The excessive generalization of fear affected by perceptual bias in experimental pain individuals: Evidence from an event‐related potential study

Abstract Introduction Excessiv generalization of fear contributes to the development and maintenance of pain. Prior research has demonstrated the importance of perception in fear generalization and found that individuals in painful conditions exhibited perceptual bias. However, the extent to which perceptual bias in pain affects the generalization of pain‐related fear and its underlying neural activity remains unclear. Methods Here, we tested whether perceptual bias in experimental pain individuals led to the overgeneralization of pain‐related fear by recording behavioral and neural responses. To this end, we established an experimental pain model by spraying capsaicin on the surface of the seventh cervical vertebra of the participant. A total of 23 experimental pain participants and 23 matched nonpain controls learned fear conditioning and then completed the fear generalization paradigm combined with the perceptual categorization task. Results We found that the novel and safety cues were more likely to be identified as threat cues in the experimental group, resulting in higher US expectancy ratings compared to the control group. The event‐related potential results showed that the experimental group exhibited earlier N1 latency and smaller P1 and late positive potential amplitudes than those in the control group. Conclusion Our findings suggest that the experimental pain individuals exhibited an excessive generalization of fear affected by perceptual bias and reduced their attentional allocation to pain‐related fear stimuli.

cal and psychological aspects of pain. Capsaicin is commonly used to induce experimental pain and has been proven to be a safe, noninvasive approach for producing stable, persistent and reproducible painful conditions to evaluate pain mechanisms (Price et al., 2018;Shenoy et al., 2011;Silberberg et al., 2015).
EEG is sensitive to fast and transient cortical processes, allowing the exploration of temporal neural activity at different sensory and cognitive levels. Fear conditioning can modulate early and late components such as P1, N1, P2, and late positive potential (LPP). The P1 component, peak between 100 and 130 ms (Luck, 2014), was used to explore the dynamic of early attentional control (Clark & Hillyard, 1996). Evidence from several studies showed that threatening stimuli elicited enhanced P1 amplitudes (Gupta et al., 2019;Krusemark & Li, 2011;Stefanou et al., 2019). The subsequent component is N1 (peak at 100-150 ms poststimulus), which indicates the extent to which fearful stimuli allocated attentional resources (Flor et al., 2002;Rothemund et al., 2012). The CS was thought to have no effect on N1 amplitudes, but profoundly reduced the latencies of N1 (Scaife et al., 2006). Another component used to assess the brain activity to fearful stimuli is P2, which is consistently associated with anterior cingulate cortex activity (Garcia-Larrea et al., 2003). The P2 component was found to be important in pain modulation during the emotional process (Petrovic & Ingvar, 2002;Price, 2000). A previous study has revealed increased amplitudes in the P2 component for unexpected fearful versus neutral faces (Yang et al., 2010). LPP, beginning at 300 ms after the onset of stimuli, was widely used to explore attentional and emotional processing (Pavlov & Kotchoubey, 2019). Numerous studies have shown enhanced LPP amplitudes in threat cues compared to safety cues (Bacigalupo & Luck, 2018;Panitz et al., 2015;Panitz et al., 2018;Pastor et al., 2015). LPP was a useful tool for examining individual differences in fear conditioning. However, the LPP amplitude showed no difference among generalized stimuli, thus suggesting that LPP was insensitive to fear generalization (Bauer et al., 2020;Nelson et al., 2015). Based on the research above, we were interested in assessing the neural mechanism of fear generalization in individuals with pain by analyzing the average ERP response across P1, N1, P2, and LPP. Zaman et al. (2015) noted that individuals with chronic pain showed an impairment of perceptual discrimination, even though a growing body of studies has applied perceptually similar stimuli to explore painrelated fear generalization (Meulders, Harvie et al., 2015; and showed an overgeneralization of fear in pain states (Meulders et al., 2014;Meulders, Jans et al., 2015). However, whether such impairment would also exist in acute pain individuals and whether it would lead to an overgeneralization of pain-related fear still remains unclear. To fill this gap, we applied a generalization protocol developed by Struyf et al. (2017), which combined the paradigm of fear generalization and a categorization task to evaluate generalization behavior and stimulus perception simultaneously. In this experiment, continuously sized circles served as CSs (Lissek et al., 2008). The middle-size circle was paired with an electrical stimulus (den Hollander et al., 2015;Glogan et al., 2019;Karos et al., 2015;Vandael et al., 2019) to elicit fear responses. During the generalization phase, one stimulus (either CS or GS) was presented in each trial and EEG was recorded. Then, a categorization task appeared, requiring participants to categorize the current stimulus as CS or GS.
After categorizing the stimulus, participants were asked to rate the US risk expectancy. A higher US expectancy ratings reveals a stronger fear response. Based on previous research, we hypothesized that compared to the control group, experimental pain individuals may exhibit perceptual bias which contributed to the overgeneralization of pain-related fear. Second, the ERP component would differentiate CS from GS and show a generalization gradient. Moreover, since the prior study had demonstrated that individuals with experimental pain could not pay as much attention to cognitive tasks as pain-free controls (Wang et al., 2020), the third hypothesis was that the amplitudes of the ERP components in the experimental group would be smaller than that in the control group. TA B L E 1 Demographic and questionnaire scores for the experimental pain group and control group (M ± SD).

Experimental pain model
To explore fear generalization in acute pain conditions, we established an experimental pain model induced by capsaicin in the EP group throughout the experiment. For the model, capsaicin powder (Sigma) was dissolved in 50% ethanol in water at a concentration of 0.1%, sprayed on the surface of the seventh cervical vertebra and covered with plastic wrap. The NP group was sprayed with pure water and covered with plastic wrap (Wang et al., 2020). The visual analog scale (VAS) was used to assess pain intensity with a target of 4, indicating that the sensation was starting to become painful. During the experiment, we continuously assessed the pain intensity between blocks. Once the participant's pain perception dropped below 4/10 on the VAS, capsaicin was applied to maintain the pain perception score above 4/10.

Conditioned and unconditioned stimuli
According to previous studies, nine white circles of gradually increasing diameters (from 5.08 to 11.18 cm, in steps of 0.762 cm; see Figure 1A) were presented on a black background (Lissek et al., 2008).
An electrical stimulus served as an unconditioned painful stimulus (pain-US) using the BIOPAC stimulator module (STM200 BIOPAC Systems Inc., Santa Barbra, CA). The US was attached to a pair of electrodes (9 mm diameter, 30 mm distance between electrodes) filled with K-Y gel. The electrodes were placed on the tibial surface of the leg (the stimulated side was balanced across participants) about 10 cm above the malleolus (van de Donk et al., 2019) to avoid muscle contractions.
The intensity of US was individually calibrated using a series of ascending electrical stimuli with a step of 0.02 V and an interstimulus interval ranging from 7 to 15 s (Carlino et al., 2010). After the presentation of each stimulus, participants scored subjective perception/pain intensity using the 11-point Likert scale. Here, 0 = "You don't feel anything at all," 1 = "You feel something, but it's not painful, it's merely a sensation," 2 = "The sensation starts to become aversive, but it's still not painful" and 10 = "This is the worst pain you can imagine." The target intensity for the US was "significantly painful and demands some efforts to tolerate," which was roughly equivalent to a score of 8 on this calibration scale (Meulders, Harvie et al., 2015)   Both the experiment and control group underwent the habituation, acquisition, and generalization phases. During the habituation phase, CS+ or CS-was presented for 1200 ms following a fixation cross. In the acquisition phase, US expectancy ratings appeared after the presentation of the stimulus, then 9 out of the 12 CS+ were paired with electrical shock as the US. In the generalization phase, after displaying the CSs or GSs, a categorization task appeared that required participants to identify the current stimulus as either CS+, CS-or as a novel stimulus, then followed by US expectancy ratings. CS = conditioned stimulus, US = unconditioned stimulus, GS = generalization stimulus.
participants were asked if they agreed to repeatedly receive the stim-

Procedure
The participant was seated in a sound-attenuated, dimly lit room. The experiment was programmed using E-prime3.0 (Psychology Software Tools Inc., Pittsburgh, USA) on a 21-inch computer monitor with a resolution of 1920 × 1080 pixels and a refresh rate of 60 Hz, placed approximately 60 cm in front of participants.
There were three consecutive phases: habituation, acquisition and generalization (see Figure 1B), in which trials were carried out in quasirandom order such that the same stimuli occurred no more than twice in a row. During the habituation phase, the CS-1, CS-2 and CS+ were presented 3 times each. Trials started with a fixation cross for 500-800 ms, followed by a CS presented for 1200 ms, with intertrial intervals (ITI) ranging from 1000 to 3000 ms. In the acquisition phase, 12 CS+, 12 CS-1, and 12 CS-2 each presented for 1200 ms and were divided into three blocks. To explore the conditioned response between the EP and NP groups, after the presentation of the CS, participants were asked to rate the level of US expectancy on a 9-point scale (1 = least likely, 5 = moderately likely, 9 = most likely) with their right hand. After answering, 9 out of the 12 CS+ were followed by the US (75% reinforcement rate, 50 ms) while the CS-was never followed by the US.
The generalization phases consisted of 10 blocks. A categorization task was performed aiming to explore the role of perception in fear generalization. In each block, CS+ was presented 6 times and CS-1, CS-2, and GSs were presented 3 times each. After the CS or GS was presented for 1200 ms, the categorization task appeared at the bottom of the circle. The participant was required to identify whether the circle was the stimulus previously judged as 1) a novel stimulus (GS), 2) the best predictor of US (CS+) or 3) the stimulus, which did not predict US (CS-). They pressed button "J" with their left hand when they categorized the current circle as GS and pressed button "K" when they thought the current circle was CS+, while button "L" was pressed when they identified the current circle as CS-. After the categorization task, participants were asked to rate the US expectancy with their right hand. To avoid extinction, 3 of the 6 CS+ in each block were paired with US (50% reinforcement rate).

EEG recording and preprocessing
The scalp EEG was recorded at a sampling rate of 2048 Hz from 64 Ag/Ag-Cl scalp electrodes mounted according to the international 10-20 system using the 64-channel BIOSEMI Active Two system. The average value of bilateral mastoids served as an online reference.
Brain activity was continuously recorded over a band-pass range of 0.01-100 Hz, with electrode impedances kept below 20 kΩ.
The offline analysis of EEG data was processed using the EEGLAB toolbox (version 13_0_0b) (Delorme & Makeig, 2004) in Matlab (R2013b, MathWorks, Natick, MA, USA). After loading the raw data, 64 electrodes were selected and located. The EEG data were downsampled to 256 Hz. Additionally, the EEG signals were band-pass filtered between 0.5 and 30 Hz and notch filtered between 49 and 51 Hz. The EEG epochs were segmented using 1200 ms time windows Epochs with activity exceeding ±100 µV were rejected using a semiautomated procedure. Finally, clean EEG data epochs were merged and averaged for each condition at the group level to obtain the grand average ERP waveform.

Data analysis
In the acquisition phase, for the behavioral data, we analyzed the US expectancy ratings using a three-way repeated-measures analysis of variance (RMANOVA) with Stimulus (CS+ and CS-) and Block (blocks 1-3 of acquisition phase) as within-subject factors and Group as the between-subject factor. In the generalization phase, we first aimed to investigate whether there was perceptual bias in the EP group compared with the NP group. We converted the categorization data into the probabilities of being categorized as CS+, GS, and CS-for each test stimulus and analyzed using a two-way RMANOVA [Stimulus (CS+, CS-, GSs) × Group] in these three categories, separately.
Then, we analyzed the US expectancy ratings in two steps. In the first step, we averaged the ratings for each type of stimulus in each group across blocks 1 to 10 in the generalization phase, using RMANOVA with Stimulus and Block as within-subject factors and Group as the between-subject factor. We expected the EP group to have higher US expectancy ratings than the NP group, indicating an overgeneralization of pain-related fear in the EP group. In the second step, we performed three-way RMANOVA [Stimulus × Category (CS+ category, GS category and CS-category) × Group] to analyze the US expectancy ratings when the stimuli were categorized as CS+, GS, and CS-, respectively, to explore the group differences and the role of perception in the US expectancy ratings.
To have an adequate signal-to-noise ratio for the ERP analysis, we

Behavioral results
In the acquisition phase, the US expectancy ratings were significantly   The follow-up contrast demonstrated that the EP group showed significantly higher responses in blocks 2, 3, 5, and 6 than those in the NP group (all ps < 0.05). These results reflected the overgeneralization of pain-related fear in the EP group ( Figure 4A).
As a follow-up to the perceptual categorization effect, we analyzed the US expectancy ratings when the test stimulus was categorized as CS+, GS, and CS-, performing a Stimulus × Category × Group RMANOVA ( Figure 4B). The ratings were modulated by the main effect  The behavioral results indicated that participants successfully learned the CS+ as dangerous stimulus and CS-as safety one in the acquisition phase. In the generalization phase, the EP group was more likely to judge the generalization stimulus as dangerous one. Regarding to the US expectancy ratings, the EP group showed greater response than the NP group, which indicating an overgenerlization in EP group.
Participants showed higher US expectancy ratings in the CS+ category than those in GS and CS-categories. We also found greater responses in the EP group compared to those in the NP group in the CS+ category. These results suggested that perceptual bias in the EP group might have an effect on the pain-related fear generalization.

ERP results
The analysis of the N1 component showed no statistically significant main effects or interaction effects in averaged amplitudes ( Figure 5).  For the ERP results, we found that the EP group showed shorter N1 latency as well as smaller P1 and LPP amplitudes than the NP group.

DISCUSSION
In the current study, we examined the role of perception in the generalization of pain-related fear as well as the underlying neural mechanisms in a painful state. The results demonstrated that indivudual in painful condition showed perceptual bias which influenced the overgeneralization of pain-related fear and they showed a different pattern of psychobiological processes.
The stronger fear responses in the EP group were consistent with previous research studied in fibromyalgia and chronic hand pain patients, which showed a greater fear response compared to healthy individuals (Meulders et al., 2014;Meulders, Jans et al., 2015). Growing studies provided a perceptual account for the observed overgeneral- ization. Hence, we combined the fear generalization paradigm with a categorization task to explore the impact of perceptual bias in the fear generalization in experimental pain. Previous study found that pain interferes with cognitive tasks via attentional disruption (Eccleston & Crombez, 1999). The categorization task can be viewed as a cognitive task, which was disturbed by the experimental pain condition in our study. This impact of pain might lead to a perceptual bias in painful sufferers. In Vriends' (2011) study, they applied CS+ and CS-to explore the fear acquisition and extinction in individuals with state anxiety and found enhanced responses to the CS+ as well as the CS-in an anxious state. Vriends et al. found that after the fear learning in the acquisition, individuals may be prepared psychologically and transfer the emotional regulation strategies to the subsequent stage of conditioned fear. Thus, the higher ratings in the GS category than the CS-category in the EP group found in our study might be that individuals in painful conditions were induced a more alert psychological preparation after the stage of acquisition and applied the adjustment strategy of fear emotion to the generalization phase, resulting in a higher fear expectation for the stimulus which they categorized as a novel one. Apart from that, when the participants identified the current test stimuli in the same category, such as the GS category, the generalized fear responses still revealed differences based on perceptual similarity in both the EP and NP group. Consistent with previous studies (Struyf et al., 2017;Zaman et al., 2019), these results indicated that fear generalization is not only a byproduct of perception bias in experimental pain individuals, but may also be driven by other cognitive processes. The "safe-than-sorry strategy" might be considered an explanation for these findings in our study. In this case, participants elicited a stronger fear response, even when they were aware that the current stimulus is a novel one but not a conditioned fear stimulus, for the reason that a misidentification of the safe stimulus as the dangerous one is better than incorrect identification of the conditioned stimulus as a safe one (Dunsmoor & Paz, 2015;Shepard, 1987). This notion reflects that generalization is a postperceptual decision process.
The N1 component is mainly related to the orientation of attention in perceptual discrimination (Ohoyama et al., 2012;Vogel & Luck, 2000). Hence, the change in N1 latency may reflect the processing speed of attentional allocation (Vogel & Luck, 2000). Based on a prior study, the pain-related fear is associated with hypervigilance (Leeuw et al., 2007), which suggests that the reduction in N1 latency is accounted for increasing vigilance in pain conditions, thus accelerating the speed of attention allocation toward the pain-related fear stimulus. This result is consistent with the behavioral results supporting our conclusion that individuals in painful conditions applied a more alert adjustment strategy of fear emotion to the generalization phase.
Moreover, the P1 component was associated with the efficiency of visual stimulus detection through the allocation of top-down attention (Desimone & Duncan, 1995;Hillyard & Anllo-Vento, 1998;Luck et al., 1990). The decreased P1 amplitudes in the EP group were possibly related to the reduction of visual processing of pain-related fear stimulus, which reflects an attentional avoidance in the experimental pain individuals. This finding is consistent with works on the fear-avoidance model, suggesting that highly pain-related fear individuals view pain as a sign of damage, which is linked to limited pain-control coping strategies and results in avoidance of pain-related movements and activities (Bunzli et al., 2017).
In the present study, we found that the behavioral results showed a different pattern from the ERP results, as the US expectancy ratings were significantly modulated by the type of stimulus. However, the P2 amplitude did not exhibit such modulation by the type of stimulus but even showed higher amplitude for novel stimulus compared with those for CS+. A dual-process theory of conditioning states a dissociation between implicit and explicit processes and that conditioning can occur independently of explicit contingency awareness (Balderston & Helmstetter, 2010;Schultz & Helmstetter, 2010). In the present study, the US expectancy rating measures the explicit process for the CS-US contingency awareness, while EEG is applied to explore the implicit processes to fear CSs (Lonsdorf et al., 2017). Behavioral and ERP results demonstrated that participants were capable of expressing contingency awareness, but did not show evidence of autonomic conditioning. Our study provides evidence for the dual-process interpretation of conditioning, suggesting that implicit and explicit learning might simultaneously exist in fear conditioning.
The present study also showed that LPP exhibited the same ERP difference pattern as the P1 component, reflecting an enhanced amplitude in pain-free conditions compared with painful conditions. These findings further demonstrate the difference in neural responses between pain and pain-free conditions, consistent with previous research into experimental pain (Wang et al., 2020), supporting the conclusion that pain disrupts early and later neural potentials. Eccleston (1994) has proposed that pain demands attentional resources.
Since experimentally induced pain has been found to impair aspects of attention (Bingel et al., 2007;Moore et al., 2013;Seminowicz et al., 2004;Wang et al., 2020), the reduction of LPP amplitude between the pain and pain-free condition may be due to the impairment of allocation of attention in pain conditions. Some limitations warrant comment. First, during the acquisition phase, the number of CS-US trials was limited, which might not form strong enough conditioning effects. Second, to reduce the fatigue effect caused by a long experiment time, the presentation time of stimuli and intertrial interval was shortened. This parameter might result in considerable effects on ERP results. Finally, the present study explored fear generalization in a control experimental setting. Hence, the conclusion in this experiment could not generally apply to patients suffering from clinical acute or chronic pain. Future research should include patients with clinical pain to explore the discrepancy in brain mechanisms between healthy and clinical pain individuals.
In summary, the current study reported that the impairment of perception contributed to a fear overgeneralization in pain conditions, which led to increasing vigilance and reduced attention to the pain-related stimulus. The present findings offer insights into psychobiological processes involved in fear conditioning and the generalization of an acute painful state, which may provide new evidence for the role of pain-related fear in the transition from acute to chronic pain.

ACKNOWLEDGMENTS
We thank all participants for their effort in taking part in our study.

DATA AVAILABILITY STATEMENT
I confirm that my article contains a Data Availability Statement even if no data are available.