Characterization of glutamatergic VTA neural population responses to aversive and rewarding conditioning in freely-moving mice

The Ventral Tegmental Area (VTA) is a midbrain structure known to integrate aversive and rewarding stimuli, a function involving VTA Dopaminergic and GABAergic neurons. VTA also contains a less known population: glutamatergic (VGluT2) neurons. Direct activation of VGluT2 soma evokes rewarding behaviors, while stimulation of their axonal projections to the Nucleus Accumbens (NAc) and the Lateral Habenula (LHb) evokes aversive behaviors. Here, a systematic investigation of the VTAVGluT2+ population response to aversive or rewarding conditioning facilitated our understanding these conflicting properties. We recorded calcium signals from VTA glutamatergic population neurons using fiber photometry in VGluT2-cre mice to investigate how the VTA glutamatergic neuronal population was recruited by aversive and rewarding stimulation, both during unconditioned and conditioned protocols. Our results revealed that, as a population, VTAVGluT2+ neurons responded similarly to unconditioned-aversive and unconditioned-rewarding stimulation. During aversive and rewarding conditioning, the CS-evoked responses gradually increased across trials whilst the US-evoked response remained stable. Retrieval 24 h after conditioning, during which mice received only CS presentation, resulted in VTAVGluT2+ neurons strongly responding to CS presentation and to the expected-US but only for aversive conditioning. The inputs and outputs of VTAVGluT2+ neurons were then investigated using Cholera Toxin B (CTB) and rabies virus, and we propose based on all results that VTAVGluT2+ neurons specialized function may be partially due to their connectivity.


Introduction The Ventral Tegmental Area
The ventral tegmental area (VTA) is a midbrain structure that has been linked with a variety of behavioral functions including aversion and reward Lammel, Lim, and Malenka, 2014;Barker et al., 2016), prediction error (Watabe-Uchida, Eshel, and Uchida, 2017;, and motivation (Arsenault et al., 2014;van Zessen et al., 2012). The heterogeneous composition of VTA includes a large proportion of dopaminergic (DA) neurons (60%), and smaller proportions of GABAergic neurons (GABA) (35%) and glutamatergic neurons (2-5%) (Yamaguchi, Sheen, and Morales, 2007;Yamaguchi et al., 2015;Nair-Roberts et al., 2008). Whilst the two first neuronal components (DA and GABA) of VTA have been well studied and characterized , the glutamatergic population has received less attention and its functional characterization needs to be elucidated to obtain a better understanding of VTA function.

VTA dopaminergic and GABAergic neuronal function during aversion and reward
In-vivo electrophysiology experiments have shown that VTA dopaminergic (DA) neurons increase their firing rate following rewarding stimulation (Lammel, Lim, and Malenka, 2014;Schultz, Dayan, and Montague, 1997;. However, if the reward is paired with a conditioned stimulus (CS), such as a sound cue, the activity of DA neurons gradually shifts from responding to the reward presentation, to responding to the CS presentation (Bromberg-Martin and Hikosaka, 2009;. At the same time, VTA GABAergic neurons can signal expected outcome (Cohen et al., 2012) by inhibiting neighboring DA neurons (Cruz et al., 2004;Tan et al., 2010;Tan et al., 2012;Eshel et al., 2015), which corresponds with the decrease in firing rate observed in DA neurons following aversive stimulation (Ungless, Magill, and Bolam, 2004). This mechanism of reward prediction via VTA GABAergic neuronal inhibition of local DA neurons is further supported by recent optogenetic experiments (Tan et al., 2012), in which direct optogenetic activation of VTA-GABA neurons leads to local VTA-DA neuronal inhibition and also to place aversion (Tan et al., 2012).

Physiology, anatomy and function of VTA glutamatergic neurons
While there is a general consensus regarding the role of VTA-DA and VTA-GABA neurons in control of aversive and rewarding behavior, the role of VTA's glutamatergic neurons is not yet understood Morales and Margolis, 2017). Glutamatergic neurons, defined by their expression of vesicular glutamate transporter 2 (VGluT2) (Yamaguchi, Sheen, and Morales, 2007;Kawano et al., 2006), can form local connections in VTA (Dobi et al., 2010) and send long-range projections to structures such as the Nucleus Accumbens (NAc) or the Lateral Habenula (LHb) Qi et al., 2016;. In addition, VGluT2 neurons are a heterogeneous population in terms of molecular and physiological characteristics: one population of VGluT2 neurons releases only the excitatory neurotransmitter glutamate, whilst another fraction can release glutamate and can also co-release dopamine or GABA (Yamaguchi et al., 2011;Morales and Margolis, 2017;Kawano et al., 2006).
Optogenetic stimulation of VTA VGluT2+ neuronal somata promotes rewarding behaviors, such as place preference and appetitive instrumental conditioning ; however, at the microcircuit level the function of their interaction with neighboring DA and GABA neurons remains unknown (Dobi et al., 2010). Conversely, VTA glutamatergic transmission has also been associated with aversive behaviors (Han et al., 2017). Indeed, optogenetic activation of axonal projections from VTA VGluT2+ neurons mainly elicit aversive responses, such as escape or avoidance. More specifically, optogenetic activation of VTA VGluT2+ terminals in the nucleus accumbens (NAc) or to the lateral habenular nucleus (LHb) both promote aversive behaviors, including aversive conditioning . Recent electrophysiological recordings of glutamatergic neurons confirmed that individual VTA VGluT2+ neurons can be activated by aversive stimulation, and either excited or inhibited by rewarding stimulation (Root, Estrin, and Morales, 2018), giving an initial insight into this seemingly paradoxical function. However, population response recordings, such as calcium imaging, can potentially help us to understand VTA VGluT2+ function in terms of conditioned aversion and reward. In particular, we use fiber photometry with the genetically-encoded Ca 2+ indicators GCaMP6s (Chen et al., 2013), which is a minimally invasive method that allows in-vivo measurements in freely-moving animals of synchronous neuronal population activity from subcortical structures (Resendez and Stuber, 2014;Guo et al., 2015), which has proved useful during conditioning (Daqing et al., 2017).
Here, we investigated the VTA VGluT2+ neuronal population response to aversive and rewarding events, both during unconditioned and conditioned stimulation. Conditioning consisted in pairing a tone (CS+) with an aversive (footshock) or rewarding (sucrose) unconditioned stimulus (US). In addition, a retrieval test was performed 24 h after conditioning to see whether VTA VGluT2+ neurons maintain a robust memory of aversive or rewarding conditioning. Finally, to better understand VTA VGluT2+ population responses to aversion and reward, we investigated their connectivity pattern using cell specific monosynaptic retrograde rabies virus tracing, allowing the mapping VTA VGluT2+ inputs.

Animals
All procedures were approved by Animal Care and Use Committees in the Shenzhen Institute of Advanced Technology (SIAT) or Wuhan Institute of Physics and Mathematics (WIPM), Chinese Academy of Sciences (CAS). Adult (6-8 weeks old) male VGluT2-ires-cre (Jax No.016963, Jackson Laboratory) transgenic mice were used in this study. All mice were maintained on a 12/12-h light/dark cycle at 25°C. Food and water were available ad libitum.
Mice were sacrificed one week after this second injection. All rabies tracing experimental procedures were completed in Biosafety level 2 (BSL2) Laboratory.

Implantation of optical fibers
A 200 µm optical fiber (NA: 0.37; NEWDOON, China) was chronically implanted in the VTA of VGluT2-ires-cre mice 2-3 weeks following virus expression for fiber photometry experiments. The optical fiber was unilaterally implanted in VTA (AP:-3.15 mm, ML: -1.10 mm and DV:-4.2 mm) with a 15° angle in the medial direction of the transverse plane.
After surgery all mice were allowed to recover for at least 2 weeks.

Histology, immunohistochemistry, and microscopy
Mice were sacrificed by overdosing with pentobarbital (1% m/v, 150 mg/kg, i.p.) and transcardially perfused with 1M cold saline followed by ice-cold 4% paraformaldehyde (PFA; Sigma) in 1M PBS. Brains were removed and submerged in 4% PFA at 4°C overnight to post-fix, and then transferred to 30% sucrose to equilibrate. The coronal brains slices (40 µm) were sectioned with a cryostat (CM1950; Leica, Germany). Freely floating sections were washed with PBS and blocked for 1 h at room temperature in blocking solution containing 0.3% Triton X-100 and 10% normal goat serum (NGS). Then the sections were incubated overnight with rabbit monoclonal anti-dsRed (1:500, #632496; Clontech; Japan); GFP(1:500, #ab290, abcam, USA); DAPI (1:50,000, #62248; Thermo Fisher Scientific, USA) diluted in PBS with 3% NGS and 0.1% TritonX-100. The sections were incubated for 1 h at room temperature with Alexa Fluor 488 or 594 goat anti-rabbit secondary antibody (1:200; Jackson Laboratory, USA). Finally, the sections were mounted and photographed using the Zeiss LSM 880 confocal microscope (Zeiss; Germany). The images were acquired using identical gain and offset settings, and analyzed with ImageJ, Image Pro Plus, and Adobe Photoshop software. ROIs were traced with reference to the "The mouse brain in stereotaxic coordinates" by George Paxinos and Keith B. J. Franklin. CTB and Rabies Virus immunoreactivity was quantified using Image Pro Plus and was verified by comparing with manual counts performed by a trained double-blind observer.

Unconditioned aversive and Conditioned aversive stimulation
VGluT2-ires-cre mice (N=8) with optical fibers implanted were placed in an unescapable acrylic box (L 25 × W 25 × H 70 cm) with a metal grid floor that delivered footshock currents (0.6 mA footshock, 0.5 s). Each mouse went through an unconditioned and then a conditioned protocol as below. During unconditioned aversive stimulation, mice were freely moving and footshocks were directly delivered with inter-trial interval durations varying within session randomly set in a range between 60-120 s. The session was approximately 10 min long and each mouse received 10 footshoocks. The conditioned sessions consisted of 5 trials where an auditory conditioned stimulus (CS; 3 kHz, sine wave, 90 dB, 5 s) was paired with an unconditioned stimulus (US; 0.5 s, 0.6 mA footshock; random inter-trial intervals 60-120 s) that began immediately after tone ended. Mice were presented with 5 CS cues alone, without footshock stimulation, 24 h after conditioning.

Unconditioned Reward and Conditioned reward
Following viral injections and optical fiber implantations, VGluT2-ires-cre mice (N=8) underwent a third surgery to implant a steel headplate for head-fixing purposes (Chen et al., 2013). The mice were habituated (~30 min/day) to the head-fix system over two-three days. During the experiment, each mouse was head-fixed and a tube delivering liquid reward was directly aimed at their mouth, through which single drops of sucrose (5% w/v) could be delivered as reward. Each mouse went through an unconditioned and then a conditioned protocol as described below. Unconditioned reward sessions were conducted during which 30 reward trials were presented with inter-trial interval durations varying within session, randomly set in a range between 25-40 s. Conditioned reward sessions consisted of one session of 30 trials in which an auditory conditioned stimulus (CS; 10 kHz, sine wave, 80 db, 5 s) was paired with one sucrose delivery, which was delivered immediately after the tone ended. Mice were presented with 30 CS cues alone, without sucrose reward, 24 h after conditioning.

Fiber photometry
Ca 2+ signals were recorded using a fiber photometry system (Thinker Tech, Nanjing).
Two weeks post AAV2/9-DIO-Gcamp6s virus injection, an optical fiber (NA: 0.37; NEWDOON, China) was implanted into VTA as described above. Behavioral event signals were recorded by a DAQ card (NI, at 1000 Hz using the same LabVIEW program.

Photometry data analysis:
Calcium Imaging signals were first extracted using Blackrock NPKM (Neural Processing MATLAB Kit), using provider instructions (Thinker Tech, Nanjing). Custom MATLAB (The MathWorks Inc. ©) scripts were developed for further analysis using R2012a. Signals were analyzed as dF/F = (F − Fb)/Fb, where Fb was defined as the baseline fluorescence before stimulation. Data were then smoothed using a 10 ms sliding windows. Time courses were calculated by aligning the time of stimulation across all individual trials and then calculating the mean change in calcium at each time window. To compare calcium activity between conditions, mean calcium activity was calculated for 0.5 s time windows centered around the time of the activity peak (2 s before stimulation vs. CS vs. US). A multivariate permutation (1000 permutations, ⍺ level of 0.05) test was used to test the statistical significance of the difference between conditions over the time course, and a threshold indicating a statistically significant difference from the baseline was applied (p<0.005).
Area Under Curve index is the sum of transient Ca 2+ activity ) over a period of 0.5 s centered around the peak of activity.

VTA VGluT2+ population increases activity to unconditioned aversive stimulation and this response remains constant over successive trials
We first investigated whether the VTA VGluT2+ neuronal population responds to unconditioned aversive stimulation. To do that we first infected VGluT2-cre animals with adenoassociated virus (AAV) expressing GCaMP indicator by injecting the AAV9-EF1a-DIO-GCaMP6s virus in VTA ( Fig. 1.A). Three weeks later an optical fiber was implanted above VTA, allowing in-vivo recording of VTA VGluT2+ calcium signals during freely-moving behavior ( Fig. 1.A-B). At the end of experiments, GCaMP6s virus expression in VTA ( Fig. 1.C) and fiber positioning were systematically checked in every mouse. (Fig. 1.D). During the unconditioned aversive experiment, mice received footshocks (0.5 s at 0.6 mA) whilst the activity of VTA VGluT2+ neurons were recorded. Immediately after the beginning of the footshock, the calcium signal of VTA VGluT2+ neurons strongly increased for each individual mouse ( Fig. 1.E, top), which was a stereotypical effect well aligned with the onset of stimulation ( Fig. 1.E, bottom). All mice expressed a similar increase of activity (4.26% DF/ F, n=8) directly after aversive stimulation, which was significantly different from baseline expression for 1.54 s before returning to baseline level ( Fig. 1.F, red part of the curve indicates p<0.05 using the multivariate permutation test). The mean signal values for all mice for a period of 0.5 s around the peak response amplitude (T=0.68 s) revealed that activity was significantly higher than baseline (BL=0.002% DF/F vs. Footshock=3.96% DF/ F, p<0.0001; Fig. 1

.G).
To observe the effect of repeated unconditioned stimulation on VTA VGluT2+ neurons, we analyzed response trends on a trial-by-trial basis and across animals ( Fig. 1.H). The peak responses in successive trials remained at a similar level ( Fig. 1.H), which was confirmed by computation of the Area Under Curve index (AUC) ( Fig. 1

.I).
In summary, this experiment demonstrated that VTA VGluT2+ neurons were strongly activated by unconditioned aversive stimulation and the amplitude of the peak response remained constant across trials.

The VTA VGluT2+ neuronal population responds to aversive conditioning
To characterize the responses of VTA VGluT2+ to conditioned aversive stimulation ( Fig. 2

.A)
we applied the following protocol: (i) Habituation Day, during which animals received tone stimulation only; (ii) Conditioning Day, during which a tone (CS) was paired with an unconditioned stimulation (US); (iii) Retrieval Day, during which only the CS was presented.
On Habituation Day VTA VGluT2+ neurons were insensitive to the CS as shown in the example of mouse #141 (Fig. 2.B-C, left), demonstrating that a neutral stimulus was insufficient to evoke a significant response in this neuronal population. During the Conditioning Day, the response of individual mice to CS was larger compared to Habituation Day. All mice had a strong evoked response to the US, similar to the unconditioned footshock experiment (Fig. 2.B-C, middle). Following conditioning, increased sensitivity to CS was evident in the mean responses of all mice ( Fig. 2.D; n=8), where CS-evoked and US-evoked response amplitudes were significantly higher than the baseline (BL-evoked=0.12%DF/F; CS-evoked=3.8% DF/F, p<0.0001; US-evoked=8.68% DF/F, p<0.0001; Fig. 2.E), and US-evoked signal was still significantly stronger than CSevoked signal (p<0.01). As a control, another group of mice infected with a GFP virus followed the same aversive conditioning protocol, and did not exhibit Ca 2+ signal variations at the time of either CS and US (Sup. Fig. 2.B-C), demonstrating that the signal is not due to artifacts.
Furthermore, the amplitude of the CS-evoked peak of Ca 2+ response increased gradually over trials ( Fig. 2 US=2.86, p<0.005; Fig. 2.J), indicating that VTA VGluT2+ neuronal responses to conditioned aversive stimulation were sustained over time. This also strongly suggests that responses may be due not only to the presence of physical stimuli but may also be due to a component of expectation. Across successive trials, there was a trend for both CS-evoked and Expected-US-evoked peak responses to decrease by a small fraction (Fig. 2.K & Sup. Fig. 2.A), although this was not significant (Fig. 2.L). Testing for a signal decrease over trials would be interesting to investigate in the future, but the results here indicate that aversive memory is robust and persistent for at least a period of 24 h.
In summary, these results show that during Conditioning Day, the amplitude of the CSevoked responses of VTA VGluT2+ neurons gradually increased over repetitions, while USevoked activity remains relatively constant. During Retrieval Day CS-evoked responses are similar to those observed during Conditioning Day and there is an Expected-USevoked response, which indicates that the VTA VGluT2+ population is sensitive and adapts to aversive conditioning and responds to aversive stimulation for at least 24 h.

VTA VGluT2 neurons respond to successive repetition of unconditioned rewarding stimulation
Since direct stimulation of VTA VGluT2+ neurons promote reward , and that unitary electrophysiological recording has shown that some VTA VGluT2+ neurons are sensitive to rewarding stimulation (Root, Estrin, and Morales 2018), we next wanted to investigate glutamatergic neuron population responses to rewarding stimulation. To answer this question, calcium signals were recorded whilst mice were head-fixed and a tube directly delivered a liquid reward in their mouth (5 % sucrose water, ITI=60-120 s, 30 trials) (Fig. 3.A). Fiber position was carefully verified in brain slices for each individual mouse at the end of experiments (Fig. 3.B). Following rewarding stimulation, a rapid increase of calcium activity was observed in VGluT2 neuronal populations of individual mice, as shown in Fig. 3.C. An increase of calcium activity just after reward delivery was observed across all mice that was significantly different from baseline measurements for 8.16 s (Fig. 3.D). Analysis of the amplitude of the peak of activity revealed that it was significantly higher than the baseline (BL=0.07 % DF/F vs. Reward=4.72 % DF/F, p<0.0001). To check for a potential gradual change of signal amplitude across stimulation, we calculated the mean in groups of 5 consecutive trials and found stable activity across trials ( Fig. 3.F-G), similar to responses to unconditioned aversive stimulation.
Here, we show that VTA VGluT2+ neurons were excited by rewarding stimulation and the amplitude of their activation remained stable across stimulation trials.

VTA VGluT2 neurons respond to reward conditioning
We have demonstrated that the VTA VGluT2+ population gradually learned to respond to a CS tone preceding aversive stimulation. To determine if rewarding stimulation can evoke a similar response pattern, mice were conditioned to a reward; the CS was a 5 s tone and the US was 5% sucrose water (random ITI 60-120 s). The experiment was conducted over three consecutive days, similarly to aversive stimulation described above: Habituation Day, Conditioning Day, and Retrieval Day (Fig. 4.A). During the Conditioning Day, an increase in calcium signal evoked by CS was observed, followed by a higher amplitude increase evoked by the US (Fig. 4.B). Group data confirmed this pattern (Fig. 4.D), revealing that the difference between baseline and CS stimulation-evoked VTA VGluT2+ population response was statistically significant and sustained for 3.88 s, and that the US stimulation evoked a larger activity increase that slowly returned to baseline level over 13.88 s. The CS-evoked and US-evoked activity peaks were significantly higher than the baseline (BL=0.02 % DF/ F, CS=2.53 % DF/F, p<0.0001; US=11.88 % DF/F, p<0.0001, Fig. 4.E), and the US-evoked activity significantly higher than the CS activity (p<0.0001). To investigate trial-by-trial changes, we plotted groups of 5 trials (Fig. 4.F); this shows that CS-evoked activity slowly increased across trials, which was confirmed by calculating the area under curve (Fig.   4.G), whereas US-evoked responses remained stable across trials (Fig. 4.F and Fig. 4.G).
In addition, it appeared that across conditioning, CS-evoked activity remained sustained until the beginning of the rewarding stimulation, which can be seen by increasing the time window to take into account the long sustained activity during CS (Trial 1-5= 0.84 AUC vs.
Trial 26-30=3.81, p<0.05, Sup. Fig. 4.B). During the Retrieval day, a calcium signal peak was visible during CS and US as shown in Fig. 4.C, but of low amplitude compared to baseline. Group mean calcium signal changes show that the CS did not evoke any clear change, whereas expected-US evoked very brief activity trend (Fig. 4.H); however, this was not statistically significant when looking at a 0.5 s time window around activity peak ( Fig. 4.I). In line with these results, no pattern was found across trials ( Fig. 4.J and Sup.

Fig. 4.C).
Together, these data demonstrate that the VTA VGluT2+ population responds to rewarding conditioning with increasing CS-evoked activity amplitude and duration, whilst US-evoked activity remained stable. But during retrieval, VTA VGluT2+ neurons did not respond. These results indicate VTA VGluT2+ population is sensitive to rewarding conditioning, but we could not demonstrate formation of rewarding memories.

VTA VGluT2 neurons may be characterized by their specific network
The VTA VGluT2+ neuronal population can respond to both aversive and rewarding stimulation and is considered a heterogeneous population. Determining which broader network they belong to may be an alternative strategy to characterize them . First, to determine whether VTA VGluT2+ neurons send collaterals to structures serving a similar function, we injected in the same Vglut2-cre mice both the retrograde tracer Cholera Toxin-B conjugated with Alexa 594 (CTB-594) in NAc and Cholera Toxin-B conjugated with Alexa 488 (CTB-488) in LHb, two structures downstream of VTA VGluT2+ neurons that are associated with aversive responses. Mice VTA were then were infected with AAV-DIO-GFP virus (Fig. 5.A). After expression, CTB was retrogradely transported from NAc and LHb terminals to VTA cell bodies (Fig. 5.B-C (Fig. 5.D-E). Only a minority of labeled neurons coexpressed VGLuT2 and CTB marker (0.11% for NAc, and 0.04% for LHb), revealing glutamatergic neurons account for a minority of VTA projections to NAc and LHb. Importantly, there was no cell in VTA expressing both CTB-494 and CTB-488 (Fig. 5.B), suggesting that VTA VGluT2+ neurons do not send collaterals to these two aversive-response associated downstream VTA targets.
But nine cells expressing simultaneously CTB-594 and CTB-488 were found, demonstrating that VTA non-glutamatergic neurons can send collaterals to NAc and LHb.
Finally, to check and detail the VTA VGluT2+ network, we mapped the structures projecting to VTA VGluT2+ neurons using Cre-dependent monosynaptic retrograde tracing. VGluT2-ires-Cre transgenic mice received AAV-CAG-DIO-histo-TVA-GFP (AAV2/9) and AAV-CAG-DIO-RG (AAV2/9) virus injection into VTA. Three weeks later, VTA was infected with RV-EvnA-DsRed (EnvA-pseudotyped, G-deleted and DsRed-expressing rabies virus) using the same coordinates (Sup. Fig. 5.A). Mice were sacrificed one week after this second injection and injection sites were verified as VTA (Sup. Fig. 5.B). Neurons projecting to VTA VGluT2+ neurons were defined as expressing red retrogradely label virus only (Sup Together, these data show that, on one side, a majority of projections to VTA VGluT2 arise from DRN, LH and MHb. On the other side, VTA VGluT2+ neurons project to NAc and LHb, which represents only a minority of VTA projections to these structures. While VTA send collaterals to NAc and LHb, they do not originate from VTA VGluT2+ neurons. This may indicate that VTA-NAc and VTA-LHb VGluT2 neurons belong to two segregated networks.

Summary of Results
We used fiber photometry to investigate how the Ventral Tegmental Area glutamatergic neuronal population was recruited by unconditioned and conditioned aversive and rewarding stimulation. We demonstrated that VTA VGluT2+ population was activated by both aversive and rewarding unconditioned stimulation, with a response amplitude remaining stable across trials. During the conditioning protocol, CS-evoked responses gradually increased over trials, briefly for aversive conditioning and in a sustained manner for rewarding conditioning; in parallel, US-evoked activity remained stable. During a retrieval test, CS-evoked and expected-US neural activities remained strong only for the aversive conditioning protocol, but not for the rewarding protocol. This suggests that aversive and rewarding conditioning signals are integrated by VTA VGluT2+ neurons through different mechanisms. Finally, to help better characterize VTA VGluT2+ neurons based on their connectivity pattern, we injected a CTB retrograde tracer in LHb and NAc nuclei, which revealed that only VTA non-glutamatergic neurons send collaterals to NAc and LHb. In parallel, by injecting rabies retrograde tracer in VGluT2-cre animals, we identified that VTA VGluT2+ neurons received inputs from variety of brain structures, with especially strong inputs from DRN, LH and MHb.

Fiber photometry and neuronal response
We used fiber photometry to perform a systematic exploration of VTA VGluT2+ neurons at the population level. This method allowed us to selectively record the activity of VTA glutamatergic neurons by injecting a cre-dependent GCaMP6s virus in VGluT2-cre mice.
Photometry is a recording method used to index synchronous neuronal activity at the Consequently, we cannot rule out that part of the signal we recorded may originate from terminals of locally infected VGluT2 VTA neurons, forming synapses with neighboring dopamine or GABA neurons at the microcircuit level. But such local interactions may represent only a small proportion of the signal detected and their contributions are negligible. Finally, unlike single neuronal recordings, a strong variation in calcium signal may be not only indicative of the activity of a population but can also be interpreted as a function of number of active neurons. An increase of fluorescence may mean that more neurons are recruited by a neuronal process.
But VTA VGluT2+ being a heterogenous population, use of both single neuron and population recording have disadvantages. Looking ahead, it will become very useful to combine photometry with technologies such as electrophysiology (Kim, 2017) to allow simultaneous investigation of single unit and population responses. This strategy could become a key method used to understand neural network function.

VTA glutamatergic neurons response to aversive and rewarding conditioning
This study focused on the population response of VTA VGluT2+ neurons during Our data also complements another recent electrophysiological study showing that VTA VGluT2+ neurons are sensitive to both aversive and rewarding stimulation (Root, Estrin, and . This provided a precise characterization of individual VGluT2 neurons based on their individual response pattern: they revealed that most VGluT2 neurons increased firing rate during aversive stimulation, and decreased during rewarding stimulation (Root, Estrin, and . In addition, they also showed that the majority of VTA VGluT2+ neurons decreased their activity during reward, and only a small fraction was activated by both reward and aversion. Our results are in accordance with and support their findings that VTA VGluT2+ neurons increase activity during unconditioned aversive stimulation. Another of their findings, that most of VTA VGluT2+ neurons decreased activity during reward, appears at odds with our results. However, this can perhaps be explained by the fact that the GCaMP6s signal is mainly correlated with neuronal activity, and it is virtually insensitive to inhibition (Chen et al., 2013). Consequently, the increase of calcium signal we recorded during reward was likely driven mainly by the small VTA VGluT2+ subpopulation described as responsive to both aversion and reward (Root, Estrin, and , and thus may not reflect the other subpopulation that is inhibited by reward. However, further investigation combining population and single neuron recording is required to test this hypothesis. This study next asked, for the first time, what the VTA VGluT2+ neuronal response to aversive conditioning and its retrieval are. We demonstrated that CS-evoked population response gradually increased across trials, whilst in parallel, the US-evoked response remained stable. We also demonstrated that VTA VGluT2+ neurons strongly respond to CS and expected-US 24 h after aversive conditioning. Some of these VTA VGluT2+ features, in particular, the gradual increased response to CS over trials, and response to an expected stimulation, may resemble VTA dopamine neurons (Nakahara et al., 2004;Satoh et al., 2003;Okihide, 2009;Nasser et al. 2017). One possible explanation our results arises from the finding that a subpopulation of VTA glutamtergic neurons, VGluT2 neurons, can coexpress VGluT2 and TH (Yamaguchi et al., 2011;Morales and Margolis, 2017). We cannot exclude that our results for the CSevoked response is due to VGluT2 neurons expressing TH. Another non-exclusive explanation may be that DA and VGluT2 neurons form connections at the microcircuit level (Dobi et al., 2010), which could modulate the response of VGluT2 neurons during CS and US pairing, especially modulating the VGluT2 population response during CS.
Finally, CS preceding-reward or CS preceding-aversive responses are slightly different, the former being sustained and the later brief, suggesting that each of these CS-evoked responses is integrated by a different subpopulation across trials. Supporting this idea, and contrary to the aversive experiment result (Fig. 2.I), the retrieval-evoked response to the CS, following rewarding conditioning, remained extremely weak, if not absent (Fig. 4.H).
This suggest that VGluT2 neurons responding to reward and aversive conditioning may belong to segregated subpopulations, probably in part corresponding to the different types of VTA VGluT2+ neurons recently characterized (Root, Estrin, and Morales, 2018). These divergences may be in part explained by specific connectivities of VTA VGluT2+ neurons sensitive to reward and aversion, in particular at the microcircuit level where interactions with dopamine and GABA neurons are known to exist (Dobi et al., 2010). For example, we can hypothesize than local VTA VGluT2+ neurons encode rewarding functions, while VTA VGluT2+ projections encode aversive ones. Future experiment should compare the functional and response profiles of these VTA VGluT2+ neurons that have diverging connectivity patterns.
Together, we showed that across trials, more and more VTA VGluT2+ neurons are similarly recruited during rewarding and aversive unconditioned stimulation, but differences emerge during conditioning. In particular, retrieval responses diverged, suggesting that neurons responding to reward and aversion conditioning may belong to different subpopulations.
Investigating the response of VTA VGluT2+ neurons during conditioning could help understand VTA function, including mechanisms that sustains learning and expectation in DA and GABA neurons.

Understanding VTA VGluT2+ neurons based on their network
We used CTB retrograde tracing to investigate VTA VGluT2+ projections to NAc and LH and found that VTA VGluT2+ represent a minority of projections to these structures. Of particular importance, we showed that, whilst VTA sends collaterals to NAc and LHb, these collaterals do not arise from VTA glutamatergic populations. Knowing that both VTA VGluT2+to-NAc and VTA VGluT2+ -to-LHb pathways are known to serve aversive function , our results raise the question of their individual functional characteristics. In particular, it would be important to separately record VTA VGluT2+ -to-NAc population response and VTA VGluT2+ -to-LHb, for example, by using fiber photometry at the terminal level, to compare their activity patterns during aversive stimulation.
We next used RV tracing to map the inputs of VTA VGluT2+ neurons and observed that particularly strong projections to VTA glutamatergic neurons were coming from DRN, LH and MHb. Our data are consistent with previous studies Faget et al., 2016), and confirm that structures such as LHb or NAc also sends projections specifically to VTA VGluT2+ populations, which may supply feedback that promotes aversive behaviors. It is known that DRN and PAG send projections to VTA-DA and VTA-GABA neurons linked to aversive and rewarding behaviors Ntamati, Creed, and Luscher, 2017;Ntamati et al., 2018); however, we observed that these structures also send parallel projections to VTA VGluT2+ , whose function remains unknown.
The diversity of VTA VGluT2+ neurons inputs and outputs support the idea that VTA VGluT2+ function is not only based on their molecular background, but also on the network the belong to . For example, functions and response patterns of dopamine neurons are highly heterogeneous and could depend of their specific projecting pattern Lammel et al., 2011;Lammel, Lim, and Malenka, 2014). In particular, knowing VTA VGluT2+ neurons stimulation can be either aversive  or rewarding , we can posit that VTA VGluT2+ projections promote aversion, while local VTA VGluT2+ neurons promote reward, likely via neighboring connections to DA and GABA neurons. An interesting future direction would be to specifically target these potential subpopulations based on their connection patterns at the circuit or microcircuit level, to investigate and systematically compare their activity profile and their molecular background.
In summary, we used fiber photometry to demonstrate that VTA VGluT2+ neuronal population response to aversive and rewarding conditioning are divergent, especially during retrieval of conditioning. This suggests that VTA VGluT2+ populations responding to reward or aversive conditioning may belong to different subpopulations. In the future, investigating VTA VGluT2+ neurons based on their local or long-range connectivity pattern may be important to better understand VTA function. In particular, deciphering the function of VTA VGluT2+ at the microcircuit level would shed new light on our understanding of how local VTA DA and GABA neurons process reward and aversive conditioning.  Time (