Experimental procedure

Two experimental groups underwent 3 sessions of aversive conditioning with 2 out of the 8 sets of facial stimuli and an olfactory context (trigeminal, non-trigeminal, control, Figure 50A). The facial stimuli and the odorants used were different for each session. All stimuli and odor presentations were carefully counterbalanced to avoid any presentation order confound. Four faces in total were used in each session: 1 female and 1 male identity, each one with two levels of anger. Male and female identities were alternatively used as CS++/CS+ to avoid any gender bias. The context was always pleasant for Group 1 (n=13) and always unpleasant for Group 2 (n=14). Each session was separated from the other by at least a day and maximum 10 days. Participants came at the same time of the day for the 3 sessions, they were asked not to eat nor drink anything 30 min before the experiment. The experiment was performed in a well-ventilated room to ensure olfactory neutrality.

161 The participants sat on an adjustable chair, facing the stimulus presentation screen, with their dominant hand on the response keyboard. Their head was maintained in a fixed position with the head support of the eyetracking system. An odor was presented as a context with a felt tip pen held 2 cm under the participants’ nose for the whole duration of the session, during the 3 phases of conditioning (i.e., habituation, conditioning and extinction). Participants were instructed to breathe normally, and they were placed under an air aspirating device in order to create an air current of 1 m³.min^-1 from the pen towards the face. This was done to maximize the amount of odorant directed to the nose, and minimize the contamination in the experimental room.

As presented in Figure 50A, each of the 4 facial stimuli was presented (Matlab toolbox, http://www.mathworks.com and Cogent2000, http://www.vislab.ucl.ac.uk/Cogent/) 32 times (4, 20 and 8 times during habituation, acquisition and extinction respectively) in a counterbalanced order for 2 seconds on a computer screen (800 x 600 pixel screen at a 57 cm distance). After each stimulus presentation, the participants were asked to rate the level of anger of the faces (Figure 50B, task). One of the sets (conditioned stimuli CS+/CS++) was reinforced at 50% during the acquisition phase with a 100dB, 200ms white noise, while the second set was unreinforced (CS-/CS-). Both sets were alternatively assigned to male or female faces, in a counterbalanced order across sessions. The whole experiment lasted about 18 minutes, and was kept short to minimize habituation effects.

Figure 49. Experimental design.

(A) Sequential presentation of trials. (B) Individual trial. CS = Conditioned Stimulus, US = Unconditioned Stimulus, w/ = with, TG+ = trigeminal, TG- = low trigeminal.

162 4.1.4 Subjective ratings

After each face presentation, participants judged the level of anger by rating the odors on a continuous scale presented on the screen, moving a cursor on a horizontal scale from 1 to 10 (“neutral” to “very angry”) with the response keyboard. Additionally, in order to assess the homogeneity odor perception across the experiment, they also had to judge the odor context on pleasantness (“very unpleasant” to “pleasant”), familiarity (“not familiar at all” to “very familiar”) and intensity (“not intense at all” to “very intense”) scales from 1 to 10 before and after the aversive conditioning, with the same system.

4.1.5 Physiological recordings

4.1.5.1 Heart rate, electrodermal activity, respiratory activity

Physiological signals were assessed with the Biopac Systems (Santa Barbara, CA) with separate settings for the electrocardiogram, electrodermal activity, and respiratory activities. Signals were transferred from the experimental room to the MP150 Acquisition Unit (16 bit A/D conversion) in an adjacent room and stored on computer hard disk (sampling rate 200 Hz). Respiratory activity was assessed by placing on the participant two respiration belts that measured abdominal and thoracic expansion and contraction. Electrodermal activity was recorded (high-pass filter: 0.025 Hz) by the constant voltage method (0.5 V). Beckman Ag–AgCl electrodes (8-mm diameter active area) filled with a skin conductance paste (Biopac) were attached to the palmar side of the middle phalanges of the second and third fingers of the participants’ nondominant hand. Heart rate was assessed by fixing Biopac pregelled disposable electrodes under the participants’ left and right wrists and on the left ankle. The signal was amplified by 1000 and low-pass filtered (30 Hz). Electrocardiographic R waves were detected offline, and intervals between heartbeats were converted into heart rate, expressed in beats per minute (BPM).

4.1.5.2 Pupil diameter

Pupil diameter was recorded with an iView X HI-SPEED Eyetracker (SensoMotoric Instruments GmbH, Germany) at a 240Hz sampling rate. Eye movements’ calibration was performed on a 800 x 600 pixel screen placed at a 57 cm distance from the participant, using a picture containing 13 dots located at known and predetermined positioned

4.1.6 Data Analyses

4.1.6.1 Behavioral data and reaction times

We assessed the efficiency of aversive conditioning by examining the anger level ratings and the reaction times for each condition. Reaction times were defined as the first key press after the face

163 presentation, in order to rate the stimulus. They were corrected for outliers, by suppressing any reaction time below 200ms, or not comprised in a +/- 2 standard deviation interval from the mean.

4.1.6.2 Electrodermal activity, Heart and Respiratory rate

Respiratory parameter: Respiratory rates and amplitude were calculated offline from the abdominal signal when possible, as the intervals between inspiration peaks and the maximal voltage amplitude at each inspiration peaks respectively.

For electrodermal activity, heart rate and pupil diameter, the mean signal recorded during the 1s before face presentation served as baseline for correcting the signal recorded after the stimulus onset period.

Electrodermal activity: Specific skin conductance responses (SCRs) to odors were measured in microSiemens and analyzed offline. They were scored as maximum peak changes in the phasic component of conductance (>0.02 microSiemens) starting in the 1- to 4-s interval after the stimulus presentation (Filion, Dawson, Schell, & Hazlett, 1991). SCRs were square root transformed to normalize the data (Edelberg, 1972).

Heart rate: The averaged heart rate response locked to the fixation cross onset is presented in Figure 51. We obtained a biphasic variation, corresponding to what was observed for a respiration locked heart variation in Delplanque and colleagues (Delplanque et al., 2009). The heart rate minimum and maximum variations were averaged within two periods, from 0 to 4.5 s and from 3 to 6 s. Then, we expressed those scores as a percentage of the BPM of the baseline. Percentage scores were introduced to standardize the differing absolute BPM variations of individual participants and thus to enable comparison between individuals and groups.

Figure 50. Biphasic heart rate trace.

Averaged by 500 ms time bins, from CS-- stimuli during conditioning phase in control condition.

164 4.1.6.3 Pupil diameter

Measurements were separated into 12 seconds epochs (1s before stimulus onset, 11s after). The collected signal was corrected offline for artifacts (eyeblinks and loss of signal), and interpolated using in-house routine written in Matlab. Trials with more than 500 ms of missing data were excluded. The signal was then baseline corrected, downsampled to 120 Hz and low pass filtered at 10 Hz. For each session, only subjects with more than half of the trials left for every condition were included. The pupil diameter results were not analyzed statistically because of substantial loss of data and poor quality of recordings (60% and 43% of data losses in groups 1 and 2 respectively) due to experimental problems.

4.1.7 Statistical analyses

All statistical analyses were mixed model MANOVAs, with Context (Pleasant vs Unpleasant) as between subject factor (BSF), and varied within subject factors (WSF) depending on the measurement. All post-hoc comparisons (PHC) were performed with Tuckey HSD tests. To control for sampling biases between Group 1 and Group 2 and enable intergroup comparisons, control analysis were first systematically performed for all the measurements in the odor condition, since the no-odor was common to both groups. Except for no-odor ratings, the main analyses were then performed on data obtained only when an odor context was present, and the control condition was not included anymore for simplicity. The analyses and respective names are detailed below.

4.1.7.1 Odor ratings (Odor_M)

Odor ratings were submitted to separate to three mixed models MANOVAs – one for each dimension: Familiarity, Hedonicity and Intensity - with Context (BSF), and Moment (Before vs After the experiment) and Trigeminality (High trigeminal vs Low trigeminal vs non-odor control) as WSFs, to check for the stability of the olfactory perception before and after the aversive conditioning procedure, to assess the variation of perception between different odors (pleasant, unpleasant, trigeminal), and control no-odor perception between the two groups.

4.1.7.2 Respiratory activity (Resp_M)

Respiration rate and amplitude were both analyzed with a mixed model MANOVA each, with Context (BSF), and Phase (Habituation vs Acquisition vs Extinction) and Trigeminality (High vs Low) as WSF, to assess the effects of odors and of the global conditioning process on the respiration pattern (odor context condition). The corresponding control analyses were mixed model MANOVAs on no-odor conditions only (control condition), with Context (BSF), and Phase (Habituation vs Acquisition vs Extinction) as WSF.

165 4.1.7.3 Other measurements

4.1.7.3.1 Non odor controls (Control_M)

For all the other measurements (Anger level ratings, Reaction times, Skin conductance response and heart rate), control analysis were mixed model MANOVAs on no-odor conditions only (control condition), with Group (BSF), and Phase (Habituation vs Acquisition vs Extinction), Anger (High vs Low anger level of the facial stimulus) and conditioning Status (CS+ vs CS-) as WSFs.

4.1.7.3.2 Odor context measurements (General_M): baseline corrected scores

For all the other measurements (Anger level ratings, Reaction times, Skin conductance response and heart rate), to assess for general effects of the aversive conditioning procedure in an olfactory context, we calculated baseline corrected scores (BASc) by subtracting the baseline obtained during habituation to each condition. Thus, only acquisition and extinction phases were considered. We then performed mixed model MANOVAs with Context (BSF), and Phase (Acquisition vs Extinction), Anger (High vs Low), Trigeminality (High vs Low) and conditioning Status (CS+ vs CS-) as WSFs.

4.2 Results

4.2.1 Odor ratings 4.2.1.1 Hedonicity

Although Context x Trigeminality interaction was significant [Odor_M, F(2,48) = 55.29; p<0.001; η² = 0.70; Figure 52A], hedonicity ratings did not differ for control conditions (PHC: p = 0.73). As expected, PHCs showed that pleasant odors were rated as more pleasant (p<0.001) than the controls, as the controls, which in turn were more pleasant than unpleasant odors (p<0.001). Trigeminal unpleasant odor (isovaleric acid) was perceived as slightly less pleasant than its non-trigeminal counterpart (indole) (p = 0.03), but this effect was not present for pleasant odors. There were no effects or interactions related to Moment.

4.2.1.2 Familiarity

A triple Context x Moment x Trigeminality interaction was significant for familiarity ratings [Odor_M F(2,48) = 3.27; p = 0.047; η² = 0.01, Figure 52B]. Pleasant odors were perceived as more familiar than the controls (p<0.001). However, none of the PHCs revealed any moment-related differences, nor differences for control conditions (PHC: p = 0.98).

4.2.1.3 Intensity

The control (no odor) was perceived as less intense than odors [Odor_M, Main Trigeminality effect, F(2, 48) = 54.89; p<0.001; η² = 0.70; PHC p<0.001, Figure 52C]. There were no intensity differences

166 between control conditions, nor unpleasant and pleasant odors, nor low and high trigeminal [Odor_M, Hedonicity x Trigeminality interaction: non-significant (n.s.) F(2,48) = 0.36; p = 0.23; η² = 0.06]. There were no effects or interactions related to Moment, therefore, the conditioning procedure did not significantly impact odors’ perception.

Figure 51. Mean subjective ratings of (A) pleasantness, (B) familiarity and (C) intensity.

P = Pleasant odors, Group 1. U = Unpleasant odors, Group2. TG+ = Highly trigeminal odors. TG- = Low trigeminal odors. Vertical bars denote Standard Errors to the Mean.

4.2.2 Anger level ratings

Control condition. Control_M showed that anger ratings did not differ significantly between the two groups of participants in the control condition. There were no effects nor interactions related to Group, except for a Group x Phase x Anger x Status [Control_M, F(2,50) = 5.36; p = 0.008; η² = 0.18].

However, none of the PHC revealed a group related difference between the controls.

BASc. General_M analysis performed on Anger ratings BASc revealed a main context effect [General_M, F(1,25) = 9.04; p<0.01; η² = 0.27],with anger being rated higher compared to the habituation baseline in a pleasant context. Since the quadruple Context x Phase x Anger x conditioning Status was significant [General_M, F(1,25) = 7.88; p = 0.004; η² = 0.24], two secondary ANOVAs with Phase (Acquisition vs Extinction), Anger (High vs Low), Trigeminality (High vs Low) and conditioning Status (CS+ vs CS-) as WSFs were then performed separately for each context group.

They show that in a pleasant context, the high anger stimuli were not conditioned, and that they were rated as appearing less angry compared to their perception of anger in conditioned stimuli appeared to be inverted [Conditioning x Anger interaction F(1,12) = 12.81; 0.004; η² = 0.52; Figure 53A]. When an unpleasant context was present, it rather modulated extinction: high anger stimuli did not undergo extinction anymore [Phase x conditioning Status x Anger interaction F(1,13) = 9.68; p

= 0.008; η² = 0.43; Figure 53B].

167

Figure 52. Mean Baseline Corrected Scores (BASc) for anger level ratings as a function of the olfactory context.

A: General_M, showing a main effect pf the context on anger ratings B. Conditioning status x Anger interaction in pleasant odor context, showing an effect of the conditioning status on anger ratings, on the high anger condition only (CS—and CS++). C:) Phase x . Conditioning status x Anger interaction in unpleasant odor context, showing impaired extinction for the high conditioned stimulus (CS++). PLEA = Pleasant odors, Group 1. UNPL = Unpleasant odors, Group2. Cond = Conditioning phase. Ext = Extinction phase. CS+,CS++

= conditioned stimuli, CS-, CS-- = non conditioned stimuli, low and high anger respectively. Vertical bars denote Standard Errors to the Mean. Significance levels for post-hoc comparisons: ‘ns’: not significant, p>0.05. ‘*’:p ≤ 0.05; ‘**’: p ≤ 0.01; ‘***’: p ≤ 0.001

4.2.3 Reaction times

Control condition. Reaction times did not differ significantly between the two groups of participants in the control condition [Control_M: Group x Phase interaction: n.s., F(2,50) = 1.87; p = 0.16; η² = 0.07], but were rather influenced by the Phase [Control_M, Main phase effect, F(2,50) = 8.66;

p<0.001; η² = 0.26], as participants habituated to the task and performed it faster along the experiment.

BASc. General_M analysis performed on Reaction Times BASc revealed a Context x conditioning Status interaction [General_M, F(1,25) = 6.48; p = 0.02; η² = 0.21; Figure 54A], showing that participants seemed to react slower for rating conditioned stimuli in an unpleasant context.

Moreover, participants were faster in the conditioning compared to the extinction phase, and for rating conditioned versus non conditioned stimuli during the conditioning phase [Phase x conditioning Status interaction, F(1,25) = 8.84 p = 0.006; η² = 0.26; Figure 54B].

168

Figure 53. Mean BASc of reaction times as a function of the olfactory context in ms.

A: Context x conditioning Status interaction, showing differential reaction times for conditioned stimuli in according to the odor context B. Phase x conditioning Status interaction, showing faster reaction times for extinction and conditioned stimuli. PLEA = Pleasant odors, Group 1. UNPL = Unpleasant odors, Group2. Cond = Conditioning phase. Ext = Extinction phase. CS+ = conditioned stimuli, CS- = non conditioned stimuli. Vertical bars denote Standard Errors to the Mean. Significance levels for post-hoc comparisons: ‘ns’: not significant, p>0.05. ‘*’:p ≤ 0.05; ‘**’: p ≤ 0.01.

4.2.4 Skin conductance responses (SCRs)

Control condition. SCRs did not differ significantly between the two groups of participants in the control condition [Control_M, Group x Phase interaction F(2,48) = 0.07; p = 0.93; η²<0.01]. None of the other effects were significant, except for a Context x Anger interaction [Control_M, F(1, 24) = 4.44; p = 0.046; η² = 0.16]. However none of the related PHCs were significant.

BASc. The General_M analysis performed on SCRs revealed both a double Trigeminality x conditioning Status interaction [General_M, F(1,24) = 6.42; p = 0.02; η² = 0.21, Figure 55A], with conditioning appearing to be differential according to trigeminality. A quadruple Context x Phase x Trigeminality x conditioning Status interaction [General_M, F(1,24) = 6.68; p = 0.02; η² = 0.22] was also revealed. In order to examine these effects in more detail, we performed a secondary mixed model MANOVA with Context, Phase, Anger and conditioning Status (CS+ vs CS-) as WSFs on SCR events measured in a highly trigeminal context only. Conditioning seemed to produce greater increases in a pleasant trigeminal context, while the increase was delayed in an unpleasant trigeminal context [Context x Phase x conditioning Status interaction, F(1,24) = 7.86; p<0.01; η² = 0.25; Figure 55B], suggesting that the extinction could have been weakened in the latter case.

169

Figure 54. Mean BASc of Electrodermal response amplitude as a function of the olfactory context in micro Siemens (µS).

A.Trigeminality x conditioning Status interaction, showing differential condtioned SCR according to trigeminality. B. Context x Phase x conditioning Status interaction in a highly trigeminal context, showing delayed conditioned SCR in extinction in an unpleasant context only. P = Pleasant odors, Group 1. U = Unpleasant odors, Group2. TG+ = Highly trigeminal odors. TG- = Low trigeminal odors. Cond = Conditioning phase. Ext = Extinction phase. CS+ = conditioned stimuli, CS- = non conditioned stimuli. Vertical bars denote Standard Errors to the Mean. Significance levels for post-hoc comparisons: ‘ns’: not significant, p>0.05. ‘*’:p ≤ 0.05; ‘**’: p ≤ 0.01;

4.2.5 Heart rate

Control condition. Mean heart rate minima and maxima did not differ significantly between the two groups of participants in the control condition [Control_M, Group x Phase interaction: n.s.; F(1,42) = 2.65; p = 0.08; η² = 0.11 and F(1,42) = 0.02; p = 0.98; η²<0.001 in minima and maxima respectively.].

None of the interactions involving Group were significant, except for a triple Group x Phase x Anger interaction [Control_M, F(2,42) = 5.08; p = 0.01; η² = 0.19] in maxima. None of the related PHCs were significant.

170 BASc. General_M showed a Phase x conditioning Status x Anger interaction in heart rate minima BASc [General_M, F(1,21) = 5.89; p = 0.02; η2 = 0.22]. This interaction was n.s. in maxima [General_M, F(1,21) = 0.45; p = 0.51; η2 = 0.02] and although three other double interactions were found [Context x Trigeminality, Context x Anger and Trigeminality x Anger, F(1,21)>4.33; p<0.04;

η2>0.17], there were no other significant conditioning Status related interactions. Conditioned stimuli induced heart rate deceleration (minima), in particular high anger stimuli during extinction, compared to non conditioned stimuli (PHC: p<0.05), independently of the valence or the trigeminality of the context odor.

4.2.6 Respiration Rate

Control condition. There was no significant difference in the respiratory rate in the control condition only [Resp_M, Group x Phase interaction: n.s.; F(2,42) = 0.08; p = 0.92; η²<0.001].

Odor context condition. Participants tended to breathe quicker initially when exposed to unpleasant smells, and this difference faded out with conditioning [Resp_M, marginal Group x Phase interaction F(2,42) = 2,59; p = 0.09; η² = 0.11]. There also was an additional tendency to breathe quicker in the presence of TG- odors [Resp_M, marginal main Trigeminality effect, F(1,21) = 3.46; p = 0.08; η² = 0.14].

4.2.7 Respiration Amplitude

Control condition. There was no difference in breathing amplitude between groups in the control condition [Resp_M, Group x Phase interaction: n.s.; F(2, 42) = 0.98; p = 0.38; η² = 0.045].

Odor context condition. Conversely to the respiration rate pattern, participants tended to breath deeper when exposed to pleasant smells [Resp_M, marginal main Hedonicity effect, F(1,21) = 4,12; p

= 0.06; η² = 0.16], but breathing amplitude was not affected by the trigeminality of the odor [Resp_M, main Trigeminal effect n.s. F(1,21) = 0.004; p = 0.94; η²<0.01].

4.3 Discussion

4.3.1 Results summary

This study provided formal evidence for the first time that odors as context can influence aversive conditioning, both behaviorally and physiologically. We showed that conditioning and extinction are affected by odor context and modulated by the strength of the emotion carried by the conditioned stimulus, namely the anger level of the faces. Pleasant odors as a context generally enhanced anger evaluations during acquisition and extinction compared to habituation, and appeared to induce some facilitation of conditioning during the acquisition phase at the physiological level (SCRs) only, with

171 effects driven by odor trigeminality. However no association specific effects were observed behaviorally. Conversely, in an unpleasant context, no modulatory changes were observed during acquisition, but rather during extinction: there was an important subsequent effect of malodors in conditioning for high anger stimuli, both behaviorally (anger ratings) and physiologically (SCRs, unpleasant trigeminal odors).

The olfactory context itself did not appear to be impacted by the conditioning process. Taken together, these data suggest that if pleasant odors could function as incongruent positive context towards the conditioned, aversive stimuli at the implicit level, malodors appear to have a delayed, yet more important influence at both explicit and implicit levels.

4.3.2 Relationship to previous work

This study used odor as a contextual influence during all the phases of conditioning, including habituation and extinction. It was therefore essential that odor perception was kept unchanged and unaffected by aversive conditioning. In fact, olfactory characteristics remained the same and there were no differences in familiarity, intensity and hedonicity before and after the process, guaranteeing the homogeneity of the affective context. All the odors were of similar intensity, and, as we chose highly polarized / contrasted odors, pleasant and unpleasant odors were perceived as expected, enabling a global consensus among participants, despite the strong interindividual variation of odor perception (Ferdenzi, Roberts, et al., 2013; Griep et al., 1997; Poncelet et al., 2010).

Moreover, familiarity followed pleasantness, as unpleasant odors were perceived as less familiar (Delplanque et al., 2008; Herz, 2003). Although the trigeminal, unpleasant isovaleric acid was rated as more unpleasant than indole (non-trigeminal), we cannot attribute this valence difference to trigeminality only, as this effect was not present for pleasant odors.

4.3.3 Partial invalidity of affective congruence

At first sight, the mechanism by which odor exerts its influence does not seem to be entirely of affective congruence nature, as this would have resulted in an enhancement of the fear conditioned response (FCR) in an aversive / unpleasant context during the acquisition phase. However analysis revealed a residual effect during extinction in unpleasant context, for high angry faces at the behavioral level and at the physiological level, which could also indicate a masking of the conditioning effect during acquisition, or a resistance to extinction. The later can be encountered in anxiety disorders (Waters, Henry, & Neumann, 2009) due to a greater reactivity to the US because of an elevated anxious state. More generally, according to Mowrer and Jones’s theory of the Discrimination Hypothesis (Mowrer & Jones, 1945), the ease of extinction is inversely related to the similarity between acquisition and extinction phases. In our experiment, when the conditioning is

172 performed with an unpleasant context in the background, the difference between acquisition and extinction solely relies in the removal of the US. The subsequent persistence of a general aversiveness conveyed both by the context odor and the high anger level contributes to the affective

172 performed with an unpleasant context in the background, the difference between acquisition and extinction solely relies in the removal of the US. The subsequent persistence of a general aversiveness conveyed both by the context odor and the high anger level contributes to the affective