High Trait Anxiety Blocks Olfactory Plasticity Induced by Aversive Learning

Abstract

Aversive learning normally induces alterations in sensory function as the brain’s sensory systems are tuned to optimize detection and discrimination of threat-predictive stimuli. Anxiety disorders can disrupt behavioral discrimination between threat-predictive and neutral stimuli, resulting in overgeneralization of negative affective responses to non-threatening situations. We thus hypothesized that anxiety could disrupt learning-induced improvement in sensory discrimination. We tested perceptual discrimination between similar odorants before and after discriminative aversive conditioning. Participants exhibiting normal levels of trait anxiety developed a larger skin conductance response (SCR) to the shock-predictive odorant and substantial improvement in their perceptual discrimination between the two odors. Repeated exposure to the odors without shock partially extinguished the SCRs but the perceptual effect persisted. By contrast, participants with high levels of trait anxiety developed comparably sized SCRs to both odors and displayed no perceptual improvement. Learning-induced perceptual plasticity can thus be impaired in people with high levels of trait anxiety.

1. Introduction

Sensory processing normally changes over time as a function of experience and environment (Bieszczad, Miasnikov, & Weinberger, 2013; Headley & Weinberger, 2015; Kass, Rosenthal, Pottackal, & McGann, 2013; Li, Luxenberg, Parrish, & Gottfried, 2006; McGann, 2015; Weinberger, 2015). This plasticity is critical because objects in the world are experienced in different positions and orientations and under different illuminants and need to be recognized as the same object despite modest sensory differences. However, for some objects small differences in their sensory features can indicate entirely different outcomes, such as a traffic light with different colors illuminated or a word spoken with different inflections. Navigating the world thus requires learning to lump together (i.e. generalize across) similar stimulus experiences that mean the same thing and to separate out (i.e. discriminate among) similar stimulus experiences that have been learned to predict different outcomes.

Early 20^th century learning theories by Lashley, Hull, and Spence emphasized the importance of discrimination and generalization in the linkage of sensory stimuli to behavior, as demonstrated experimentally using classical or operant conditioning. For instance, a pigeon trained to peck at a key illuminated with one wavelength of light would also tend to peck when the key was illuminated with lights of other wavelengths, different luminance, or other variables. Training with different experimental paradigms could teach subjects to respond broadly to a wide range of related stimuli (generalization behavior) across multiple stimulus dimensions or to become highly selective about which stimulus they responded to (discrimination behavior). Naturally this behavioral discrimination depends on the ability of the subject to tell the stimuli apart in the first place (Klopfer, 1968), but in the behaviorist era sensory capabilities were largely treated merely as practical limitations on discriminative behavior. Later developments in neuroscience, artificial intelligence, and behavior analysis fostered the development of a cognitive model of mind in which the perception of a stimulus could be treated separately from behavioral responses to it (Pinker, 2021). Cognitive models emphasized the incorporation of learned information into the mental process of stimulus perception and extended the classical framing of discrimination and generalization of stimulus-driven behavior to encompass effects of attention (Broadbent, 1961; Mackintosh, 1975), perceptual decision-making (Green & Swets, 1966), and contextual framing (Holland, 1992; Tversky & Kahneman, 1985) on stimulus perception. The perceived stimulus then informs organism-level decision-making about behavioral responding. From this perspective, it is natural to view stimulus perception as dynamic and informed by prior experience, consistent with classical findings about the impressive abilities of highly trained sensory experts (James, 1890; Poggio, Fahle, & Edelman, 1992) and potentially explanatory of cases where patients with disordered cognition exhibit atypical perception (Braff et al., 1978; Braff, Swerdlow, & Geyer, 1999; Schneider et al., 2002)

One weakness of this cognitive approach is that it often partitions the perceptual processing of an incoming stimulus set (e.g. sensory decision-making about whether two similar stimuli should be perceived as the same or different) into a mental module separate from and upstream of behavioral and emotional responses to the stimuli. However, a growing body of neurobiological evidence demonstrates that even the very earliest sensory processing in the brain can reflect the ecological importance of a stimulus (see McGann 2015 for review). For instance, we have previously reported that discriminative fear conditioning in which a mouse learns that one particular odor predicts an aversive electric shock can selectively increase neurotransmitter release from the olfactory nerve in response to that odor (Kass, Rosenthal et al. 2013). Changes in early sensory processing happen even when the outcome-predictive sensory stimuli are readily detectable and quite different from each other, yet the change in neurosensory processing of odors can predict how individual mice generalize or discriminate their fear among odors (Kass & McGann, 2017). This suggests that the affective content of the learned odor-shock relationship is not merely “tuning” the sensory system to become better at perceptually discriminating between odors but may also be intimately linked to the emotional response to the threat-predictive odor (Moberly et al., 2018). Biological cross-talk between the emotional response to a threat-predictive odor and the early sensory processing of that odor suggests that altered emotional responses to sensory stimuli, such as in anxiety, might disrupt the normal alterations in stimulus perception that occur during learning.

Anxiety stands out as a potential disruptor of sensory changes during emotional learning because it reflects a failure of discrimination-generalization decision-making. It is often conceptualized as an inappropriate generalization of a feeling of threat to stimuli or situations that physically resemble or conceptually relate to threatening circumstances but are not themselves threatening (Dunsmoor & Paz, 2015; Dymond, Dunsmoor, Vervliet, Roche, & Hermans, 2015; Lissek et al., 2005; van Meurs, Wiggert, Wicker, & Lissek, 2014). In laboratory assessments, patients with anxiety disorders or high levels of trait anxiety tend to generalize their perceptions of threat across a larger range of potentially threat-predictive stimuli than control subjects (Dunsmoor, Ahs, & LaBar, 2011; Gazendam, Kamphuis, & Kindt, 2013; Haddad, Pritchett, Lissek, & Lau, 2012; Lissek et al., 2014; Morey et al., 2015; Sep, Steenmeijer, & Kennis, 2019), with similar outcomes in the clinic (Lissek & Grillon, 2010; Wong & Lovibond, 2018). Though the neural underpinnings of anxiety remain poorly understood, brain regions like the amygdala and locus coeruleus are increasingly understood to play a role not only in anxiety and fear learning but also in sensory processing (Fast & McGann, 2017; Giustino & Maren, 2018; McCall et al., 2017; Morey et al., 2015; Morris, McCall, Charney, & Murrough, 2020).

Several laboratories have established that human subjects can rapidly improve their perceptual discrimination and detection of odor stimuli following odor-cued aversive conditioning (Åhs, Miller, Gordon, & Lundström, 2013; Li, Howard, Parrish, & Gottfried, 2008). Parallel data in the mouse has demonstrated that odor-cued aversive learning alters neural representations of the threat-predictive odor in the early olfactory system (Kass & McGann, 2017; Kass et al., 2013), that alterations in these representations can improve odor discrimination (Kass, Guang, Moberly, & McGann, 2016), and that early olfactory processing is shaped by both amygdala and locus coeruleus (Fast & McGann, 2017). We thus asked whether the aversive conditioning-induced olfactory plasticity that typically occurs in humans would be disrupted in people with high levels of anxiety. Specifically, we hypothesized that when faced with two very similar odors that differentially predict an aversive shock, highly anxious subjects would both generalize their learned skin conductance response to both odors (because anxiety disrupts behavioral and emotional generalization) and fail to improve in discriminating between those odors in purely sensory testing. By contrast, we expected typical subjects would exhibit more selective skin conductive responses to the shock-predictive odor over the control odor and would improve their sensory discrimination performance between the two odors.

Fear learning-induced sensory improvements would be of obvious benefit to the organism by narrowing the scope of fear to the most relevant circumstances. However, it is not clear whether these improvements would be expected to be reduced during extinction learning, when the organism learns that the stimulus no longer predicts a negative outcome. Should the sensory improvements persist after extinction then the original learning experience would have lasting effects on the organism as it navigates it sensory world. This question may be especially important in the context of anxiety, not only because we hypothesize that high anxiety people exhibit impaired learning-induced sensory plasticity but also because pathological anxiety is frequently treated using a form of extinction training known as extinction training. We thus tested whether in typical subjects who improved at discriminating between similar odors following aversive conditioning if further presentations of the odors without shocks would reduce the conditioned physiological response to the odors but leave the sensory improvement intact. If true, this would suggest that sensory plasticity during learning normally makes an ongoing contribution to the perception of potential threats in the world, and that disruption of this process by anxiety would be expected to have lasting effects.

2. Materials and methods

2.1. Participants

Participants consisted of 210 students (119 female) with a mean age of 21.4 (SD=4.2) from Rutgers University in New Brunswick. Students were recruited either via the Rutgers human subject pool system or via flyers placed around campus, prior to COVID-19 outbreak. Participants recruited via Rutgers human subject pool received Research Participation Units (RPU’s) as part of their requirement for completing Intro to Psychology course. Individuals recruited via flyers received a small monetary compensation of $30. Participation was limited to non-pregnant, non-smokers and participants with no respiratory issues that could potentially affect olfactory processing during the experiment (e.g. significant nasal congestion, history of nasal surgery). Participants were asked to consume nothing but water and not to chew gum one hour in advance of participation. All participants were instructed to avoid wearing scented products the day of the experiment, and a small number of recruited subjects who were noticeably scented were excluded from the experiment at the time of data collection. All participants gave written informed consent and experiments were conducted in accordance with the Declaration of Helsinki on Medical Studies Involving Human Subjects and with protocols approved by the Rutgers Institutional Review Board.

Participants completed the State-Trait Anxiety Inventory (STAI) (Spielberger, Gorsuch, Lushene, Vagg, & Jacobs, 1983) which ranges from a score of 20 – 80, with higher scores representing higher anxiety level. Previous analysis of STAI scores using Angoff, Receiver Operating Characteristic, and Borderline methods has demonstrated that a trait anxiety score cutoff of 48 optimally classifies college students exhibiting pre-clinical levels of anxiety in college students (Sarkin & Gülleroğlu, 2019). Participants in this study (min STAI Trait = 20, max = 69) were thus dichotomized into High Trait Anxiety (HTA, N=49, mean Trait anxiety score M=54.13, SD=4.98) or Normal Trait Anxiety (NTA; N=159, mean Trait anxiety score: M=36.31, SD=6.68) based on an a priori STAI-T score of at least 48, which corresponded to the 80^th percentile. Dichotomization of data, even based on a priori criteria, can produce misleading outcomes if the underlying phenomenon is not in fact dichotomous (MacCallum, Zhang, Preacher, & Rucker, 2002). However, taxonometric analysis confirmed that at least within the limits of our dataset, a categorical model separating high and normal STAI Trait scores is a more likely fit to our data than a dimensional model treating trait anxiety as a continuum (Supplementary Materials, Fig. S3).

2.2. Olfactory Stimuli and Discrimination Testing

The aliphatic esters Hexanal (≥98% purity) and Heptanal (≥95% purity) were obtained from Sigma-Aldrich and stored capped with nitrogen. Hexanal and heptanal, which differ only by one carbon chain length, were selected because they are perceptually difficult to discriminate for humans (Laska & Teubner, 1999) and mice (Kass et al., 2016). Odorants were diluted 1:10,000 and 7.8:10,000, respectively, in mineral oil (Sigma Aldrich #8042–47-5) to adjust for their difference in vapor pressure. These odors are used by the food industry as flavoring agents and have been described as fruity (Burdock, 2016; Lewis, 2016). A pilot study using a 200 mm long visual analog scale (VAS) with “No odor perceived” and “Extremely Strong” at opposite ends determined the two odors at these concentrations were isointense (N = 15, Hexanal M = 129.93 mm, SD = .31, Heptanal, M = 130.87mm, SD = .37, paired-samples t-test t (14) = −.17, p = .864). The participants also reported, using a VAS with the same parameters but “Extremely Pleasant” and “Extremely Unpleasant” at opposite ends that these odors had an equivalent hedonic value (N = 15, Hexanal M = 82.07 mm, SD = .37, Heptanal, M = 76.73 mm, SD = .36, paired-samples t-test t (14) = .53, p = .61). Of note, many of the participants in our study, when asked to verbally characterize the odors, described the odors as reminding them of the odor of molding clay.

Odors were delivered binarally via a custom-built, computer-controlled olfactometer with a delivery cylindrical glass tube positioned 2 cm in front of the nose. The tube was approximately 5” long and 1” wide, and odor was delivered at a flow rate of 1.5 L/min in room air. A constant vacuum positioned concentrically around the odor delivery tube prevented odorant from reaching the subject between odor presentations. Shunting this vacuum resulted in rapid odor onsets and offsets. The olfactometer was calibrated daily prior to data collection each day via photoionization detector (ppbRae 3000, RaeSystems). Participants were positioned on an adjustable chin rest to ensure a constant distance between the odor delivery tube and the nose throughout the experiment. Odor presentations and visually delivered instructions were controlled by custom programs written in PsychoPy (Peirce, 2007).

Perceptual discrimination was tested using a three-alternative forced choice test (Laska & Teubner, 1999) consisting of six “triangles,” each composed of three sequential 6 sec odorant presentations, two of which were the same and one of which was different. The participant was asked to identify which odor was different from the other two via keypress on a computer keyboard. Perceptual discrimination between odorants was quantified as the percentage of correct answers on the 6-question triangle test (chance performance is 33%) both before and after conditioning. Subject responses were recorded via keyboard with keys labeled 1, 2 and 3 corresponding to the order of odorant presentations within each triangle. The order in which the odors were presented within each triangle was pseudo-randomized such that the participant was not presented the same arrangement of odor presentations more than once. The inter-stimulus interval (ISI) within triangles was 10 seconds between odor presentations and the inter-trial interval (ITI) between triangles was 30 seconds. Participants’ reaction time was recorded and used to exclude perceptual testing data from responses that were made implausibly rapidly or slowly. For all perceptual discrimination analyses, except for those involving the investigation of the effects of extinction and sham extinction on perceptual changes, we have pooled the data from the participants in the paired acquisition group with those who received paired acquisition and were tested after a time delay (sham extinction group) since there was no difference in perceptual discrimination improvements between the participants in these groups (Mann-Whitney; U=301.00, z=−1.09, p=.27).

2.3. Electrical Stimulation

During aversive conditioning participants received a total of eight 200 msec-long electrical stimulations delivered to the forearm (Phelps et al., 2004). Electrical stimulations were delivered via an SD9 stimulation unit from Natus Technologies (Phelps et al., 2004) through a two-point bar electrode positioned along the palmaris longus tendon. Microlyte gel was used to improve current stability. Shocks were delivered in 200 msec. trains of constant-voltage pulses at 50 Hz delivered at least one minute apart. Shock voltage was determined for each subject individually during initial setup by delivering calibration shocks that started at 20V and increased in 5V increments until the participant deemed the shock to be “uncomfortable but not painful” (maximum voltage selected was 80 V). The average voltage level across all participants was 40.83 volts (SD = 12.75). This produces a sensation like a static electric discharge occasionally experienced in daily life. Participants received no information regarding the odor-shock contingency or how frequently the shock would be delivered. All participants underwent this process of electrode placement and shock level selection, even those assigned to the “no shock” control group.

2.5. Skin Conductance Responses

Participants’ odor-evoked, anticipatory skin conductance response was measured via two 10mm Ag/AgCl electrodes placed on the middle phalange of the first and second finger of the subject’s non-dominant hand (Phelps, Delgado, Nearing, & LeDoux, 2004). A bead of microlyte gel was applied to the recess in the electrodes to improve signal stability. A constant, imperceptible 0.5 V signal was passed between the two electrodes and the conductance was measured using a Coulbourn, LabLinc V SCR module. SCR signals were digitized at a sampling rate of 100 kHz using a CED micro1401, recorded using Spike2 software, and analyzed using custom Matlab software.

Participants’ skin conductance response was measured throughout the experiment (Phelps et al., 2004). Anticipatory, odor-evoked SCR amplitudes were quantified by subtracting the average amplitude of a 4 sec. pre-stimulus time window (baseline) from the peak amplitude during the first 5.8 seconds of the 6 second odor presentation (the final 200 ms were excluded from all trials because a shock was concurrently presented on some trials). For all analyses of physiological learning (except those involving comparisons between extinction and sham extinction groups) we pooled the acquisition trials across groups that received pairings of odor and shock (paired acquisition, extinction and sham extinction).

2.6. Aversive Conditioning & Control Paradigms

Figure S1 displays the aversive conditioning trial structure for experimental and control paradigms. During odor-cued aversive conditioning participants in the odor-shock paired group received 16 trials, on half of which one of the test odorants (known as the CS+ stimulus) co-terminated with a mild electric shock (delay conditioning) while on the other half the other test odorant (the CS− stimulus) was presented without a shock. Delay conditioning paradigms such as this are ubiquitous in both animal fear conditioning (Jones, Choi, Davis, & Ressler, 2008; Kass et al., 2013; Sotres-Bayon, Bush, & LeDoux, 2007) and human aversive conditioning (Li et al., 2008; Lissek & Grillon, 2010; Phelps et al., 2004). Hexanal and heptanal served as CS+ and CS− stimuli, counterbalanced across subjects. The duration of the ITIs varied pseudo-randomly from 45–75 seconds (M=1 min) to prevent timing effects. To control for the effects of odor exposure, we included an odor only group where a smaller number of participants underwent the same paradigm as participants in the paired group, except no shocks were delivered. To control for any potential effects of electrical stimulation, we also included a shock only control group, who underwent the same paradigm as participants in the paired group but no odors were presented during the “conditioning” paradigm. To test whether the effects of aversive conditioning were reversible we included an extinction group, in which after aversive conditioning participants received an additional 16 trials of the CS+ and 16 trials of the CS-odors without any shocks. To control for the effects of additional passage of time in the extinction group, we also included a sham extinction group that after aversive conditioning did simple cross word puzzles for the same duration of time as the 32 trials of extinction. Lastly, to control for the effects of additional odor exposure in the extinction paradigm, we included an odor only group which received the same 32 trials of odors following aversive conditioning.

2.7. Data Quantification and Analysis

Two participants who were strongly scented upon arrival were excluded from the study. All SCR data was examined visually and participants who had obvious excessive artifacts (N = 24) or unmeasurable SCRs, defined as the absence of any SCR during the experiment time window (N = 5) or no response evoked by the shock US (≤0.01 μS) for at least 5 out of 8 shocks (N=3), were excluded from SCR analysis. One participant in the odor only control group was excluded from SCR analysis because his SCR scored as 4.40 standard deviations above the group mean for one odor but not the other. For the perceptual discrimination task, participants who scored 5/6 or 6/6 correctly at baseline (N=54) were not included in the analyses of perceptual discrimination to ensure exclusion of any participant who could reliably discriminate the odor prior to conditioning. Additionally, those subjects whose reaction time during the perceptual task was at least 2.5 standard deviations above or below the overall mean during baseline (5 subjects) or post training (5 subjects) were also excluded from perceptual discrimination analyses.

To quantify the difference in CS⁺-evoked and CS⁻evoked SCR after learning, we averaged the SCR evoked on the last three trials of CS⁺ and subtracted the average SCR from the last three trials of CS⁻. The averaging of the last three trials minimized the impact of random trial order effects. Analysis of averaged group-level “learning curves” is conceptually fraught (Gallistel, Fairhurst, & Balsam, 2004), especially when confounded by differences in conditioning trial sequence across participants, but Fig. S2 shows that both NTA (Fig. S2A) and HTA (Fig. S2B) participant groups had reached asymptotic conditioning by trial 6. This confirmed that analysis of the last three CS⁺ and CS⁻ trials would not be confounded by differences in learning rate across groups.

SCR and perceptual discrimination data distributions were not normally distributed, so nonparametric tests were used for all analyses that included these variables (Altman & Bland, 2009). Two-sample Kolmogorov-Smirnov tests were used for pairwise comparisons of data distributions to test the hypothesis that both datasets were sampled from the same underlying distribution without making assumptions about the distribution shape. Analogous comparisons between multiple groups were made using the Kruskal-Wallis test, a non-parametric test similar in function to the one-way ANOVA. The alpha value to reject the null hypothesis was p = 0.05, which was reduced to correct for the number of comparisons in each analysis using the Bonferroni correction. All data points used in analysis are depicted individually in the figures below using cumulative frequency histograms, which best illustrate empirical distribution shapes, and scatterplots, which best convey a visual impression of the number and density of the data points in each distribution. Statistical analysis was performed in SPSS and Origin Pro. For analyses of data with normal sampling distributions such Visual Analog Scale data, parametric tests were used and all alpha values were 0.05.

The mean late acquisition SCRs evoked by the CS+ and then by the CS− between experimental and control groups for NTA, using an Independent-Samples Kruskal-Wallis Test for each, the CS+ and the CS− evoked SCRs. Bonferroni adjusted pairwise comparisons followed all Kruskal-Wallis tests, to assess differences between groups with a significance value set at p < .05. The same analysis was conducted to assess the effects of HTA on physiological discrimination, comparing the NTA and HTA groups. The effects of aversive conditioning on anticipatory SCR discrimination was assessed between experimental and control NTA groups and between HTA and NTA paired acquisition groups via Kruskal-Wallis. Lastly, we compared the absolute SCRs evoked by the CS+ and the CS− within group for each of the experimental and control groups as well as HTA and NTA groups via Related-Samples Wilcoxon Signed Rank Test. For the effects of extinction training on physiology, we compared the mean late extinction SCRs via Related-+Samples Wilcoxon Signed Rank Test. Finally, we compared the effects of extinction training on the extinction group to the effect on physiology on the control, odor only group via Independent-Samples Mann-Whitney U.

3. Results

3.1. Physiological baselines were similar for NTA and HTA participants

Prior to aversive conditioning, the high trait anxiety (HTA; N=49) and normal trait anxiety (NTA; N=159) groups were quite similar. They had a similar age (HTA Mdn=20.83 years, NTA Mdn=20.04 years; Mann-Whitney U=4183.00, p=0.39) and a comparable sex ratio (HTA F: 32, M: 17, NTA F:87, M:72, X² (1)=1.72, p=0.19). The NTA and HTA participants did not differ in resting skin conductance level (U=4073.00, p=0.44), nor in the voltage of the electrical stimulation selected by the subject to achieve an “unpleasant but not painful” sensation during pre-experiment calibration (HTA Mdn=40 V, NTA Mdn=40 V, U=3531.00, p=.63). These results suggest that the perception of the intensity of the shock was comparably aversive for the HTA and NTA groups.

3.2. Effects of discriminative aversive conditioning in NTA participants

3.2.1. Discriminative aversive conditioning evokes discriminative SCR in NTA participants

We assessed whether participants with normal levels of trait anxiety in the paired acquisition group would develop discriminative SCRs (larger SCRs evoked by whichever odor predicted the shock—the CS⁺ —than by the odor that did not—the CS⁻), while participants in the odor only and shock only control groups would not develop discriminative SCRs. To test this hypothesis, we first compared the average of the anticipatory (pre-shock) SCR on just the last three CS+ trials (“late acquisition” trials) and compared between experimental and control groups for NTA participants. Data were notably non-normal, so statistical testing was performed using non-parametric testing with Bonferroni correction applied for multiple pairwise comparisons. As shown in Fig. 1A, for whichever odor served as the CS+, there was a significant effect of experimental group on SCR amplitude during late acquisition trials, H(2)=30.07, p<.001 (independent samples Kruskal-Wallis test). Participants in the paired group (N=80) displayed a higher amplitude anticipatory SCR evoked by the CS+ odor compared to participants in the odor only (N=26, p<.001, d=.63) and shock only group (N=28, p<.001, d=.60). There were no differences in amplitude of CS+ evoked SCR responses between participants in the odor only or shock only groups (p=1.00), indicating that neither passive exposure to the odors nor exposure to electrical stimulation had an effect on participants’ arousal response. These results confirmed that the CS⁺ evoked larger SCRs in participants that underwent aversive conditioning than in control groups.

Figure 1. Effects of discriminative aversive conditioning on SCR for NTA group.

A) Mean of absolute odor evoked SCRs in response to late acquisition CS+ and CS− odors across groups (Paired N=80, Odor Only N=26, Shock Only N=28). B) Cumulative frequency histogram demonstrating the distribution of late averaged late acquisition CS+ and CS− evoked SCRs across groups. Dotted line represents the median.

Conditioning produced a similar pattern of results for the SCR evoked by the CS− odor after conditioning. As shown in Fig. 1A, there was a significant effect of experimental group on CS⁻-evoked SCRs H(2)=21.26, p<.001 (Kruskal-Wallis). Pairwise comparisons revealed that participants in the paired group displayed a higher amplitude anticipatory SCR in the presence of the CS− odor compared to participants in the odor only (p=.026, d=.47) and shock only group (p<.001, d=.53). The significantly larger SCR responses to the CS− odor indicates that the participants in the paired group generalized their arousal response from the CS+ odor that did predict the shock to the (quite similar) CS− odor that did not. However, the SCR response to the CS− odor was not as large as the response to the CS+ odor (Wilcoxon Signed Rank test for related data, T=703.00, p<.001, d=0.44), demonstrating that participants in the NTA paired group did discriminate between the two odors in their learned SCR responses. The raw data are presented as a cumulative frequency histogram in Fig. 1B. The divergence of the distributions of SCR response amplitudes between the CS+-evoked responses (red) and CS− evoked responses (black) illustrate the extent of the discriminative conditioned responses, while the distance between these lines and those of the control groups show the impact of conditioning overall. As expected, there was no significant difference between the CS− evoked SCRs during late acquisition between the odor only and shock only group (p=.562). There was also no difference between the CS+ and CS− evoked SCRs for the odor only group (T=118.00, p=.144) or shock only group (T=241.00, p=.387), demonstrating that only participants that received aversive conditioning exhibited discriminative SCRs.

3.2.2. Extinction learning decreases discriminative SCR in NTA participants

We assessed whether the effects of discriminative aversive conditioning on physiological learning would be decreased by further presentations of the CS+ and CS− odors in the absence of shock. Extinction training partially decreased the conditioned SCR to the CS+ (Fig. 2) such that their SCRs were significantly smaller at the end of extinction training than at the end of acquisition training in the same subjects (N=19, Wilcoxon, T=37.00, p=.020, d=0.53), (Fig. 2A). However, these responses remained considerably larger than zero (Med=0.14, SD=0.19, One-Sample Wilcoxon Signed Rank W=.183, p<.001), demonstrating that the aversive conditioning of the SCR had not fully extinguished. Even though the SCRs were still elevated following extinction training, there was a significant reduction in physiological discrimination between the CS+ and CS− , (Fig.2B) (Related-Samples Wilcoxon, T=43.00, p=.036, d=0.79). In the odor only control group, an additional 32 presentations of odor only made no difference in the physiological discrimination of the CS+ and the CS− (Related-Samples N=6; Wilcoxon, T=8.00, p=.60). This immediate extinction training paradigm thus reduced the physiological discrimination between the CS+ and CS− odors but did not completely eliminate the arousal evoked by the previously conditioned odor stimuli.

Figure 2. Effects of extinction learning on SCR.

A) Mean CS+ evoked SCRs during the last 3 trials of acquisition and extinction for NTA participants in the extinction group (N=19). B) Cumulative frequency distribution of SCRs discrimination for NTA participants in the extinction group.

3.2.3. Discriminative aversive conditioning improves perceptual discrimination in NTA subjects

We investigated whether subjects with normal levels of trait anxiety would improve their perceptual discrimination between two similar odors after discriminative aversive conditioning. To this end, subjects performed the hexanal vs heptanal, three-alternative forced choice odor discrimination test both before and after conditioning (or control treatment). We tested for changes in performance by subtracting the pre-conditioning score, expressed as a percentage correct, from the post-conditioning score. A positive value in this discrimination metric indicates that participants improved their ability to perceptually discriminate between the two odors. In the NTA groups, prior to aversive conditioning there were no differences in perceptual accuracy between participants assigned to the paired acquisition group (N=39, Mdn=50%), the odor only group (N=23, Mdn=50%), or shock only group (N=18, Mdn=50%), (Kruskal-Wallis, H(2)=3.78, p=.15). However, as can be seen in Fig. 3, the experimental group exhibited a statistically significant change in perceptual discrimination (Kruskal-Wallis, H(2)=16.65, p<.001). Bonferroni adjusted pairwise comparisons revealed that participants in the paired group were significant better at discriminating between the CS+ and CS− odors compared to participants in the odor only (p=.008, d=0.41), and shock only (p=.001, d=.06), groups.

Figure 3. Change in perceptual discrimination for NTA.

A) Bars depict mean change in percent correct discrimination trials by experimental group (Paired N=39, Odor Only N=23, Shock Only N=18, Extinction N=14). Error bars refer to 1 standard error. Dashed line represents zero on x-axis. B) Cumulative frequency histogram demonstrating the distribution of change in percent correct discrimination scores across groups. Horizontal dotted line represents the median. Vertical dashed line represents no change in perceptual discrimination.

Each group was also compared to its own baseline performance. Subjects in the paired group exhibited a significant improvement (related samples Wilcoxon Signed Rank Test, T=444, p<.001, d=.78). The median subject in the Paired group scored 50% correct prior to conditioning (i.e. 17% above chance) and 67% correct (i.e. 34% above chance), reflecting a substantial improvement after conditioning. A post-hoc analysis revealed an actual effect size of d=0.78 and an achieved power (1- β) of 0.99. There was no correlation between perceptual improvement and the amplitude of shock selected by the subject (r=.28, p=.30), nor between the difference in SCR between the CS+ and CS− and the degree of perceptual change (r=0.242, p=.39). Participants in the odor only group experienced a slight reduction in perceptual discrimination (one-sample Kolgmogorov-Smirnov Test, D=.216, p=.007, d=0.15), while no change in perceptual discrimination occurred in the group that received only shocks (D=.140, p=.20).

The subset of participants in the extinction group that no longer exhibited discriminative SCRs (defined as a difference between the CS+ and CS− of less than 0.1 μS) nonetheless showed significant improvement (Fig. 3) in perceptual discrimination between the two odors compared to their own baseline discrimination (N=14, Related-Samples Wilcoxon, T=59.00, p=.018, d=0.67). This extinction-resistant perceptual improvement was similar to that observed for participants in the acquisition and sham extinction (who received a delay rather than extinction training after acquisition) groups, as there was no statically significant difference between the level of perceptual improvement (Independent-Samples Kruskal-Wallis Test, H(2)=1.34, p=.51). The control odor only group in which participants received the same number of odor trials as participants in the extinction group also showed no perceptual improvement (Related-Samples Wilcoxon Signed Rank Test, T=14.50, p=.93). Thus, extinction training did not reduce the perceptual improvements that followed aversive conditioning despite eliminating the difference in SCR evoked by the CS+ and CS− odors.

3.3. Effects of Trait anxiety on SCR

3.3.1. Generalization of physiological responses to CS+ and CS− in High Trait Anxiety subjects

We investigated whether subjects with high levels of trait anxiety would generalize their learned skin conductance responses to two similar odors that differentially predicted an aversive shock. Figure 4 illustrates the longitudinal change in SCR for both NTA and HTA subjects by comparing the mean SCR response on the very first CS+ presentation (prior to any odor-shock pairing) to the average SCR on the last three CS+ presentations. Both subsets of participants exhibited a significant increase in CS+-evoked SCR compared to their own first trial (related samples Wilcoxon Signed Rank Test (NTA: T=3028.00; p<.001, d=.69; HTA: T=269.00, p=.001, d=0.51), demonstrating that both groups did develop a learned arousal response to the CS+ odor. However, the HTA participants exhibited a different pattern than the NTA participants. Unlike the NTA participants, in HTA participants the CS− evoked SCRs that were no smaller than those evoked by the CS+ in the same subjects (related samples Wilcoxon Signed Rank Test, T=143.00, p=.84), suggesting that HTA subjects generalized their aversion to both odors (Fig. 5). Overall, participants’ level of trait anxiety was weakly and negatively correlated with their conditioned SCR difference, as revealed by a Spearman’s rank-order correlation (r_s (104)=.215, p=.029), see figure 5C for distributions of conditioned SCR difference by anxiety level. Despite their similar unconditioned SCR responses to shock, HTA participants also exhibited significantly smaller conditioned responses to the CS+ odor prior to conditioning than NTA participants (Mann-Whitney; U=534.00, z=−3.29, p=.001, d=−0.33). Moreover, the increase in SCR evoked by conditioning was much less in the HTA participants than in those with typical levels of trait anxiety (T=3028.00, p<.001, d=.69). These data are consistent with previous reports that subjects with high trait anxiety exhibit blunted SCR responses during aversive conditioning (Jezova, Makatsori, Duncko, Moncek, & Jakubek, 2004; Naveteur & Baque, 1987; Naveteur, Buisine, & Gruzelier, 2005). This blunting was most evident in learned responses as the unconditioned skin conductance response to the first electrical stimulation in the conditioned group was not statically significantly different for HTA and NTA participants (HTA N = 15, Mdn=0.17 μS, NTA N=35, Mdn=0.37 μS, U=1037.00, p=.096).

Figure 4. Conditioning of SCR for NTA and HTA groups.

A) Mean SCRs evoked by the first and averaged last three CS+ trials for NTA (N=80) and HTA (N=24) groups. Error bars indicate 1 standard error. B) Cumulative frequency distribution of CS+ evoked SCRs for first, and last three CS+ trials, across anxiety groups. Dotted line represents the median.

Figure 5. Effects of trait anxiety on SCR for NTA and HTA groups.

A) Effects of trait anxiety on physiological responses of participants in the paired group to CS+ and CS− odors (NTA=80, HTA = 24). B) Cumulative frequency histogram demonstrating the distribution of late acquisition CS+ and CS− evoked SCRs across NTA and HTA groups. Dotted line represents the median. C) Cumulative frequency histogram demonstrating the distribution of conditioned SCR difference by STAI Trait score.

HTA participants did not exhibit sufficient difference in their SCRs to the CS+ and CS− after the initial conditioning to be analyzed for the effects of extinction learning. However, the HTA participants after conditioning were statistically extremely similar to the NTA participants who had undergone extinction training in that a) both NTA participants after extinction training (N =19) and HTA participants after conditioning (N=24) exhibited similarly modest SCRs to the CS+ (Mann-Whitney, U=205.0, p=0.57) and b) both groups exhibited similar responses to the CS+ and CS−.

3.3.2. Participants with high trait anxiety do not improve on perceptual discrimination

We investigated whether, unlike NTA participants, participants with high levels of trait anxiety would fail to improve their perceptual discrimination between two similar odors following discriminative aversive conditioning. Prior to conditioning, NTA and HTA participants were equally poor at discriminating the test odors hexanal and heptanal during the baseline perceptual testing. The median score for the NTA participants in the odor-shock paired group (N = 39) was 50% correct and the median score for the HTA participants (N = 15) was also 50% correct, both modestly better than chance (33% correct) with no statistical difference between these results (M-W test, U = 295.5, p = .952).

However, unlike the NTA participants, High Trait Anxiety participants in the paired group exhibited no change in perceptual discrimination after conditioning (T=36, p=.81), with nearly identical scores before and afterwards (Fig. 6). In order to rule out the possibility that this null result was due to a lack of statistical power in the smaller HTA subset of participants, we conducted a sensitivity analysis with power (1-β)=0.8 and α=.05, and our sample of N=15, using GPower (Faul, Erdfelder, Lang, & Buchner, 2007). This analysis confirmed that our sample size should have been sufficient to detect an effect size of d=0.693, which is smaller than the 0.78 effect size actually observed in the NTA group. Moreover, the average performance on the discrimination task actually dropped slightly after conditioning in the HTA participants, with no sign of improvement at all.

Figure 6. Discrimination test scores for NTA and HTA participants.

A) Mean percent correct trials on triangular task before and after conditioning for normal trait anxiety (N=39) and high trait anxiety (N=15) participants in the paired group. Error bars indicate 1 standard error. Dashed line indicates performance at chance level (33%). Cumulative frequency histogram demonstrating the distribution of percent correct discrimination scores across anxiety groups. Dotted line represents the median.

4. Discussion

NTA participants exhibited increased SCRs in response to both odors after conditioning, but their SCRs were significantly larger for whichever odor predicted the mild wrist shock. The learning experience enables them to distinguish between the threat-predictive and non-threat-predictive stimuli and make an appropriately discriminative physiological response when encountering either stimulus again. By contrast, HTA participants did not become better at telling the odors apart and exhibited comparably heightened SCRs for both stimuli, thus exhibiting both a failure of sensory plasticity and a generalization of the aversive response to the non-threat predictive stimulus. Importantly, the perceptual improvement exhibited by NTA participants persisted after extinction training had significantly reduced their conditioned SCR. This suggests that the learned sensory discrimination between stimuli that differentially predicted aversive outcomes is sustained as NTA individuals navigate and interpret the world even when no negative outcome is expected. HTA individuals never gain the sensory benefits that normally inure after learning to discriminate the similar odor, and thus navigate the world with less ability to tell a threat-predictive stimulus from a non-threat-predictive stimulus.

Prior to any conditioning, the HTA participants were very similar to the NTA participants, including in the shock amplitude they selected and in the unconditioned SCR evoked by the shock. However, the NTA participants on average exhibited a small SCR to the CS+ odor on its very first presentation, prior to any shocks besides the calibration shocks during set up, but the HTA participants did not exhibit an SCR to the CS+ odor (Figure 4). Conditioning induced learned SCR responses to both the CS+ and CS− odors in the HTA participants (Fig. 5) that were not as large in absolute terms as those exhibited by NTA participants but were very large increases in proportion to their pre-conditioning odor-evoked SCRs (Fig. 4). These results are consistent with multiple previous reports that HTA individuals, exhibit blunted skin conductance responses during aversive conditioning (Jezova et al., 2004; Naveteur & Baque, 1987; Naveteur et al., 2005; Wilken, Smith, Tola, & Mann, 2000). The extinguished NTA group serves as an important comparison for the HTA group, in that it also exhibited reduced SCR responses to the CS+ compared to the NTA subjects after conditioning and also showed similar SCR responses to both CS+ and CS− odors, nevertheless showed significant perceptual improvement. Neither a large SCR nor a difference in SCR between CS+ and CS− is necessary for learning-induced sensory discrimination.

Following extinction training, we observed heightened olfactory discrimination in NTA subjects despite their reduced SCRs compared to themselves after acquisition training. To our knowledge, this is the first investigation of extinction in humans after odor-cued discriminative fear conditioning. It may be that extinction training was incomplete and that a more extended extinction paradigm could have fully extinguished the conditioned SCR and the sensory improvement. Olfactory stimuli do evoke greater arousal than visual stimuli (Parma, Ferraro, Miller, Åhs, & Lundström, 2015) and evoke memories that are more emotionally salient compared to other sensory modalities (Herz & Schooler, 2002), supporting the idea that CS+ odor-evoked physiological responses might require more extinction trials than stimuli in other modalities. However, even if more training might eventually have eliminated the effect, our data indicate that the sensory enhancement is present for at least the first 16 presentations of the CS+ after learning in this paradigm. Additional extinction training trials or sessions (e.g. to facilitate extinction memory consolidation) would provide valuable information about the ultimate duration of the sensory effects. It may also be important to consider whether anxiety-related subject variables like participant age and sex impact this form of acquisition and extinction learning, but this study did not have appropriate statistical power to test these hypotheses.

Odor-cued aversive conditioning is known to improve odor discrimination and detection in typical human subjects (Åhs et al., 2013; Li et al., 2008), including effects that persist over time (Parma et al., 2015). It is not clear whether this effect could be specific to olfactory sensory processing. Olfactory cues can evoke emotionally richer experiences compared to other sensory modalities (Arshamian et al., 2013; Herz & Schooler, 2002; McGann, 2017) that can include trauma-related memories and can trigger clinical symptoms like anxiety, flashbacks and panic attacks (Cortese, McConnell, Froeliger, Leslie, & Uhde, 2015; Daniels & Vermetten, 2016; Hinton, Pich, Chhean, & Pollack, 2004; Kline & Rausch, 1985; McCaffrey, Lorig, Pendrey, McCutcheon, & Garrett, 1993). However, the tendency toward generalization in high trait anxiety individuals has been observed in other sensory modalities for cued aversive conditioning, especially with initially ambiguous stimuli. In the visual modality, individuals with high trait anxiety displayed equally large startle responses to a visual cue (an image of a face) that served as a CS+ and a similar visual cue (i.e. a different face) that served as the CS−, while low trait anxiety participants displayed more startle to the CS+ than the CS− (Haddad et al., 2012). Importantly, in that study high trait anxiety participants displayed differential levels of startle response towards a second CS− (the image of an oval shape) which was unambiguously different from the CS+, highlighting the tendency of high trait anxiety individuals to generalize under conditions of ambiguity. By contrast, a different study using perceptually unambiguous visual cues as stimuli found no evidence of trait anxiety effects on generalization (Torrents-Rodas et al., 2013). We believe the present results will not be unique to olfactory stimuli but will apply to ambiguous stimuli across sensory modalities.

Sources of ambiguity in these experiments include not only physical ambiguity, where stimuli are so physically similar as to be nearly indistinguishable, but also contingency ambiguity, where it is unclear whether two differing stimuli actually predict different outcomes. Resolving contingency ambiguity requires learning to make a discrimination/generalization decision including which stimulus dimensions are predictive of outcomes and which are not, and they may thus be similarly subject to disruption in HTA individuals. For example, Wong & Lovibond used a classical fear paradigm in which a black dot was presented at various positions inside a yellow square that was only sometimes paired with an aversive shock. They found that high trait anxiety participants generalized their expectancy of shock to a range of dot positions, but only when they failed to form a theory about the relationship between the position of the dot and the delivery of a shock (Wong & Lovibond, 2018). We propose that such inference formation may depend in part on changes in sensory processing that is impaired in HTA individuals.

4.1. Conclusion

Our results indicate that people with high levels of trait anxiety lack a form of aversive learning-induced sensory plasticity that is robust in people with typical levels of trait anxiety. The heightened generalization of anxiety across situations in anxious people may also manifest as sensory generalization across similar stimuli. People with high trait anxiety may be less likely to perceive important distinctions between situations that predict a negative outcome and those that do not, contributing to the maintenance to anxiety disorders. Therapeutic interventions like exposure therapy typically presume that the patient recognizes differences between threatening and non-threatening situations and merely needs to learn the difference in their outcomes. It may be an important complement to this approach emphasize the physical, sensory differences between these situations as well.

Highlights.

• High trait anxiety blocks aversive learning-induced sensory plasticity

• The heightened generalization of anxiety manifests as sensory generalization

• High trait anxiety may disrupt ability to distinguish safe from danger-associated cues

• Extinction does not reduce the enhanced perceptual discrimination

• Exposure therapy interventions might benefit from addressing sensory generalization