being crucial for adaptive responding following pavlovian contingency degradation.
A wealth of literature has shown that the mPFC plays a crucial role in mediating the degradation of instrumental action (A)–O contingencies in rodents (Woon et al., 2020). Indeed, neurotoxic lesions of the prelimbic subdivision of the mPFC (A32D) prevented this update in rats (Balleine and Dickinson, 1998; Corbit and Balleine, 2003; Coutureau et al., 2012), as did chemogenetic silencing in mice (Swanson et al., 2017), an effect that is believed to be dopamine dependent, as suggested by both DA lesions and inactivation of D1/D2 receptors (Naneix et al., 2009; but see Lex and Hauber, 2010), and mediated by ventral hippocampal (vHPC) inputs, as shown by chemogenetic synaptic silencing of vHPC terminals (Piquet et al., 2023). A relatively low cross-species homology in prefrontal functioning hampered a clear-cut translation of these findings in non-human primates (Duan et al., 2021), although reports implicating subdivisions of the mPFC in instrumental degradation exist (Jackson et al., 2016). Whether the mPFC is implicated in the degradation of pavlovian S–O contingencies, on the other hand, remained a long-standing question. In here, we show that chemogenetic inhibition of mPFC (A32D + A32V), but not of orbitofrontal (OFC, VO + LO), CaMKII+ neurons impairs the rats’ ability to both learn the degradation and appropriately respond between non-degraded and degraded cues in unrewarded testing conditions. This suggests that mPFC neurons play a key role in both encoding and recalling degraded S–O associations. Worth mentioning, an important division of labor has been proposed in which the prelimbic (A32D) and infralimbic (A32V) subdivisions of the mPFC regulate the expression and suppression of adaptive behaviors (Quirk et al., 2000; Milad and Quirk, 2002; Rosenkranz et al., 2003; Killcross and Coutureau, 2003; Maren and Quirk, 2004). This division of labor remains to be investigated in the framework of pavlovian contingency degradation and surely represents an interesting direction for future investigations.
The fact that chemogenetic inhibition of OFC CaMKII+ neurons did not affect the animals’ ability to learn the degradation or discriminate between cues in unrewarded testing conditions is a result in apparent contrast with what previously reported (Ostlund and Balleine, 2007; Alcaraz et al., 2015). Different variables might account for discrepancies between the current and these previous studies. First of all, there are evident differences relative to target sites, with lesioning procedures spreading either more laterally or more anteriorly within the OFC, as compared to our viral strategy, which consistently targeted VO and LO, leaving the anterior insular, medial orbitofrontal, prelimbic and infralimbic cortices untouched. Considering that relatively close prefrontal areas—or even the same area alongside its axis—often account for different behavioral effects (Alsiö et al., 2020; Fisher et al., 2020; Cerpa et al., 2023), we cannot exclude that inconsistencies in neuroanatomical targeting might represent an important source of variability. Secondly, when examining behavioral protocols, core experimental differences arise between the current and the abovementioned studies, including the number of sessions and reward deliveries, the testing procedure (rewarded vs unrewarded conditions), and—most importantly—the degradation procedure itself. Specifically, differently from Ostlund and Balleine, who only manipulated the likelihood of receiving the outcome after the cue (backward transition probability) while keeping the reward probability during the cue (forward transition probability) unchanged, we modified both forward and backward probabilities. Consequently, the impairment observed in their study could be tied to exclusive modulation of the backward probability. At last, we cannot exclude that differences in the silencing techniques (postconditioning lesions vs DREADD-mediated inhibition), as well as in the behavioral measures adopted (entries/min vs entries as a % of baseline responding), might also represent a potential source of variability.
A critical aspect of the neuronal mechanisms involved in adaptive behavior involves catecholaminergic neuromodulation, which can be functionally linked to detecting novelty, producing and interpreting prediction errors and sustaining working memory. Neurons in the LC are believed to track uncertainty by phasically responding to novel stimuli (Bouret and Sara, 2004; Tait et al., 2007; McGaughy et al., 2008; Tervo et al., 2014; Uematsu et al., 2017; Jahn et al., 2018; Cope et al., 2019). Compelling theoretical models hypothesize that the LC interacts with the prefrontal cortex to support adaptive behaviors (Sara and Bouret, 2012; Cerpa et al., 2021). Specifically, when a change in contingencies is detected, a rise in prefrontal NA activity is believed to trigger behavioral adaptations (Aston-Jones et al., 1997; Bouret and Sara, 2005; Sadacca et al., 2017). However, these NA bursts are believed to be highly segregated, in line with the undergoing conception of the LC as a non-homogenous nucleus, in both its structure and function (Chandler et al., 2019; Poe et al., 2020).
In here, using NA-specific CAV-2-PRS constructs (del Rio et al., 2019; Hayat et al., 2020), we found that direct and discrete LC NA projections to the mPFC and the OFC exist, thereby adding to the growing evidence of a highly modular LC architecture (Agster et al., 2013; Chandler et al., 2013, 2014, 2019). In a seminal paper, Chandler et al. already revealed largely segregated population of cells projecting to various cortical regions using retrograde tract tracing, in particular OFC vs M1, mPFC vs M1, and ACC vs M1 (Chandler et al., 2014). However, to the best of our knowledge, we are the first to directly compare mPFC versus OFC LC NA projections, as well as to describe the gradient of LC NA projections throughout the anteroposterior axis of the LC. Most importantly, we found that chemogenetic inhibition of LC NA projections to the mPFC, but not to the OFC, impaired the rats’ ability to correctly respond between non-degraded and degraded cues in unrewarded testing conditions. This result suggests that NA might be specifically responsible for the distinctive behavioral outcomes observed in the first set of experiments and, more specifically, for the inability to recall degraded S–O associations in unrewarded testing conditions.
Disentangling the role of different prefrontal subregions and neurotransmitters is a key step to advance our knowledge of adaptive and maladaptive behaviors. In this sense, the current study belongs to a recent and ever-growing body of work (Cerpa et al., 2021, 2023; Chernoff et al., 2023; Cremer et al., 2023) that not only is integrating and slowly revising the classical conceptions linking the mPFC to instrumental learning and the OFC to pavlovian learning, but suggesting a major role for NA in the neuromodulatory computations that underlie higher-order cognitive processes. For instance, a similar, but opposite, dissociation of effects was previously observed in another recent work from our laboratory (Cerpa et al., 2023). In such study, also using the CAV-2-PRS-HA-hM4Di-hSyn-mCherry vector, we showed that LC NA projections to the OFC (VO + LO), but not to the mPFC (A32D + A32V), are required for both encoding and recalling the identity of an expected instrumental outcome, specifically when that identity has been reversed. Furthermore, another recent study, this time using a pharmacological approach, highlighted a double dissociation between the behavioral effects of NA transmission across these prefrontal subregions, this time with respect to risky choice and impulsive action (Chernoff et al., 2023).
The cellular and molecular mechanisms underlying these NA-mediated effects remain to be elucidated. However, new overarching theories of LC function, like the global model failure system theory (Jordan, 2023), postulate that cortical top-down predictive inputs to the LC, i.e., prediction errors (PE), are followed by bottom-up responsive outputs that ultimately control synaptic integration, driving behavioral change. Under this conception, novel and uncertain environments would prompt high rates of PE, resulting in high LC outputs. The current work, together with our previous study (Cerpa et al., 2023), provides behavioral evidence of a strong cortical parcellation in these bottom-up responses, in line with recent conceptions of prefrontal functioning relatively to goal-directed behavior (Turner and Parkes, 2020; Cerpa et al., 2021). We propose that, while OFC NA projections might be implicated in adaptations to action violations and outcome identity changes, as happens in reversal learning, mPFC NA projections might respond to context-dependent incentive memory violations, as for the case of instrumental and pavlovian degradation. Whether DA transmission in the mPFC is also required for the latter form of adaptive responding remains to be tested. Finally, we must acknowledge that only male rats were used in the current study. The LC displays some anatomical and physiological variations between male and female rats (Joshi and Chandler, 2020); therefore, a thorough characterization in future studies would need to integrate the sex factor.
In summary, the current findings provide clear evidence of prefrontal parcellation of function in the framework of adaptive responding. Our results question the previously assumed role of the OFC in pavlovian contingency degradation and point instead at the mPFC as a key player. Notably, we show that this effect is driven—at least in part—by NA innervation of the subregion, adding to current theories suggesting a major, but complex and modular, role for the LC in the regulation of flexible, goal-directed decision-making.
Footnotes
This work was supported by the French National Research Agency (CE37-0019 NORAD and CE14-0020 FRONTOFAT to E.C.). The vectors used in Experiments 3, 4, and 5 were generously provided by Eric J. Kremer and Marina Lavigne (IGMM, Montpellier, France). We thank Angélique Faugère for her help with immunofluorescence and Yoan Salafranque for his expert animal care.
↵*A.P. and H.P. contributed equally to this work.
The authors declare no competing financial interests.
- Correspondence should be addressed to Etienne Coutureau at etienne.coutureau{at}u-bordeaux.fr.
